Teach me about this AI fad

[FIREin2018]

I know nothing about AI.
I feel like my parents 30 yrs ago when computers came out.

1a) How do I get started with ChatGpt?
1b) what's this about it buying more prompts?

2)  is chatGpt the best?
I see Whatsapp has AI  built into their app. How's that?

3)  I don't want it to create pics for me.
I want it to search all the weekly area supermarket ads and output the best deals.
Oh, also add best deals from weekly hardware stores (home Depot,Lowe's, harbor freight).
How do I do that?

[dividendman]

AI is not some new thing.

AI is generally useful to large companies that have access to a lot of data for it to learn off of. It's not really useful for an individual to do some task.

An easy example is google autocomplete, it can kind of guess what you're going to type next based on all the data. It's good for sharpening photos, finding better routes for uber, synthesizing possible drugs etc. But it's not useful for things like finding the best kickboxing instructor that charges less than $50/hr in this geolocation or some other kind of problem an individual is going to have.

All of that said, I don't use AI directly. Only indirect use through vendors like Google.

None of the "AI" that exists comes anywhere close to general artificial intelligence that could solve a myriad of general problems for you.

[scottish]

3)  I don't want it to create pics for me.
I want it to search all the weekly area supermarket ads and output the best deals.
Oh, also add best deals from weekly hardware stores (home Depot,Lowe's, harbor freight).
How do I do that?

That's too bad, it's good at creating pics!    But none of the chat bots/large language models will be up to date on weekly hardware store ads.   Maybe they could give you deals from 2023 though... ;-)

It's mostly hype.    AI is proving good at natural languages and image processing in particular.    The folks at deepmind have also made some really impressive systems that play games like chess and go.    Alot of companies are trying to flog it for applications such as writing source code, but you can't trust the results if you need accuracy.   YMMV.

[twinstudy]

I don't trust AI to replace most high-level jobs. It's good at generating positive content (i.e., synthesising facts and information based on others' writing) but it is not good at generating negative content (successfully avoiding mistakes, and correctly advising others on how to avoid mistakes).

I work in law and I'd love to see AI successfully counsel a taxpayer on how far she can go in bending a particular section of the tax legislation without falling foul of its anti-avoidance provisions. I don't think AI can do that at all. I don't think it could even reliably synthesise and apply cases dealing with the very point: it could locate and quote from them, but I doubt it could apply principles to facts.

Nor will AI have a good grasp of basic but unspoken concepts. For example, diffusion of responsibility. It is much easier to claim that an omission is a mistake than to excuse oneself for a positive action which is fraud.

Till AI understands those concepts, it will be of no use in litigation, other than to do baby tasks like research, document summaries and discovery, which are things you can get an intern to do for cheap anyway.

I'm sure every field has its own things that AI will struggle with.

[FINate]

AI is a massive field made up of multiple techniques developed over the past 70ish years. The Wikipedia article on AI is a decent starting point: https://en.wikipedia.org/wiki/Artificial_intelligence

You're almost certainly already using AI and don't realize it. If you're searching on Google, getting driving directions via an app, using a voice assistant, or a bunch of other technology then you're using different forms of AI.

Generative AI, aka Generative Pre-trained Transformers (e.g. ChatGPT) has had a moment since around 2020. I suspect this is what you're really asking about. While I do think we're close to Peak AI Hype, it has actually become quite useful for while-collar workers for its ability to do a lot of the grunt work of gathering, collating, and formatting information. This includes fields like Software Engineering where GPT can now be used to generate the first iterations of code that are then refined by hand.

IMO, the best way to learn about generative AI is to learn by doing. Setup a ChatGTP account and start asking it questions about AI and how it works, how it's being used, and so on.

[daverobev]

AI is a massive field made up of multiple techniques developed over the past 70ish years. The Wikipedia article on AI is a decent starting point: https://en.wikipedia.org/wiki/Artificial_intelligence

You're almost certainly already using AI and don't realize it. If you're searching on Google, getting driving directions via an app, using a voice assistant, or a bunch of other technology then you're using different forms of AI.

Generative AI, aka Generative Pre-trained Transformers (e.g. ChatGPT) has had a moment since around 2020. I suspect this is what you're really asking about. While I do think we're close to Peak AI Hype, it has actually become quite useful for while-collar workers for its ability to do a lot of the grunt work of gathering, collating, and formatting information. This includes fields like Software Engineering where GPT can now be used to generate the first iterations of code that are then refined by hand.

IMO, the best way to learn about generative AI is to learn by doing. Setup a ChatGTP account and start asking it questions about AI and how it works, how it's being used, and so on.

Yeah so OP just needs to know this predictive/generative AI is "stupid". It regularly outputs nonsense because it was trained on 'the internet' which includes wrong answers that then get corrected (so it gives a bad answer, but if you query it it then gives an updated answer... just like the forums it has been trained on).

Code is a somewhat good use because it can give a framework that a good coder can fix up much faster than starting from scratch.

At the current state of play, it's - IMHO - only really valuable precisely because of enshittification. Google has monetised 'search' so that you don't get what you're looking for until after you've seen a load of adverts or otherwise Google-preferred links. ChatGPT or whatever can cut through this... but at a very high electricity cost. The whole thing is currently pulling so much electricity to do what a decent search engine could mostly provide... at a fraction of the cost and energy consumption.

But yeah, general purpose AI this is not. The worst thing is probably that it's used to make shitty bots that plague forums posting divisive shit (see: Twitter). I love the idea that you get better results on searching for things if you restrict it to 10+ years ago. Sigh.

[FINate]

AI is a massive field made up of multiple techniques developed over the past 70ish years. The Wikipedia article on AI is a decent starting point: https://en.wikipedia.org/wiki/Artificial_intelligence

You're almost certainly already using AI and don't realize it. If you're searching on Google, getting driving directions via an app, using a voice assistant, or a bunch of other technology then you're using different forms of AI.

Generative AI, aka Generative Pre-trained Transformers (e.g. ChatGPT) has had a moment since around 2020. I suspect this is what you're really asking about. While I do think we're close to Peak AI Hype, it has actually become quite useful for while-collar workers for its ability to do a lot of the grunt work of gathering, collating, and formatting information. This includes fields like Software Engineering where GPT can now be used to generate the first iterations of code that are then refined by hand.

IMO, the best way to learn about generative AI is to learn by doing. Setup a ChatGTP account and start asking it questions about AI and how it works, how it's being used, and so on.

Yeah so OP just needs to know this predictive/generative AI is "stupid". It regularly outputs nonsense because it was trained on 'the internet' which includes wrong answers that then get corrected (so it gives a bad answer, but if you query it it then gives an updated answer... just like the forums it has been trained on).

Code is a somewhat good use because it can give a framework that a good coder can fix up much faster than starting from scratch.

At the current state of play, it's - IMHO - only really valuable precisely because of enshittification. Google has monetised 'search' so that you don't get what you're looking for until after you've seen a load of adverts or otherwise Google-preferred links. ChatGPT or whatever can cut through this... but at a very high electricity cost. The whole thing is currently pulling so much electricity to do what a decent search engine could mostly provide... at a fraction of the cost and energy consumption.

But yeah, general purpose AI this is not. The worst thing is probably that it's used to make shitty bots that plague forums posting divisive shit (see: Twitter). I love the idea that you get better results on searching for things if you restrict it to 10+ years ago. Sigh.

Indeed, it is stupid, as in it isn't the same as general human intelligence. And it makes mistakes, including hallucinating (i.e. making shit up) which is troubling because no one really understands why it happens. But this doesn't mean it's not extremely useful and substantively different than pre-enshittified Google.

I was playing around with ChatGPT yesterday asking it things like "Show me a graph of inflation adjusted housing prices since 1980. Cite your sources." And in less than 60 seconds it would explain what it is doing, generate a graph, and cite the source for the data used in the graph. I then refined it by asking for break down by region, comparing California vs. the rest of the US, relative to income, and so on. Finding graphs like this on Google was always difficult to impossible -- the hope was always that someone happened to create the exact thing you're looking for, otherwise it's a long process of importing source data into a spreadsheet, cleaning it up, then creating the graph.

To be clear, if I were giving a talk or writing a paper I wouldn't throw up ChatGPT generated charts, that would be stupid. Instead, it's a way to quickly explore ideas and data. Then once you have a better idea of what you're after, you get the source data and do it for real. That's a real benefit over and above what Google offers.

[GilesMM]

Have you tried the Flipp app for grocery deals?

[FINate]

Here's another example of how generative AI is being used, which highlights why I think Google is very very worried.

My truck needs new tires soon. I don't drive it much other than to get up into the mountains on questionable roads, big snow days for skiing, and the like. I have proper winter tires for ski season, but there are times in the shoulder seasons when I've needed reasonable winter performance. So I'm looking for a capable all terrain tire that can handle sharp rocks, decent mud performance for short stretches, and performs reasonably well on snow and ice.

I kinda have an idea already of the top contenders, but this makes for a good test of ChatGPT. So I tried asking it "what are the top all-terrain tires for light trucks for snow and ice" and it came back with a very reasonable list of 5 tires, mostly what I'd expect. But then I asked it to refine for tires with 3PMSF rating which eliminated one. Then I asked it again to refine for a specific size and load rating, and it eliminated one more.

Getting to a list of 3 top contenders for very specific criteria in just a couple minutes is very good. There's one new tire for 2024 not on the list that I think could be, so their LLM may not be super fresh. And I wouldn't rush out to buy one without further research. But going through a similar process with Google alone is super frustrating and time consuming. The video results are the absolute worst, just a bunch of filler and self promotion for 60 seconds of actual content. And  the listicles and influencer content larded up with ads and what not is maddening.

So using ChatGPT as a first-order search to help define and narrow what you're looking for is extremely useful. This is also the point in the search experience that is most valuable to advertisers, so Google is right to be worried. Again, I wouldn't just trust its output as the gospel truth, since it's outputs are only as good as the inputs. But Google is much more useful once you have very specific tires to research.

[GilesMM]

Michelin CrossClimate are amazing all season tires.

[FINate]

Michelin CrossClimate are amazing all season tires.

Agreed, but not the right tire for my application.

[vand]

AI is tremendously helpful professionally if you are tech literate but actual coding isn't a core part of your skillset. I make heavy use of it to write scripts in any number of languages for both my job and non-work projects.

It's also great for compiling datasets without having to trawl the entire web.

As ever, the people who can get the most from AI are those who can ask good questions.  The better and more precise the questions and prompting/course correcting, the better the results.

What it can't do is- make you care about your job, chase people when something needs chasing up, or otherwise teach you the soft skills that make you truly awesome at your job.  It's a tool at the end of the day - like having access to your own personal savaunt in whatever area you need. People who choose to use it will be advantaged over those who don't... The good news is that it's easier and more intuitive than most imagine.  If you're asking questions on an internet forum then you already have the skills needed to start leveraging AI.

An example - I recently used chatGPT to help me write my entire investing/FI dashboard in Googlesheets. Help with custom functions for web-scraping, custom tax calculation functions, as well as numerous other bit with formulae, formatting etc.  I could probably have eventually put it all together pre-AI, but having chatGPT on hand to ask how to do this and that just made it so much easier.

[vand]

As for OP Q3, I asked chatGPT the following:

Can you summarize the current best special offers and discounts on groceries across Tesco and Sainsbury's in the UK?

Then

How can I create a weekly reminder that refreshes this list each time?


This was already enough to give me something workable to set up and use going forward. You can refine it with further clarifications and prompting, but you get the gist... 

[Ron Scott]

I wouldn’t judge AI from the current crop of chatbots like ChatGPT, Claude, CoPilot, Perplexity, etc. Other uses seem more advanced.

We are starting to see some interesting results in self-driving AI tools for cars. Pfizer’s COVID vaccine was largely made possible by MRNA AI and scientists are working to make the most of AI in drug development. DeepMind’s protein folding achievements were created by AI. Robots can navigate varying terrain because of AI and execute against some goals using their own chosen tactics. The first human patient implanted with a brain-chip from Neuralink AI is able to control a computer mouse using his thoughts. And AI is starting to make decent videos based on a text prompt.

It seems to me we are in the early infancy stages of AI development. Most of us will live to see some amazing developments, but I’m not convinced the real deal is just around the corner.