Twitter

[NorCal]

Um, you were saying about AI not necessarily being racist and otherwise biased?

https://gizmodo.com/twitter-layoffs-elon-musk-ai-ethics-1849743051

Getting rid of the people who's job it is to try to make sure the AI isn't racist doesn't seem like a good start.

So your view is that an example of a human being doing bad things should make us more willing to trust human moderators?

Sure seems like humans have a demonstrated history of disregarding ethics.

Human beings are not monoliths.

Neither are artificial intelligences. <-- which was my original point.

I agree with you that there is clearly a difference in how you and I think about the world. You are taking the negatives behaviors of some specific entities and generalizing them to assert that all entities with a set of characteristics must exhibit those same negative behaviors.

AI is cool and everything, but I don't think it's automatically a net positive. I also don't think we know all the harms that are going to result. A human being can pull the power plug if necessary. If you let AI go, without people overseeing it and at least trying to prevent the bad stuff from happening, then bad stuff is going to happen. Is a human overseer perfect? No. But its better than nothing.

I think maybe you know different human beings than I do. I think humans are cool and all, but in lots of situations humans aren't automatically a net positive. When we let humans go without some sort of oversight bad stuff can and does happen. We have thousands of years of recorded history demonstrating all the bad stuff humans do, starting with murder and war, progressing to genocide, and throwing in nuclear meltdowns exacerbated by (human) operator error.

Would AI moderation be perfect? No. It'll make mistakes. But it's better than nothing. And given how much bias and racism humans have been demonstrated to exhibit -- and how much harm (human moderated) social media seems to done to our democracy -- the status quo clearly is not sustainable.

There are a lot of misconceptions about what "AI" actually does and does not do.  It's a fancy buzzword that is mostly a complex series of regression models.  Essentially it applies statistical models to the world to find the best "fit" of sample data to a desired outcome. 

Left to its own devices, AI generates a lot of outputs that we would consider highly racist or would otherwise offend our value systems.  In situations like mortgage lending, AI will generate outcomes that violate fair lending laws, even if race is not an input.

AI has many beneficial uses, but how it's used and managed matter a great deal.  It's a lot more nuanced than human = bad and AI = good. 

[maizefolk]

It's a lot more nuanced than human = bad and AI = good.

I'm perfectly happy to agree with you on this.

My argument above is an exercise in creating a mirror image of the argument that was being made both another that 1) AI is a monolith and 2) human = good and AI = bad.

Left to its own devices, AI generates a lot of outputs that we would consider highly racist or would otherwise offend our value systems.  In situations like mortgage lending, AI will generate outcomes that violate fair lending laws, even if race is not an input.

Edit: This is another example. Some AI models, trained with some datasets, will have racially disparate outcomes in their recommendations on  mortgage lending decisions. I completely agree.

But there is a big jump from this to "AI is necessarily racist".

And again. This isn't unique to AI models or even worse in AI models than having humans do the same work (which is typically the data used to train models). For example, look at how human beings appraise houses based on the race of the people who show them around the house and the race of the people in photos hanging in the house. https://www.nytimes.com/2022/08/18/realestate/housing-discrimination-maryland.html

[NorCal]

My prediction is that Twitter will be a business school case study in the quickest way to make $44B evaporate.

Musk has put himself in a situation where he's a guaranteed loser.

He's promised the right "absolute free speech".  Or at least the far right has interpreted it that way.  Anything he does to moderate content will now be considered a betrayal.  Even keeping Donald Trump banned will be considered a betrayal.

It seems a decent number of users on the left are already fed up and leaving the platform.  It may or may not be a full trend, but it could turn into one.  Network effects work in reverse too.  Just ask MySpace.

And the digital advertising market is cooling pretty rapidly anyways.  Advertisers are already broadly pulling back on their spending and don't need much incentive to move their money to different platforms.  Major brands would pull the plug if content moderation is even perceived to loosen up.  And Twitter probably just fired most of the team needed to keep up with their existing content moderation anyways.

And Musk put enough debt on the buy-out to pretty much wipe out Twitter's existing cash-flow.  It's just a question of whether the revenue declines are more or less than the savings from firing 3,500 people. 

[NorCal]

It's a lot more nuanced than human = bad and AI = good.

I'm perfectly happy to agree with you on this.

My argument above is an exercise in creating a mirror image of the argument that was being made both another that 1) AI is a monolith and 2) human = good and AI = bad.

Left to its own devices, AI generates a lot of outputs that we would consider highly racist or would otherwise offend our value systems.  In situations like mortgage lending, AI will generate outcomes that violate fair lending laws, even if race is not an input.

Edit: This is another example. Some AI models, trained with some datasets, will have racially disparate outcomes in their recommendations on  mortgage lending decisions. I completely agree.

But there is a big jump from this to "AI is necessarily racist".

And again. This isn't unique to AI models or even worse in AI models than having humans do the same work (which is typically the data used to train models). For example, look at how human beings appraise houses based on the race of the people who show them around the house and the race of the people in photos hanging in the house. https://www.nytimes.com/2022/08/18/realestate/housing-discrimination-maryland.html

In the context of content moderation, I would suspect that AI is highly useful in pointing human moderators to the right topics to review, but is nearly useless in making independent moderation decisions.

[jeninco]

It's a lot more nuanced than human = bad and AI = good.

I'm perfectly happy to agree with you on this.

My argument above is an exercise in creating a mirror image of the argument that was being made both another that 1) AI is a monolith and 2) human = good and AI = bad.

Left to its own devices, AI generates a lot of outputs that we would consider highly racist or would otherwise offend our value systems.  In situations like mortgage lending, AI will generate outcomes that violate fair lending laws, even if race is not an input.

Edit: This is another example. Some AI models, trained with some datasets, will have racially disparate outcomes in their recommendations on  mortgage lending decisions. I completely agree.

But there is a big jump from this to "AI is necessarily racist".

And again. This isn't unique to AI models or even worse in AI models than having humans do the same work (which is typically the data used to train models). For example, look at how human beings appraise houses based on the race of the people who show them around the house and the race of the people in photos hanging in the house. https://www.nytimes.com/2022/08/18/realestate/housing-discrimination-maryland.html

In the context of content moderation, I would suspect that AI is highly useful in pointing human moderators to the right topics to review, but is nearly useless in making independent moderation decisions.

Maybe? I mean "tons of people are using the rude n-word and c-word" could make it into the poorly filtered data set as "this is fine" or as "nope, flag that".

[maizefolk]

In the context of content moderation, I would suspect that AI is highly useful in pointing human moderators to the right topics to review, but is nearly useless in making independent moderation decisions.

You may well be right. Frankly it looks like Musk is leaning more heavily into crowd sourced fact correction and moderation than replacing human moderators with AI. (I base this on him cheering on the crowd sourced fact correction of the white house giving Biden credit for raising social security payouts a couple of days ago, so I could be completely wrong.)

My problem here isn't with people saying "AI won't solve the moderation problem." It might. It might not. I'd be fascinated to see it tried properly with modern Large Language Models but I wouldn't bet money either way on the outcome.

It's only with people saying "AI is racist."

[NorCal]

It's a lot more nuanced than human = bad and AI = good.

I'm perfectly happy to agree with you on this.

My argument above is an exercise in creating a mirror image of the argument that was being made both another that 1) AI is a monolith and 2) human = good and AI = bad.

Left to its own devices, AI generates a lot of outputs that we would consider highly racist or would otherwise offend our value systems.  In situations like mortgage lending, AI will generate outcomes that violate fair lending laws, even if race is not an input.

Edit: This is another example. Some AI models, trained with some datasets, will have racially disparate outcomes in their recommendations on  mortgage lending decisions. I completely agree.

But there is a big jump from this to "AI is necessarily racist".

And again. This isn't unique to AI models or even worse in AI models than having humans do the same work (which is typically the data used to train models). For example, look at how human beings appraise houses based on the race of the people who show them around the house and the race of the people in photos hanging in the house. https://www.nytimes.com/2022/08/18/realestate/housing-discrimination-maryland.html

In the context of content moderation, I would suspect that AI is highly useful in pointing human moderators to the right topics to review, but is nearly useless in making independent moderation decisions.

Maybe? I mean "tons of people are using the rude n-word and c-word" could make it into the poorly filtered data set as "this is fine" or as "nope, flag that".

Agreed.  Keyword blacklists are easy and useful.  I just view don't view that as "AI". 

[ender]

My prediction is that Twitter will be a business school case study in the quickest way to make $44B evaporate.

Musk has put himself in a situation where he's a guaranteed loser.

He's promised the right "absolute free speech".  Or at least the far right has interpreted it that way.  Anything he does to moderate content will now be considered a betrayal.  Even keeping Donald Trump banned will be considered a betrayal.

It seems a decent number of users on the left are already fed up and leaving the platform.  It may or may not be a full trend, but it could turn into one.  Network effects work in reverse too.  Just ask MySpace.

And the digital advertising market is cooling pretty rapidly anyways.  Advertisers are already broadly pulling back on their spending and don't need much incentive to move their money to different platforms.  Major brands would pull the plug if content moderation is even perceived to loosen up.  And Twitter probably just fired most of the team needed to keep up with their existing content moderation anyways.

And Musk put enough debt on the buy-out to pretty much wipe out Twitter's existing cash-flow.  It's just a question of whether the revenue declines are more or less than the savings from firing 3,500 people.

I feel like there's decent odds of this too.

I am also curious whether Twitter will even meaningfully survive from a technical perspective. As much as a lot of people love to trivialize things like "it's just X" building and operating Twitter is not actually a trivial endeavor.

[PDXTabs]

Maybe? I mean "tons of people are using the rude n-word and c-word" could make it into the poorly filtered data set as "this is fine" or as "nope, flag that".

And AI is probably bad at some of the more subtle social usage of certain words. Some groups, and even geographies, get a free pass on both of those words. I generally refrain from using the C word while in North America, but not elsewhere.

[NorCal]

Maybe? I mean "tons of people are using the rude n-word and c-word" could make it into the poorly filtered data set as "this is fine" or as "nope, flag that".

And AI is probably bad at some of the more subtle social usage of certain words. Some groups, and even geographies, get a free pass on both of those words. I generally refrain from using the C word while in North America, but not elsewhere.

I'm not a programmer myself, but I did work at a startup circa 2008-2009 that was solving some of these problems for a different industry.  Somewhere between 20-30% of the staff had PhD's in theoretical mathematics, linguistics, or nuero-linguistic programing.  Probably 60%+ had master's degrees in similar fields.

We had some fascinating discussions, although I was way too dumb for many of them. 

A lot of these problems have since been "solved" in different and automated ways.  But here's a hypothetical problem that automated systems would come up against:

1. Person X says something horrible that would result in an account suspension.
2. Person Y reiterates post X, but isn't clear if they support it or are shaming the original post.
3. Person Z reshares the content to point out that it is undesirable and should result in platform penalties.

All three of these people shared the same content.  Now go try and tell a computer how to tell differentiate between the acceptable and unacceptable posts. 

At my startup, we had software that could help with this type of problem.  But it wasn't a program you'd let loose on the data unsupervised.  You would have to train it, validate it, and have real people reviewing the decisions.  At best, you could get to something that was semi-automated.

[ChpBstrd]

Maybe? I mean "tons of people are using the rude n-word and c-word" could make it into the poorly filtered data set as "this is fine" or as "nope, flag that".

And AI is probably bad at some of the more subtle social usage of certain words. Some groups, and even geographies, get a free pass on both of those words. I generally refrain from using the C word while in North America, but not elsewhere.

I'm not a programmer myself, but I did work at a startup circa 2008-2009 that was solving some of these problems for a different industry.  Somewhere between 20-30% of the staff had PhD's in theoretical mathematics, linguistics, or nuero-linguistic programing.  Probably 60%+ had master's degrees in similar fields.

We had some fascinating discussions, although I was way too dumb for many of them. 

A lot of these problems have since been "solved" in different and automated ways.  But here's a hypothetical problem that automated systems would come up against:

1. Person X says something horrible that would result in an account suspension.
2. Person Y reiterates post X, but isn't clear if they support it or are shaming the original post.
3. Person Z reshares the content to point out that it is undesirable and should result in platform penalties.

All three of these people shared the same content.  Now go try and tell a computer how to tell differentiate between the acceptable and unacceptable posts. 

At my startup, we had software that could help with this type of problem.  But it wasn't a program you'd let loose on the data unsupervised.  You would have to train it, validate it, and have real people reviewing the decisions.  At best, you could get to something that was semi-automated.

In theory, wouldn't the AI operate faster than person Y can respond to or even read the content by person X?

Also, wouldn't the presence of AI change the culture so that people wouldn't feel the need to challenge such statements for the few minutes they are up, and wouldn't want to take the risk of being flagged themselves? Current culture is that "Someone is WRONG on the Internet and I Have to Fix It or Else it'll Be There Forever!"

[NorCal]

Maybe? I mean "tons of people are using the rude n-word and c-word" could make it into the poorly filtered data set as "this is fine" or as "nope, flag that".

And AI is probably bad at some of the more subtle social usage of certain words. Some groups, and even geographies, get a free pass on both of those words. I generally refrain from using the C word while in North America, but not elsewhere.

I'm not a programmer myself, but I did work at a startup circa 2008-2009 that was solving some of these problems for a different industry.  Somewhere between 20-30% of the staff had PhD's in theoretical mathematics, linguistics, or nuero-linguistic programing.  Probably 60%+ had master's degrees in similar fields.

We had some fascinating discussions, although I was way too dumb for many of them. 

A lot of these problems have since been "solved" in different and automated ways.  But here's a hypothetical problem that automated systems would come up against:

1. Person X says something horrible that would result in an account suspension.
2. Person Y reiterates post X, but isn't clear if they support it or are shaming the original post.
3. Person Z reshares the content to point out that it is undesirable and should result in platform penalties.

All three of these people shared the same content.  Now go try and tell a computer how to tell differentiate between the acceptable and unacceptable posts. 

At my startup, we had software that could help with this type of problem.  But it wasn't a program you'd let loose on the data unsupervised.  You would have to train it, validate it, and have real people reviewing the decisions.  At best, you could get to something that was semi-automated.

In theory, wouldn't the AI operate faster than person Y can respond to or even read the content by person X?

Also, wouldn't the presence of AI change the culture so that people wouldn't feel the need to challenge such statements for the few minutes they are up, and wouldn't want to take the risk of being flagged themselves? Current culture is that "Someone is WRONG on the Internet and I Have to Fix It or Else it'll Be There Forever!"

What you're talking about is effectively a pre-clearance system.  Each post would have to pass by some automated content moderater before showing up in other people's feeds.  Or something similar.  There's no technical reason this couldn't happen.  But that's not what these companies care about.

These companies want people to post on their platforms frequently.  They want anger and rage because that gets more views and is more frequently shared.  They know that the closer a post gets to the content moderation line (or even a hair over it) will get way more views/shares/likes/advertising dollars than a benign post that generates zero controversy.

These companies don't want to moderate content at all.  It's expensive and it takes away the content that generates views/shares and eyeball time.

These companies are just doing a different version of what Foxnews does.  Generate enough outrage until the advertisers get skittish.  Then moderate just enough until the advertisers come back. 

Tristan Harris is a good person to look up on the topic.  I can't find the original long-form interview I heard him do, but here's a shorter TED talk he did:

https://www.bing.com/videos/search?q=tristan+harris+ted+talk&&view=detail&mid=8DEAA696C3DC01BA32A68DEAA696C3DC01BA32A6&&FORM=VRDGAR&ru=%2Fvideos%2Fsearch%3Fq%3Dtristan%2Bharris%2Bted%2Btalk%26qpvt%3Dtristan%2Bharris%2Bted%2Btalk%26FORM%3DVDRE

[maizefolk]

According to musk (so take it for what it is worth), twitter was losing $4M/day. However, last quarter twitter was still a public company and they reported losing $344M in three months. Roughly $1.4B annually so $1.5B seems plausible.

Twitter laid off 3,700 people. The big tech company I'm most familiar with uses $400k as a rule of thumb for their total annual cost of a silicon valley based FTE. If that rule of thumb also applies to twitter laying off 3,700 people would save $1.5B/year. That would put twitter back at roughly break even and puts their annual burn rate at approximately where it was in Q2 of 2021 and the total number of twitter employees at roughly where it was in 2018.

Of course that doesn't leave twitter any buffer to absorb either 1) advertiser boycotts or 2) the $1B/year in additional debt service payments owed as part of Musk's leveraged buyout without going back into the red.

[Gremlin]

It's a lot more nuanced than human = bad and AI = good.

I'm perfectly happy to agree with you on this.

My argument above is an exercise in creating a mirror image of the argument that was being made both another that 1) AI is a monolith and 2) human = good and AI = bad.

Left to its own devices, AI generates a lot of outputs that we would consider highly racist or would otherwise offend our value systems.  In situations like mortgage lending, AI will generate outcomes that violate fair lending laws, even if race is not an input.

Edit: This is another example. Some AI models, trained with some datasets, will have racially disparate outcomes in their recommendations on  mortgage lending decisions. I completely agree.

But there is a big jump from this to "AI is necessarily racist".

And again. This isn't unique to AI models or even worse in AI models than having humans do the same work (which is typically the data used to train models). For example, look at how human beings appraise houses based on the race of the people who show them around the house and the race of the people in photos hanging in the house. https://www.nytimes.com/2022/08/18/realestate/housing-discrimination-maryland.html

This is something I happen to have a fair bit of knowledge about.

What's implied in your comment is that it's a problem with the training datasets in these cases.

Which is absolutely true (along with factor creation and filtering, amongst other things).  One of the biggest issues that exists in creating appropriate training sets, is that they are, by their very nature, biased.  The biases may very well be unintended.  In many circumstances, they may be considered to be unavoidable or acceptable (and in many cases, depending on what you're decisioning, this may be an appropriate call).

Where you have data, those biases can be measured and understood - whether they are ignored, controlled for, or eliminated is often a trade-off between ethics and financial realities.  Where there is a data gap bias (ie a group of individuals whose data is severely underrepresented or even completely unrepresented in the training data), this can become an intractable problem very quickly.  By definition, ML models hunt for anomalies and treat them differently.  If you look like a data gap, you will be treated differently, and almost certainly, adversely.

Even 'universal' datasets suffer from this.  Models built solely on census data, probably the most 'universal' datasets out there, are often notoriously biased against the homeless.  They are data gaps in the training set. 

It's not the algorithm's fault - modelers will often talk about 'garbage in, garbage out' as a mantra against poor quality data.  But it's equally true that 'bias in, means bias out'.  And I've never, ever seen a dataset that controls for data gap bias.  I remain skeptical that it's even possible for one to exist.  Your post implies that this should be happening as a matter of course - I'll challenge that and suggest that if you have a solution for this problem, you'd better patent it, because you will be a billionaire if you do.

So if none of the datasets out there are truly unbiased, and the algorithms are incapable of assessing 'data gap bias', then the algorithm built will necessarily be biased.  Because you can't dissociate the AI decisioning from the data is was built from, then the AI is biased.

Depending on the context, significant data gap biases exist for certain races, the homeless, women, minors, people escaping domestic violence, those in witness protection, immigrants, LGBTQI+ and seniors.  Oh, and people named Karen (quite seriously!)...

[maizefolk]

This is something I happen to have a fair bit of knowledge about.

What's implied in your comment is that it's a problem with the training datasets in these cases.

Which is absolutely true (along with factor creation and filtering, amongst other things).  One of the biggest issues that exists in creating appropriate training sets, is that they are, by their very nature, biased.  The biases may very well be unintended.  In many circumstances, they may be considered to be unavoidable or acceptable (and in many cases, depending on what you're decisioning, this may be an appropriate call).

Where you have data, those biases can be measured and understood - whether they are ignored, controlled for, or eliminated is often a trade-off between ethics and financial realities.  Where there is a data gap bias (ie a group of individuals whose data is severely underrepresented or even completely unrepresented in the training data), this can become an intractable problem very quickly.  By definition, ML models hunt for anomalies and treat them differently.  If you look like a data gap, you will be treated differently, and almost certainly, adversely.

Even 'universal' datasets suffer from this.  Models built solely on census data, probably the most 'universal' datasets out there, are often notoriously biased against the homeless.  They are data gaps in the training set. 

It matters what applications we are talking about. If you can truly randomly sample from the same population you'll be evaluating on to generate training data, it is possible to avoid a lot of problems of data gaps that would otherwise be present <-- Obviously this doesn't solve the problem of how to get accurate and unbiased labels for that data, as discussed above but you're bringing up the new point of undersampling of specific groups.

I agree with you. Trying to get a random sample of all human beings in the USA is a very hard problem. That's why political polling is broken. That's why, as you point out, despite hiring more than 600,000 people for months, the US census still misses lots of people and systematically under counts certain groups. Another example I'm aware of is that something like 8-10% of adult black men are missed in a typical census.

Trying to get a random sample of all tweets on twitter is a significantly less hard problem since all the tweets exist within a single database. The big issue one could potentially run into is that any training dataset will necessarily be generated from past tweets and, in order to be useful, the model will need to be applied to future tweets. So shifts in language/topic could cause problems if the dataset isn't being consistently updated and revalidated.

Approaches like zero shot learning, which are starting to show up more in natural language processing, should also be more robust to out-of-training-data-range cases than a lot of historically common ML approaches. But I don't think we need to bring those into our assumptions in a simple case like all the tweets on twitter where it really is possible to select a true random samples of the population of interest.

It's not the algorithm's fault - modelers will often talk about 'garbage in, garbage out' as a mantra against poor quality data.  But it's equally true that 'bias in, means bias out'.  And I've never, ever seen a dataset that controls for data gap bias.  I remain skeptical that it's even possible for one to exist.  Your post implies that this should be happening as a matter of course - I'll challenge that and suggest that if you have a solution for this problem, you'd better patent it, because you will be a billionaire if you do.

So if none of the datasets out there are truly unbiased, and the algorithms are incapable of assessing 'data gap bias', then the algorithm built will necessarily be biased.  Because you can't dissociate the AI decisioning from the data is was built from, then the AI is biased.

Depending on the context, significant data gap biases exist for certain races, the homeless, women, minors, people escaping domestic violence, those in witness protection, immigrants, LGBTQI+ and seniors.  Oh, and people named Karen (quite seriously!)...

I'm not sure how I implied that controlling for biases in data occurs "as a matter of course." I agree that in many cases getting a real and accurate training dataset is quite hard if not impossible. My point was that the biases were driven by the training data (typically generated by humans) rather than something inherent in AI/ML approaches. There are approaches to work around not having important examples/groups in ones training dataset. But bias from human labelers is harder to completely compensate for.

Here is seems useful to ask how unbiased should AI need to be before we're comfortable using it? Should it be no bias ever? Or just provably no more biased (and ideally less biased) than the typical human who would otherwise be asked to make the same decisions and judgement calls?

[SotI]

If Elon brings in a wider range of content moderators to minimize group think, I would consider this beneficial.

That seems pretty difficult.  For example, how do you minimize group think when a major political party's platform is a claim they have zero credible evidence to support?

I guess I am looking at it from a European and not US-centric view. I am used to working internationally and find a variety of views helpful, as hardly anything is ever "either this or that". 

[seattlecyclone]

According to musk (so take it for what it is worth), twitter was losing $4M/day. However, last quarter twitter was still a public company and they reported losing $344M in three months. Roughly $1.4B annually so $1.5B seems plausible.

Twitter laid off 3,700 people. The big tech company I'm most familiar with uses $400k as a rule of thumb for their total annual cost of a silicon valley based FTE. If that rule of thumb also applies to twitter laying off 3,700 people would save $1.5B/year. That would put twitter back at roughly break even and puts their annual burn rate at approximately where it was in Q2 of 2021 and the total number of twitter employees at roughly where it was in 2018.

Of course that doesn't leave twitter any buffer to absorb either 1) advertiser boycotts or 2) the $1B/year in additional debt service payments owed as part of Musk's leveraged buyout without going back into the red.

From the sound of things, the layoff was done very haphazardly. A lot of institutional memory is simply gone at this point. The remaining employees are surely facing a combination of a higher workload and a dramatic shift in corporate culture. What percentage of those who remain are not actively circulating their resume at this point? Employee attrition is sure to follow. At one of these complex tech companies there's only so big of a percentage of your staff you can afford to lose before the reliability of the various services starts to suffer. User attrition is also likely to occur. Plenty of folks don't trust Elon Musk to safeguard user privacy to the same extent as previous management did. Fewer users plus increased downtime is not a winning combination re: advertiser dollars. I feel sorry for those workers let go into a recessionary environment, and I worry about the effect on the already-depressed labor market that will result from this many qualified tech people looking for employment at the same time. Those concerns aside, I'm kind of sitting back with a bowl of popcorn, watching one man try to light $44 billion on fire as quickly as possible.

[LennStar]

Who wants to bet that "being Musked" become a phrase for self-destructive firing?

Also, if the screenshot was not faked, Musk has promised the "nuclear" shame option against advertisers that no longer want to advertise for lack of moderation of alt-rights. Nice business you have there, would be such a shame if anything happens to it...

Also he doesn't understand that most people are not as egozentric as he and don't care for the blue check, definitely not enough to pay 100 bucks a year for it. Or even if they did, he is devaluing it immensely, not least by making "pay" the only identity check.

Even if nothing good comes out of it, at least it will make a heck of a case study for business schools.

[nick663]

According to musk (so take it for what it is worth), twitter was losing $4M/day. However, last quarter twitter was still a public company and they reported losing $344M in three months. Roughly $1.4B annually so $1.5B seems plausible.

Twitter laid off 3,700 people. The big tech company I'm most familiar with uses $400k as a rule of thumb for their total annual cost of a silicon valley based FTE. If that rule of thumb also applies to twitter laying off 3,700 people would save $1.5B/year. That would put twitter back at roughly break even and puts their annual burn rate at approximately where it was in Q2 of 2021 and the total number of twitter employees at roughly where it was in 2018.

Of course that doesn't leave twitter any buffer to absorb either 1) advertiser boycotts or 2) the $1B/year in additional debt service payments owed as part of Musk's leveraged buyout without going back into the red.

There is also the assumption in there that the 3,700 people weren't bringing any additional value to the company.  Cutting them will certainly have an impact on revenue eventually which means more cuts have to be made.

Also, a layoff of that size (along with Elon's public personality) will make it hard to recruit people if they do want to hire.  Now they will have to pay more while picking from a smaller pool of candidates as "Twitter" will not be the selling point it used to be.

[NorCal]

According to musk (so take it for what it is worth), twitter was losing $4M/day. However, last quarter twitter was still a public company and they reported losing $344M in three months. Roughly $1.4B annually so $1.5B seems plausible.

Twitter laid off 3,700 people. The big tech company I'm most familiar with uses $400k as a rule of thumb for their total annual cost of a silicon valley based FTE. If that rule of thumb also applies to twitter laying off 3,700 people would save $1.5B/year. That would put twitter back at roughly break even and puts their annual burn rate at approximately where it was in Q2 of 2021 and the total number of twitter employees at roughly where it was in 2018.

Of course that doesn't leave twitter any buffer to absorb either 1) advertiser boycotts or 2) the $1B/year in additional debt service payments owed as part of Musk's leveraged buyout without going back into the red.

There is also the assumption in there that the 3,700 people weren't bringing any additional value to the company.  Cutting them will certainly have an impact on revenue eventually which means more cuts have to be made.

Also, a layoff of that size (along with Elon's public personality) will make it hard to recruit people if they do want to hire.  Now they will have to pay more while picking from a smaller pool of candidates as "Twitter" will not be the selling point it used to be.

If I know one thing about recruiters, they know which companies to target for a large volumes of leads. 

Even in this hiring environment, I guarantee most Twitter employees are getting multiple LinkedIn messages a day with job opportunities.  And what good reason would a Twitter employee have for staying at this point?

I can think of very few reasons the best will stay.  Those that remain will probably be those promoted above their level of competence when their boss was fired.  Or maybe a few that are waiting for equity to vest (assuming they didn't completely mess up the equity-comp plan in the acquisition).

It kinda reminds me of the corporate environment you see when a Private Equity company tries to jump in and play Silicon Valley.  The best talent is gone in a year and the remaining shell of a company can no longer execute or grow.

[Paul der Krake]

The laid off employees who put their names on layoffs.fyi are currently being bombarded by series A/B/C startups who are still hiring. It's not clear how long that's going to last.

They will struggle to get jobs that pay close to what Twitter was paying them, because that tier of company has mostly stopped hiring. Goodbye cushy $350k comp package for 6 years of engineering experience, hello industry average $175k.

A lot of people are very quick to declare that what Elon Musk is doing is an unmitigated disaster, when the dude is quite possibly the most successful serial entrepreneur the world has ever seen. It could very well be a total disaster, but it's way too early to tell. 

[Paul der Krake]

Somewhat related, I work for another high profile tech company that also did layoffs last week, and let me tell you, the internal vibes are not good. The younger grads especially are taking it pretty hard.

It's quite different from the Twitter case because:
- it was handled extremely well
- the severance is excellent
- the business case for doing the layoffs is strong

Yet it still blows big time for everyone, even those who weren't cut. Barely any work got done in the days following the announcement.

[clifp]

Somewhat related, I work for another high profile tech company that also did layoffs last week, and let me tell you, the internal vibes are not good. The younger grads especially are taking it pretty hard.

It's quite different from the Twitter case because:
- it was handled extremely well
- the severance is excellent
- the business case for doing the layoffs is strong

Yet it still blows big time for everyone, even those who weren't cut. Barely any work got done in the days following the announcement.

I'm not sure there is a good way of doing large layoffs. It seems is mostly choosing the least bad one.  One company I worked at cut 5-10% every month or two for a year before eventually going bankrupt. I am pretty sure that's the worse way the organization is in a constant state of fear and rumors are rampant. 

A couple things worth noting, when Jobs came back to Apple in 1997, he laid off 4,100 employees, 31% of the workforce.  Evidently, 50% RIF gets Twitter to breakeven.  Yes, ad revenue will go down, so either more cuts or more revenue are needed.

The WARN law required 60 days warning or severance in lieu of the warning. Twitter gave their employees 14 weeks, 5 weeks more than required.

Twitter started in 2006, I joined in 2008, Musk in 2009.  In all those years the only impactful improvements to the platform I saw was going from 140-280 characters. blue check marks, and adding (limited) video.  People have been begging for a edit button since the beginning.  People who think Elon has no idea what he is doing aren't paying attention.  I'd argue that other than Donald Trump, nobody has been a more successful Twitter user than Elon Musk. His 115 million followers, dwarfs the number of followers of previous Richest Man in the world, like Bezos or Gates.

Elon's Twitter presence, in 2012 was factor in my forking over more than twice as much as I'd ever spent for a car, to an unproven car manufacturer.  (I ordered it Dec 31,2012 cause the price went up $10K Jan 1) . I watch him fix Telsa problem by paying attention to complaints, specifically the high price of body repairs in Model S. They are still expensive to fix body damage, but way better than before he got involved.

Elon makes plenty of juvenile, offensive, half-baked, and some just stupid tweets, although not has many as Trump has done.  However, he never sounds corporate and is seldom boring.

Finally, Twitter was ridiculously overstaffed, they had 1,500 involved in moderation and amazing 3,023 engineers.  In contrast, SpaceX has 12,000 employees with I'm guessing a similar 3,000-4,000 engineers.  By any metric SpaceX engineers have made an order of magnitude more innovations than Twitter engineers.  To paraphrase Churchill about Twitter engineers. Never in the course of engineering history, have so many, done so little, for so long.

Elon Musk, won't let the remaining 1,500 or so engineers take 15 years to add a bloody edit button to the platform.

[LennStar]

The edit button was never a techical issue but a policy one. As with many other non-changes. You might find them wrong, but that doesn't make the engineers bad.
Though I admit that 3000 is an awful lot for a better chat server. Mastodon definitely had less ;)

Which brings me to the main point why I think you are wrong. The worth of twitter is in it's users and that it is the only (widely known) type of it's kind, at least in the "West".
But even so about 1/3 of the regularily active people I follow have opened up mastodon accounts in the last week. It readily admit my bubble is heavy on people prone to do such a thing (privacy advocates, netpolicy nerds etc.), but I also have seen several people who definitely don't fall into that category, like artists.

People are ready to go, and even if they don't, they certainly won't pay for the Blue. Which means their tweets will be ranked very very badly.
Especially for the artists that is a huge thing.

I hold up my opinion that Musk is on a good way to make Twitter a company that is worth 1/10th of what he paid for it.

[Sibley]

I'm also seeing that some people who were laid off are being asked if they'll come back. Not a sign that the layoff was done well.

https://www.businessinsider.com/some-tweeps-already-being-asked-to-come-back-to-twitter-2022-11

And re the AI - I don't think its a good idea to rely entirely on AI for content moderation, for ANY social media site. There's too much nuance in human communication, AI is going to get it wrong at times. People will also get it wrong at times. AI to cover the easy stuff, and people involved to review the AI on the hard stuff, will help. It won't be perfect of course, but probably better than either alone.

I did see some discussion on Twitter of alternatives. Apparently there's something called Mastodon that some are trying out, at least in the very small corner of Twitter that I see. Name is cool at least, I know nothing else about it.

[clifp]

The edit button was never a techical issue but a policy one. As with many other non-changes. You might find them wrong, but that doesn't make the engineers bad.
Though I admit that 3000 is an awful lot for a better chat server. Mastodon definitely had less ;)

Which brings me to the main point why I think you are wrong. The worth of twitter is in it's users and that it is the only (widely known) type of it's kind, at least in the "West".
But even so about 1/3 of the regularily active people I follow have opened up mastodon accounts in the last week. It readily admit my bubble is heavy on people prone to do such a thing (privacy advocates, netpolicy nerds etc.), but I also have seen several people who definitely don't fall into that category, like artists.

People are ready to go, and even if they don't, they certainly won't pay for the Blue. Which means their tweets will be ranked very very badly.
Especially for the artists that is a huge thing.

I hold up my opinion that Musk is on a good way to make Twitter a company that is worth 1/10th of what he paid for it.

No question that value of the Twitter is its users, in particular journalists and politicians.  In the early days it was ok for learning about tech.  But Twitter's search has always been bad, and it has only gotten worse since virtually nobody uses hashtags anymore. So the discovery process is broken.   As for artists, Instagram is vastly better, and Pinterest is geared toward artist and has nearly as many users as Twitter.

I just took a look at Mastodon, it ain't the future. even with an 83% jump in users this month to 843K MAU, is closer in size to MMM than to Twitter.  Asking people to choose which one of 50 servers to join is worse than Discourse, where you normally just get an invite.

No question Elon overpaid, bad timing on his part, and being reckless on his acquisition terms.  I'm not sure if ever had the chance to walk away and just pay a $1 billion but that was certainly the financially smart thing to do.

As for the engineers, I don't blame the engineers (although Twitter does seem like a great place to quiet quit in Silicon Valley)  But the engineering management and business management have been pathetic. There are probably a dozen policy decision around an edit button. At some point you choose 2 or 3 roll them out to subset of the user community and settle on the one that people like the best, rinse and repeat.  This is how Google, Facebook, Amazon and the rest of the tech world make decisions, it ain't rocket sceince.

I have no idea how this will play out financially for Elon. All I'm saying is that I expect to see the platform functionality improve more over the next 2 years than the last 14.  At some point, for enough features and low enough price point $5 or $8 I'll consider upgrading.

[maizefolk]

Twitter started in 2006, I joined in 2008, Musk in 2009.  In all those years the only impactful improvements to the platform I saw was going from 140-280 characters. blue check marks, and adding (limited) video.  People have been begging for a edit button since the beginning.  People who think Elon has no idea what he is doing aren't paying attention.  I'd argue that other than Donald Trump, nobody has been a more successful Twitter user than Elon Musk. His 115 million followers, dwarfs the number of followers of previous Richest Man in the world, like Bezos or Gates.

I don't disagree with the overall point you are making (not a lot of innovation or new features to show for that many SWEs).

But I think I joined roughly the same time as you and thinking back on it the service has changed a lot more than that in the last 14 years.

When I first joined retweets weren't officially supported the way they are today. Someone would just write "RT @SomeUser Text of original tweet". And it was a lot time after "official" retweets became a thing that quote tweeting was added as an option. Embedding even still photos wasn't originally supported people would link out to third-party image hosting. They also rewrote their whole backend from Ruby on Rails to Scala/Java to fix scaling problems.

A bunch of their "innovation" in recent years has been in how to show people tweets in their timeline that keep them engaged and scrolling (a person who one of your followers once liked replied to a tweet so we're showing it in your timeline). I'd argue that has actually made the use experience a lot worse (although at least the simply chronological view is still an option). Anyway. Like I said, not disagreeing with your conclusion. It was just interesting to remember what twitter has and hasn't changed in that long period of time.

[nick663]

Elon makes plenty of juvenile, offensive, half-baked, and some just stupid tweets, although not has many as Trump has done.

This is an incredibly low benchmark to use haha.

Finally, Twitter was ridiculously overstaffed, they had 1,500 involved in moderation and amazing 3,023 engineers.  In contrast, SpaceX has 12,000 employees with I'm guessing a similar 3,000-4,000 engineers.  By any metric SpaceX engineers have made an order of magnitude more innovations than Twitter engineers.  To paraphrase Churchill about Twitter engineers. Never in the course of engineering history, have so many, done so little, for so long.

Those are 2 very different businesses.  I don't think you can begin to compare them like that.

[PDXTabs]

Finally, Twitter was ridiculously overstaffed, they had 1,500 involved in moderation and amazing 3,023 engineers.  In contrast, SpaceX has 12,000 employees with I'm guessing a similar 3,000-4,000 engineers.  By any metric SpaceX engineers have made an order of magnitude more innovations than Twitter engineers.  To paraphrase Churchill about Twitter engineers. Never in the course of engineering history, have so many, done so little, for so long.

Those are 2 very different businesses.  I don't think you can begin to compare them like that.

Sure, but by comparison Wikipedia has less than 300 staff including contractors AFAIK.
https://en.wikipedia.org/wiki/Wikipedia:Wikimedia_Foundation

I'm not sure that Twitter actually needs 3k engineers. But maybe? We'll see.

[clifp]

Twitter started in 2006, I joined in 2008, Musk in 2009.  In all those years the only impactful improvements to the platform I saw was going from 140-280 characters. blue check marks, and adding (limited) video.  People have been begging for a edit button since the beginning.  People who think Elon has no idea what he is doing aren't paying attention.  I'd argue that other than Donald Trump, nobody has been a more successful Twitter user than Elon Musk. His 115 million followers, dwarfs the number of followers of previous Richest Man in the world, like Bezos or Gates.

But I think I joined roughly the same time as you and thinking back on it the service has changed a lot more than that in the last 14 years.

When I first joined retweets weren't officially supported the way they are today. Someone would just write "RT @SomeUser Text of original tweet". And it was a lot time after "official" retweets became a thing that quote tweeting was added as an option. Embedding even still photos wasn't originally supported people would link out to third-party image hosting. They also rewrote their whole backend from Ruby on Rails to Scala/Java to fix scaling problems.

A bunch of their "innovation" in recent years has been in how to show people tweets in their timeline that keep them engaged and scrolling (a person who one of your followers once liked replied to a tweet so we're showing it in your timeline). I'd argue that has actually made the use experience a lot worse (although at least the simply chronological view is still an option). Anyway. Like I said, not disagreeing with your conclusion. It was just interesting to remember what twitter has and hasn't changed in that long period of time.

I honestly don't remember the tweets and retweets changing, but I assume you are right.  I don't personally post many pictures much less videos, but there have been improvements on media. So I'm guilty of hyperbole no question. 

Still, there is a laundry list of things that are frustrating about the platform.  If I want to find out what you said about topic, I use the forum search. For Twitter I use Google, but often I just give up, and rely on my faulty memory of some interesting factoid I learned on Twitter.   Google is so easy to use that I'm not very forgiving of folks who don't use it for fact-checking.  For instance on another forum someway said Tesla is way less profitable that Ford or GM.   A quick Google check showed that uh no Tesla made 3.29 billion vs GM 3.3 billion and Ford 800 million dollar loss, last quarter. There is no such thing as a quick Twitter check of what Elon said, you have to go back and read of all of his Tweets.

The advertisers who pulled out of Twitter made a mistake I think.  The discussion of what Elon will do with Twitter, has eclipsed politics on Twitter.  This is 2 days before a midterm  crazy.

[maizefolk]

The laid off employees who put their names on layoffs.fyi are currently being bombarded by series A/B/C startups who are still hiring. It's not clear how long that's going to last.

They will struggle to get jobs that pay close to what Twitter was paying them, because that tier of company has mostly stopped hiring. Goodbye cushy $350k comp package for 6 years of engineering experience, hello industry average $175k.

A lot of people are very quick to declare that what Elon Musk is doing is an unmitigated disaster, when the dude is quite possibly the most successful serial entrepreneur the world has ever seen. It could very well be a total disaster, but it's way too early to tell.

Did you see the news about Meta?

Meta Platforms Inc. is planning to begin large-scale layoffs this week, according to people familiar with the matter, in what could be the largest round in a recent spate of tech job cuts after the industry’s rapid growth during the pandemic.

The layoffs are expected to affect many thousands of employees and an announcement is planned to come as soon as Wednesday, according to the people. Meta reported more than 87,000 employees at the end of September.

I'm guessing Meta is also in the tier of company paying people $300k+ salaries, so there may be even more people competing for the same jobs the laid off twitter folks are being approached about.

Agreed with your last point. Will be interesting to wait and see how things turn out. I'm grateful to not have my own livelihood riding on the outcome though.

[Telecaster]

The $8/month to keep verified status seems like a smart move. It's not financially burdensome to the majority of verified accounts. A quick google search says there are 420,000 verified accounts. If half of them opt to pay $8/month, that's a quick $20MM annual profit with no added expense.

Musk wouldn't have bought Twitter if he didn't see it benefitting him in the long run. He knows how to provide customers with a superior product, so I wouldn't be surprised to see a Twitter IPO for 5x once his improvements are fully in place. Everything you're seeing in the media now is hyped up to get clicks and Musk as the villain is going to get far more attention than Musk as the hero.

I'm not so sure about either of those.   For one, Twitter needs more like $1 billion, probably more like $2 billion in new revenue.   The people with checkmarks are usually famous or notable people--which are the people the rest of us proles want to follow.   Losing half of those might be a death blow to the business model.   But even if the checkmarks double it still wouldn't move the needle financially. 

Twitter has been slow to grow users, hence slow to grow revenue.   I'm not sure what would cause people who already aren't on Twitter to flock to Twitter.   Musk's threats to rain down thermonuclear war on his customers probably won't help.   So I don't see a 10X in revenue, which is about what it would take to get to 5x the buyout price.   

[Paul der Krake]

The laid off employees who put their names on layoffs.fyi are currently being bombarded by series A/B/C startups who are still hiring. It's not clear how long that's going to last.

They will struggle to get jobs that pay close to what Twitter was paying them, because that tier of company has mostly stopped hiring. Goodbye cushy $350k comp package for 6 years of engineering experience, hello industry average $175k.

A lot of people are very quick to declare that what Elon Musk is doing is an unmitigated disaster, when the dude is quite possibly the most successful serial entrepreneur the world has ever seen. It could very well be a total disaster, but it's way too early to tell.

Did you see the news about Meta?

Meta Platforms Inc. is planning to begin large-scale layoffs this week, according to people familiar with the matter, in what could be the largest round in a recent spate of tech job cuts after the industry’s rapid growth during the pandemic.

The layoffs are expected to affect many thousands of employees and an announcement is planned to come as soon as Wednesday, according to the people. Meta reported more than 87,000 employees at the end of September.

I'm guessing Meta is also in the tier of company paying people $300k+ salaries, so there may be even more people competing for the same jobs the laid off twitter folks are being approached about.

Agreed with your last point. Will be interesting to wait and see how things turn out. I'm grateful to not have my own livelihood riding on the outcome though.

Just saw the Meta (unconfirmed) leak, and frankly can't say I'm surprised. There have been industry rumors for months, performance reviews are being ratcheted upwards, and morale is low. I heard from a couple people a few days ago that work travel for employees was being canceled on short notice, and that's rarely a good sign.

And yes, Meta is one of the highest paying firms for engineering talent, and their hiring bar is very high.

I recommend reading this open letter to Zuck from a couple weeks ago. Technically it's just one dude's opinion about one company, but there is a huge subtext. A lot of industry signs that his broader point is taken very seriously by the entire industry.

[Sibley]

Did you see the news about Meta?

Meta Platforms Inc. is planning to begin large-scale layoffs this week, according to people familiar with the matter, in what could be the largest round in a recent spate of tech job cuts after the industry’s rapid growth during the pandemic.

The layoffs are expected to affect many thousands of employees and an announcement is planned to come as soon as Wednesday, according to the people. Meta reported more than 87,000 employees at the end of September.

I'm guessing Meta is also in the tier of company paying people $300k+ salaries, so there may be even more people competing for the same jobs the laid off twitter folks are being approached about.

Agreed with your last point. Will be interesting to wait and see how things turn out. I'm grateful to not have my own livelihood riding on the outcome though.

Well, I guess Zuckerburg is getting the squeeze because his pet project is spending like mad and not taking off. There are downsides to being a public company.

[Sibley]

I recommend reading this open letter to Zuck from a couple weeks ago. Technically it's just one dude's opinion about one company, but there is a huge subtext. A lot of industry signs that his broader point is taken very seriously by the entire industry.

That is pretty amazing. I am not current with that industry, but I get the impression that they're confused or underwhelmed re the VR stuff and are telling Zuck to come back to earth.

[LennStar]

Finally, Twitter was ridiculously overstaffed, they had 1,500 involved in moderation and amazing 3,023 engineers.  In contrast, SpaceX has 12,000 employees with I'm guessing a similar 3,000-4,000 engineers.  By any metric SpaceX engineers have made an order of magnitude more innovations than Twitter engineers.  To paraphrase Churchill about Twitter engineers. Never in the course of engineering history, have so many, done so little, for so long.

Those are 2 very different businesses.  I don't think you can begin to compare them like that.

Sure, but by comparison Wikipedia has less than 300 staff including contractors AFAIK.
https://en.wikipedia.org/wiki/Wikipedia:Wikimedia_Foundation

I'm not sure that Twitter actually needs 3k engineers. But maybe? We'll see.

That comparisn is so wrong on so many levels ;)
First of all you are not comparing to Wikipedia, you are comparing to the Wikimedia Foundation. Those two are very emphatically 2 different things. So much that afaik even today, even though many want it, even though they have enough money, the Foundation has not paid a single staffer to write anything on Wikipedia, not even fact-checking about people (like did he really had an affair and other potentionally harmful things).
And actually the biggest problem for Wikipedia is getting writers. Part of it is that the tech is still very newbie-unfriendly and another part is the "unfriendly" male dominated athmospere.
 

[Moonwaves]

The edit button was never a techical issue but a policy one. As with many other non-changes. You might find them wrong, but that doesn't make the engineers bad.
Though I admit that 3000 is an awful lot for a better chat server. Mastodon definitely had less ;)

Which brings me to the main point why I think you are wrong. The worth of twitter is in it's users and that it is the only (widely known) type of it's kind, at least in the "West".
But even so about 1/3 of the regularily active people I follow have opened up mastodon accounts in the last week. It readily admit my bubble is heavy on people prone to do such a thing (privacy advocates, netpolicy nerds etc.), but I also have seen several people who definitely don't fall into that category, like artists.

People are ready to go, and even if they don't, they certainly won't pay for the Blue. Which means their tweets will be ranked very very badly.
Especially for the artists that is a huge thing.

I hold up my opinion that Musk is on a good way to make Twitter a company that is worth 1/10th of what he paid for it.

I set up a mastodon account at the weekend (on the mastodon.ie instance, there was lots of fun and games over the weekend as a big part of Irish twitter moved over). Of course lots of people in twitter's Dublin office just lost or will be losing their jobs, so there's a fair amount of solidarity for them involved.

I have to say, so far it seems to be delightful. And no quote-tweet equivalent, which is great, I think. If you want to respond to a toot, you have to engage with the person who wrote it.

[Telecaster]

That comparisn is so wrong on so many levels ;)
First of all you are not comparing to Wikipedia, you are comparing to the Wikimedia Foundation. Those two are very emphatically 2 different things. So much that afaik even today, even though many want it, even though they have enough money, the Foundation has not paid a single staffer to write anything on Wikipedia, not even fact-checking about people (like did he really had an affair and other potentionally harmful things).
And actually the biggest problem for Wikipedia is getting writers. Part of it is that the tech is still very newbie-unfriendly and another part is the "unfriendly" male dominated athmospere.

Twitter doesn't pay its content creators either.   

[PDXTabs]

Finally, Twitter was ridiculously overstaffed, they had 1,500 involved in moderation and amazing 3,023 engineers.  In contrast, SpaceX has 12,000 employees with I'm guessing a similar 3,000-4,000 engineers.  By any metric SpaceX engineers have made an order of magnitude more innovations than Twitter engineers.  To paraphrase Churchill about Twitter engineers. Never in the course of engineering history, have so many, done so little, for so long.

Those are 2 very different businesses.  I don't think you can begin to compare them like that.

Sure, but by comparison Wikipedia has less than 300 staff including contractors AFAIK.
https://en.wikipedia.org/wiki/Wikipedia:Wikimedia_Foundation

I'm not sure that Twitter actually needs 3k engineers. But maybe? We'll see.

First of all you are not comparing to Wikipedia, you are comparing to the Wikimedia Foundation. Those two are very emphatically 2 different things. So much that afaik even today, even though many want it, even though they have enough money, the Foundation has not paid a single staffer to write anything on Wikipedia, not even fact-checking about people (like did he really had an affair and other potentionally harmful things).
And actually the biggest problem for Wikipedia is getting writers. Part of it is that the tech is still very newbie-unfriendly and another part is the "unfriendly" male dominated athmospere.

Indeed, but what does that have to do with how many engineers you need to maintain your data-centers and keep your software up-to-date?

[GuitarStv]

Does anyone find it odd that Musk is fully endorsing Republicans after saying that Twitter needed to be politically neutral?

https://www.reuters.com/world/us/musk-recommends-voting-republicans-us-midterm-elections-tweet-2022-11-07/

[NorCal]

Does anyone find it odd that Musk is fully endorsing Republicans after saying that Twitter needed to be politically neutral?

https://www.reuters.com/world/us/musk-recommends-voting-republicans-us-midterm-elections-tweet-2022-11-07/

I think his behavior can be explained in one of two ways:
1. He is completely unaware how his actions are being perceived.
2. He saw how much eyeball-share Donald Trump added to Twitter from 2016-2020, and he is trying to recreate that experience on the platform by emulating Trump's antics in his own way.

Given how much attention we're giving him even in our little remote corner of the internet, I'm voting for option #2. 

[LennStar]

Does anyone find it odd that Musk is fully endorsing Republicans after saying that Twitter needed to be politically neutral?

https://www.reuters.com/world/us/musk-recommends-voting-republicans-us-midterm-elections-tweet-2022-11-07/

No, he has done so before. And he give a shit about what he said yesterday anyway.
Just a few days ago he said he would go nuclear against the advertiser that were presured by left groups into no longer paying for ads. Other tweets of him are full of far right vocabulary (though by now it's mainstream Republican lingo).
It was less obvious before be bought the thing, so I guess there is some validity in it that it is a conscious(?) effort to get in the right wing groups as paying customers. But he was fairly anti-left from teh start. No wonder, since they want him, the richest man on earth, to pay taxes!!! For lazy shitbags that don't even work 10 hours a day and cannot afford a Tesla.

[ministashy]

For all this talk of AIs and the like, the stuff I've been hearing from the experts is that Musk has two nearly insurmountable problems:

1.  Social media platforms pay content creators in order to attract users, so that they can sell their eyeballs to advertisers.  They don't demand content creators pay them for that privilege.  And Twitter has a relatively tiny userbase already compared to Facebook, TikTok, Snapchat, Youtube, and so on, so they're already starting from a disadvantaged position.  Musk demanding that verified users pay for their accounts is going to drive away any hip new content creators to other platforms (and probably more than a few established ones as well).

2.  No matter how much or little moderation Musk puts in place, he's going to get backlash.  If he moderates content to please advertisers, the right wing is going to see this as a betrayal and leave en masse.  If he doesn't moderate content to please the right wing, advertisers are going to leave en masse, because no company in their right mind wants to be associated with racism, homophobia, etc.

[Sibley]

Does anyone find it odd that Musk is fully endorsing Republicans after saying that Twitter needed to be politically neutral?

https://www.reuters.com/world/us/musk-recommends-voting-republicans-us-midterm-elections-tweet-2022-11-07/

I think his behavior can be explained in one of two ways:
1. He is completely unaware how his actions are being perceived.
2. He saw how much eyeball-share Donald Trump added to Twitter from 2016-2020, and he is trying to recreate that experience on the platform by emulating Trump's antics in his own way.

Given how much attention we're giving him even in our little remote corner of the internet, I'm voting for option #2.

Is Twitter getting more users?

[NorCal]

Does anyone find it odd that Musk is fully endorsing Republicans after saying that Twitter needed to be politically neutral?

https://www.reuters.com/world/us/musk-recommends-voting-republicans-us-midterm-elections-tweet-2022-11-07/

I think his behavior can be explained in one of two ways:
1. He is completely unaware how his actions are being perceived.
2. He saw how much eyeball-share Donald Trump added to Twitter from 2016-2020, and he is trying to recreate that experience on the platform by emulating Trump's antics in his own way.

Given how much attention we're giving him even in our little remote corner of the internet, I'm voting for option #2.

Is Twitter getting more users?

They're losing users.  I saw one estimate by a tracking firm at around 1M users out of 400M IIRC.

But that doesn't mean they're not getting more engagement from other users.  Just think of how many are now religiously checking what Elon said next.  And maybe they'll just get stuck in a doomscroll while they're at it.  These are the metrics that matter in social media. 

I'm still fully believe that this is a loser plan.  But there is a logic to it if you follow how social media makes money.

[teen persuasion]

Does anyone find it odd that Musk is fully endorsing Republicans after saying that Twitter needed to be politically neutral?

https://www.reuters.com/world/us/musk-recommends-voting-republicans-us-midterm-elections-tweet-2022-11-07/

I think his behavior can be explained in one of two ways:
1. He is completely unaware how his actions are being perceived.
2. He saw how much eyeball-share Donald Trump added to Twitter from 2016-2020, and he is trying to recreate that experience on the platform by emulating Trump's antics in his own way.

Given how much attention we're giving him even in our little remote corner of the internet, I'm voting for option #2.

Is Twitter getting more users?

They're losing users.  I saw one estimate by a tracking firm at around 1M users out of 400M IIRC.

But that doesn't mean they're not getting more engagement from other users.  Just think of how many are now religiously checking what Elon said next.  And maybe they'll just get stuck in a doomscroll while they're at it.  These are the metrics that matter in social media. 

I'm still fully believe that this is a loser plan.  But there is a logic to it if you follow how social media makes money.

I noticed that over half of the tweets I saw in my (employer acct) feed today were labeled "liked" or retweeted by Elon Musk.  Very surprising.

Then I noticed that at least one of the accounts I follow had joined the Elon Musk impersonator club (renamed their account) - I was likely seeing multiple impersonator account likes and retweets, and can't distinguish them from Elon's real account.  Original tweets would show the name AND handle, but retweets and likes don't.

There's a lot of verified accounts having problems with impostors, too.  Multiple reports do nothing.

[ChpBstrd]

I'm not sure what was the point of buying Twitter for $44B only to turn it into just another 4-chan, 8-chan, Telegram, Parlor, or Truth Social.

THE thing separating Twitter from those apps was its popular appeal. Why wade into a much smaller and more crowded market, with its relative lack of advertisers? Why give up some of the network effects and first-mover advantages Twitter had by encouraging more mainstream users to bail to another platform? To earn right-wing users back from all those smaller services perhaps?

Is doing away with moderation a bet that the mainstream users and advertisers will have nowhere else to go, and so will tolerate the racial slurs, misinfo, sexism, calls for violence, troll farms, bots, and election interference?

In a sense, this strategic move is a bold bet that section 230 of the Communications Decency Act can never be repealed. When the entire social media ecosystem is owned by a handful of right-leaning billionaires and is used as an elections weapon, Democrats will eventually start asking "why are we enabling this with section 230"?

Musk is betting that won't happen, or Democrats will never be in power again, and section 230 will see no serious opposition anytime soon. He's also betting Twitter's network effects will prevent defections toward a moderated competitor (When the human-moderated Twitter model has already proven to be unprofitable, who is going to start it?).

[Paul der Krake]

Section 230 can never be repealed. It would instantly obliterate every tech company subject to US jurisdiction.

[NorCal]

Section 230 can never be repealed. It would instantly obliterate every tech company subject to US jurisdiction.

But it could be heavily amended.  There's lots of daylight between "tech bears no responsibility for anything on their platforms" and "tech is fully responsible for every rando's comments". 

[Paul der Krake]

Section 230 can never be repealed. It would instantly obliterate every tech company subject to US jurisdiction.

But it could be heavily amended.  There's lots of daylight between "tech bears no responsibility for anything on their platforms" and "tech is fully responsible for every rando's comments".

For sure! But realistically you can't go much further than "best effort" policies that rely on imperfect automation and human reports. There are enough available eyeballs in the world to do much better.

I think the general public vastly underestimates how much money and effort already goes into moderation and cleanup. It doesn't matter what your website does, if you leave any free-form field anywhere on it (or worse, file uploads), it will be abused.

Harassment, porn, bandwidth theft, shitcoin mining, snuff murders, you will get everything. The big boys already spend billions of dollars a year keeping those services from falling into absolute anarchy, and they still get their executives dragged in front of the Senate so that Elizabeth Warren can wag her finger at them. They do it because it's still better business than not doing it, but the cost (to say nothing of the human cost to the moderators who have to look at this shit) is absolutely staggering.

No other industry gets this much scrutiny over what users do with their products, and that's fine. If the leader of the KKK buys a Ford Focus, that's not great for Ford but nobody is calling for extra scrutiny on Ford dealerships' terms of service, because a car is not a megaphone. But look, people are really good at evading rules, so cut the corporations some slack. They can't possibly police everything and get it right every time.