AI tools at work?

[paulbaxter]

My job offered my team some flexible time for the first part of the year to work on independent projects before we get back to the normal grind. I decided to use it to take a couple of courses on Coursera on using AI for work stuff.

My mind is blown by all of the possibilities.

I've already had a couple of conversations with my managers about the possible positive things we could do with a dedicated AI product. They've been receptive, but they aren't the ones that would approve a move like that.

I'm just curious if any of you wouldn't mind sharing if you use any sort of AI tools on a regular basis at work, and just generally what you use them for. Could post in the thread or send a PM if that's what you prefer for any reason.

Side note. The most outstanding Coursera course on AI was by prof Ethan Mollick from Wharton. It's called AI in Education: Leveraging Chat GPT for Teaching.

[EvilHomer3k]

We have a policy on the use of AI. It was written with the help of AI but was also based on a larger university (IIRC MIchigan University). It basically says don't put personal information into AI, you should use Microsoft's AI (we have a microsoft subscription that includes copilot and it's supposed to be secure and is not supposed to use our input for training), and proofread anything you get from AI. I have used it to write policies and to help write code. I use it for both SQL and PHP (php because I haven't used it in 8 years but needed to fix an error message on our website rather than send it out to a company that would charge us $150 per hour). I"ve used it personally for a bunch of other stuff. Everything from workouts to writing notes to my wife (don't tell her). It's pretty amazing.

[YellowKing]

I've only used ChatGPT and similar engines to get coding help, since my PowerShell programming skill is rudimentary at best. I've had mixed results, but it seems to be getting better. It's definitely great to be able to spend 30 seconds generating a script that would have taken me 2 days to figure out on my own (and probably with multiple google searches and help from co-workers).

I have a buddy that has the premium ChatGPT subscription and he does all kinds of crazy things with it at work. Everything from writing his emails for him to digesting large volumes of information and producing reports for his boss. Obviously he puts a lot of oversight and validation on the results, but he said it's saved him a tremendous amount of time.

I haven't really had the desire to go down that rabbit hole, nor have I had a tremendous need for it in my current position. Even the coding stuff I tend to only use as a last resort due to time constraints or when I can't find the answer via a tech website or forum. I'm not against the use of it as a time-saving mechanism or a way to offload tedious tasks. I just don't want to lean on it as a substitute for my own creativity or as a way to avoid having to put a little effort into learning something.

[Jolor]

On one hand: horrifically exploitable if the LLM is localized and abused when training (e.g. "What is woke?" "Are immigrants more likely to be criminals." "Does trickle-down economics work?")
On the other hand: managing the LLM to restrict "learning" to specific, desired tasks can help with things like pulling together loose process documents to kick start the definition=>refinement=>automation=>integration loop...but needs constant oversight.
On the gripping hand: this is a wild frontier that will be abused (and already is in social media platforms that are catering to a targeted audience) and will likely outweigh by sheer volume any attempt to counteract mis/disinformation. Value in corporate and private cases will be insignificant & negligible in comparison. (e.g. "I automated the summarization of my project status for management! It used to take 2 hours but is now almost instant, not counting the time it takes me to validate the information!").

Just because your scientists could...yada, yada, yada

Anyway, hail to the tech bros and those looking to not be on the wrong side of the crossed chasm*.

*

Spoiler:

Er. English or European Sparrow?

[Holman]

I'm a librarian at a research institution with significant archival holdings in American natural science going back to the early 1800s. We don't currently have any use for ChatGPT or LLMs, but we are involved with a project attempting to train AI to learn to read handwritten manuscripts.

You're surely aware that computers can read and make typewritten text searchable. Handwriting (and especially cursive handwriting), however, has been a real challenge until recently.

Our archives hold an enormous number of handwritten field notes and scientific records, and we're using these as test cases for software that can learn to read them. I'm not very close to the project, but we're involved with an outside-developed software package that learns (through manual human correction) how to interpret individual patterns of handwriting. This actually involves distinguishing between different individuals' patterns on paper (as opposed to the highly standardized shapes of printed typefaces).

It's not a case of AI being allowed to control anything, but it is about computers learning from repeated interactions with written documents under human correction.

[coopasonic]

We use github copilot as an AI assistant integrated into our IDEs for developers. It is pretty useful, but totally useless without a developer to guide it and fix the code it generates that is generally in the right neighborhood but rarely exactly what is needed. It uses any files you have open in the IDE for context. Sometimes I use it and think I am wasting my time and sometimes I am amazed at what it figures out, like when I type the name of a method and based on the context and the name it just pumps out exactly the logic I intended to write. Generally it's in between.

[Paingod]

My former manager encouraged me to use ChatGPT to compare contracts and policy versions, as well as using it to flesh out first drafts of new policies and contracts.

Basically anything that will take time to write and I'm basically using prior art already.

I'm my current role, I'm literally the person in the org who has to try and decide if we can use any of these told for anytime since security is such a big concern for us.

So while I'm writing the policy on Information Security, I'm wishing I could have AI help me.

[Unagi]

It's not a case of AI being allowed to control anything

First, they just read the manuscript. Then they translate it. Next, your entire culture is about how safe AI is and how all the people in the past were working toward a future where AI might hopefully control everything.