How are you using ChatGPT/AI at work?

[GilesMM]

I have been tiptoeing into a work environment recently and was surprised to see how far ChatGPT was already making inroads.   Some of the smartest people I was working around were using it for programming, Excel formulii and scripting for Powerpoint slides.  Very slick.  It was a wake up call for me that stuff I usually do the hard (slow) way could be better, fatter and easier with ChatGPT.


How you are you all using it to work smarter?

[sixwings]

My teams use it for helping with excel scripting/macros/VBA, writing procedures/documentatation, some email drafting, etc. We have officially banned it for programming as we don't particularly want openAI to have access to our code in their databases. We have other software to help with that which is much more secure (but we are a large company, i suppose for a small start up or something that's trying to get to market quickly that may be less of a concern).

It's a fine tool to use for non-critical peripheral processes but not something to use for programming your core products.

[Raenia]

we don't particularly want openAI to have access to our code in their databases.

This is why I never expect to be able to use ChatGPT or similar tools at work - too much risk that proprietary information / internal processes end up in their database. That would never be allowed.

[Villanelle]

My spouse has to give a surprising number of speeches at work, and he has always struggled to write them.  He also writes a fair amount of letters-of-recommendation and things of that nature.  He's been using ChatGPT to come up with initial drafts.  he still spends a long time--far more than necessary, especially for the speeches, IMO--editing, but it does speed up the process for him.   

[Adventine]

I recently tried using ChatGPT to practice foreign languages. It works for practicing reading and writing (desktop version) and listening and speaking (on the app). It's an excellent way to supplement formal language learning, especially if you don't have native speakers to practice with.

[BiggerFishToFI]

Programmer here, but mostly use it for help with proposal writing and emails where I want to sound especially professional.

Wish I was as eloquent as a robot, but I'm not.

[By the River]

I prompted Chatgpt with the subject and a couple of audit findings that I had.  It quickly popped out an audit report.  Of course, not in our format or style but I did use some of the wording in my report.   

[GilesMM]

I prompted Chatgpt with the subject and a couple of audit findings that I had.  It quickly popped out an audit report.  Of course, not in our format or style but I did use some of the wording in my report.   

I just had it completely rewrite a letter I carefully constructed to a federal regulator. I asked for a "professional yet entertaining style".  OMG it is fabulous!

[LaineyAZ]

I asked ChatGPT a question about funding sources for a proposed local community project.  Answers were impressive and I'm definitely going to follow up.

[SmashYourSmartPhone]

How you are you all using it to work smarter?

I don't.  We've run private copies of some of the LLMs against our codebase, and while the results are hysterically funny, they're also entirely wrong, and our problems aren't the general sort you can find on Stack AIFlow.

Programmer here, but mostly use it for help with proposal writing and emails where I want to sound especially professional.

In other words, "long and verbose without saying anything of substance."  Don't use LLMs to inflate your word count.  Just say less.  Because odds are good that if you've generated a pile of fluff, someone else will use a LLM to summarize the important bits, which is what you would have written in the first place.

Except having used a lot of energy in the process pointlessly.

"Sound good, go for it!"

"ChatGPT, will you make this sound more professional?" =>

"Having evaluated the proposal in its entirety, we have come to the conclusion that the project is acceptable, and grant our permission to execute as proposed. (add words as needed)"

"ChatGPT, can you summarize this turd?" =>

"Sound good, go for it!"

[maizefolk]

I've largely replaced myself as the native english speaker who edits others writing to sound like that of a native english speaker with ChatGPT.

Have students in their first year of using python or R use ChatGPT as the first line service to explain code error messages or issues so they only come to me for the really interesting errors or problems.

First drafts of abstracts for papers after feeding in chunks of the results/discussion.

Am trying it out for letters of recommendation with mixed success but I may not have figured out how to feed in the right information for it to work with.

[GilesMM]

I've largely replaced myself as the native english speaker who edits others writing to sound like that of a native english speaker with ChatGPT.

Have students in their first year of using python or R use ChatGPT as the first line service to explain code error messages or issues so they only come to me for the really interesting errors or problems.

First drafts of abstracts for papers after feeding in chunks of the results/discussion.

Am trying it out for letters of recommendation with mixed success but I may not have figured out how to feed in the right information for it to work with.

That is the secret - knowing what specifically to feed the 'bot to get a good result.  It takes a lot of practice.

[Adventine]

I've largely replaced myself as the native english speaker who edits others writing to sound like that of a native english speaker with ChatGPT.

Have students in their first year of using python or R use ChatGPT as the first line service to explain code error messages or issues so they only come to me for the really interesting errors or problems.

First drafts of abstracts for papers after feeding in chunks of the results/discussion.

Am trying it out for letters of recommendation with mixed success but I may not have figured out how to feed in the right information for it to work with.

That is the secret - knowing what specifically to feed the 'bot to get a good result.  It takes a lot of practice.

Are you guys using the free ChatGPT version or the paid version which is supposedly more "intelligent" and can integrate plugins? 

[Chris Pascale]

While tutoring, I'll sometimes have a student punch a problem into the program. It's not ready for prime time.

[EvenSteven]

I just tried it with a simple question I couldn't remember off the top of my head:

User
does HindIII create a 5 prime overhang?

ChatGPT
No, HindIII does not create a 5' overhang. HindIII is a type II restriction enzyme, and it creates a 3' overhang or "sticky end." It specifically recognizes and cleaves DNA sequences with the following pattern:

5'-AAGCTT-3'

The cleavage site is between the two adenine (A) nucleotides, resulting in a 3' overhang with the sequence 5'-AA.

The "sticky end" created by HindIII is complementary to other DNA fragments that have been cut with HindIII, allowing them to easily anneal together when mixed. This feature is often used in molecular biology techniques like DNA cloning and gene manipulation.

HindIII does create 5' overhangs, not 3', so not super reliable yet.

[maizefolk]

Here's the output from GPT4 when I ask it the same question (I'm assuming you're using 3.5?)

User: does HindIII create a 5 prime overhang?
ChatGPT:

Yes, the restriction enzyme HindIII recognizes a specific six-base pair DNA sequence and cleaves it in a manner that leaves a 5 prime (5') overhang. The recognition sequence for HindIII is 5'-AAGCTT-3', and the enzyme cuts this sequence between the adenine (A) and the first cytosine (C) as follows:

5'-A|AGCTT-3'
3'-TTCGA|A-5'

The vertical line "|" indicates the site of the cut. The result is a 5' overhang (or sticky end) with the sequence 5'-AGCTT. Sticky ends are useful in molecular biology for joining DNA fragments together, as they can easily pair with complementary overhangs from other DNA molecules cut with the same enzyme.

[EvenSteven]

Here's the output from GPT4 when I ask it the same question (I'm assuming you're using 3.5?)

User: does HindIII create a 5 prime overhang?
ChatGPT:

Yes, the restriction enzyme HindIII recognizes a specific six-base pair DNA sequence and cleaves it in a manner that leaves a 5 prime (5') overhang. The recognition sequence for HindIII is 5'-AAGCTT-3', and the enzyme cuts this sequence between the adenine (A) and the first cytosine (C) as follows:

5'-A|AGCTT-3'
3'-TTCGA|A-5'

The vertical line "|" indicates the site of the cut. The result is a 5' overhang (or sticky end) with the sequence 5'-AGCTT. Sticky ends are useful in molecular biology for joining DNA fragments together, as they can easily pair with complementary overhangs from other DNA molecules cut with the same enzyme.

Yeah, using 3.5

[GilesMM]

I queried about the forum and got this:


While the Mr. Money Mustache forums are a valuable resource for those interested in personal finance, early retirement, and frugality, like any online community, they have their share of challenges and problems. Some of the common issues and concerns associated with the MMM forums include:
 
Quality of Advice: The forums are user-driven, which means that the quality of advice and information can vary widely. While there are many knowledgeable and experienced members, there can also be instances of misinformation or advice that may not be suitable for all individuals or situations. Users should exercise caution and do their own research before making financial decisions based solely on forum discussions.
 
Online Anonymity: Some users choose to remain anonymous or use pseudonyms, which can make it challenging to verify the credibility of the advice or expertise of the person providing it.
 
Conflicting Opinions: Like any online community, there can be disagreements and conflicting opinions among members. While healthy discussions and debates are encouraged, sometimes these disagreements can become less constructive and more contentious.
 
Lack of Diverse Perspectives: The MMM forums have a primarily North American and Western audience, which can result in a lack of diverse cultural and socioeconomic perspectives. This may not address the specific needs and challenges of people from different backgrounds or regions.
 
Risky Investment Strategies: Some discussions may promote aggressive investment strategies, such as heavy reliance on individual stocks or speculative investments, which may not be suitable for all investors. It's essential for users to assess their own risk tolerance and financial goals.
 
Simplistic Advice: The simplicity and directness of some MMM principles, while effective for many, may not be appropriate for all situations. Individual circumstances vary, and not all advice is universally applicable.
 
Overemphasis on Extreme Frugality: The MMM philosophy places a strong emphasis on frugality, which can sometimes lead to an unrealistic expectation of how quickly one can achieve financial independence. Some people may find the extreme frugality promoted on the forums challenging or unappealing.
 
Overemphasis on Early Retirement: The forums can sometimes create the impression that early retirement is the ultimate goal, which may not align with everyone's personal aspirations or circumstances.
 
Echo Chamber Effect: Like many online communities, the forums may foster an echo chamber effect where like-minded individuals reinforce each other's beliefs and may be resistant to alternative viewpoints.
 
Neglect of Emotional and Psychological Aspects: Achieving financial independence is not just a mathematical problem; it often involves complex emotional and psychological factors that may not be adequately addressed in forum discussions.

[Just Joe]

It would have been funnier if ChatGPT had called out specific forum participants who have especially strong opinions about various subjects.

[Tasse]

I'm a professional editor of academic manuscripts, and I tried out a large language model editor marketed for this specific purpose. In the single paragraph of edits it would let me preview for free, it made two changes that "sounded better" but changed the meaning of the text. It really drove home for me that LLMs do not understand text they work with, meaning they cannot edit (or write) to clearly convey a specific meaning. They write/edit stuff that "sounds right" based on their training dataset. Humans using that strategy is the reason there is a market for my job.

It could be useful for correcting simple and repetitive errors (e.g. "Fig 1" -> "Fig. 1"), but so is ctrl+F.

---

Neglect of Emotional and Psychological Aspects: Achieving financial independence is not just a mathematical problem; it often involves complex emotional and psychological factors that may not be adequately addressed in forum discussions.

On the contrary, I use this forum almost exclusively to discuss the emotional and psychological factors of pursuing FIRE.

[Joel]

One of my staff has had a lot of luck using chatgpt for power bi coding.

[scottish]

Here's the output from GPT4 when I ask it the same question (I'm assuming you're using 3.5?)

User: does HindIII create a 5 prime overhang?
ChatGPT:

Yes, the restriction enzyme HindIII recognizes a specific six-base pair DNA sequence and cleaves it in a manner that leaves a 5 prime (5') overhang. The recognition sequence for HindIII is 5'-AAGCTT-3', and the enzyme cuts this sequence between the adenine (A) and the first cytosine (C) as follows:

5'-A|AGCTT-3'
3'-TTCGA|A-5'

The vertical line "|" indicates the site of the cut. The result is a 5' overhang (or sticky end) with the sequence 5'-AGCTT. Sticky ends are useful in molecular biology for joining DNA fragments together, as they can easily pair with complementary overhangs from other DNA molecules cut with the same enzyme.

I also noticed GPT-4 is better at giving authoritative answers.     I compared GPT-3 to an enthusiastic script kiddie who will try anything he can find on the internet, while GPT-4 is more like a power user.

Would you rely on it's answers though?    Neither of them are good enough for my work.

[maizefolk]

In my experience GPT4 is also a LOT better at telling me it doesn't know things. Yesterday* after Just Joe's post I tried a couple of times to get it to comment on individual MMM forum users and it consistently said some version of "I'm sorry, I don't know enough to comment."

I am sure it still makes lots of factual mistakes. But so do the students and collaborators whose writing I often edit (and in fairness I probably make mistakes too, I just don't realize it). So ChatGPT4 is already at the point it is useful to me in helping to write or summarize text in areas I know a lot about. For topics I don't know enough to fact check it as it goes, fact checking it would be slower than just doing the work myself.

*When I try to do the same thing today, asking about individual forum members appears to trigger bing integration and does give answers referencing individual posts they've made.

[EvenSteven]

In my experience GPT4 is also a LOT better at telling me it doesn't know things. Yesterday* after Just Joe's post I tried a couple of times to get it to comment on individual MMM forum users and it consistently said some version of "I'm sorry, I don't know enough to comment."

I am sure it still makes lots of factual mistakes. But so do the students and collaborators whose writing I often edit (and in fairness I probably make mistakes too, I just don't realize it). So ChatGPT4 is already at the point it is useful to me in helping to write or summarize text in areas I know a lot about. For topics I don't know enough to fact check it as it goes, fact checking it would be slower than just doing the work myself.

*When I try to do the same thing today, asking about individual forum members appears to trigger bing integration and does give answers referencing individual posts they've made.

Uh-oh, how good is it at recognizing sarcasm?

[GilesMM]

Harvard says AI boosts productivity 40% when used appropriately...


https://venturebeat.com/ai/enterprise-workers-gain-40-percent-performance-boost-from-gpt-4-harvard-study-finds/