[cogman10]

Bingo. You see this in public shareholder calls all the time. Today, you can't find a tech company that doesn't say something like "We are integrating AI with LLM into our technology to blah". They don't say that because it's a meaningful or good product, they say that because investors give you a 10% stock bump just for saying "LLM".

The same thing happened with crypto, machine learning, web 2.0, etc. Investors love a good buzzword and will reward a company handsomely for using them.

Hell, that's like 90% of the reason FTX became as big as it was.

[rurp]

It was wild seeing companies have their biggest stock jump ever simply by talking about crypto magic beans on an earnings call.

[cogman10]

My company just did a shareholder meeting. Of everything we described (and we had a KILLER year) basically the only questions that were asked revolved around our usage of LLMs. It didn't matter that we made a ton of money, new products, and happy customers. What mattered was how we are adding AI.

[consumer451]

Well, please tell us, how are you adding AI?

I say this only one-quarter sarcastically, I would genuinely like to know. The only real applications of LLMs that I have seen so far are devs "learning" new languages.

[cogman10]

Ha.

Part of our software has a data ingestion and normalization process that can be somewhat costly. Basically, on a near per client basis we get data that needs to be transformed into our data model. We are currently investigating into using LLMs to setup our normalization DSL (or at least fill it out as best it can before a human signs off on it).

There's also an investigation going on to allow a client to ask an LLM about there data. I'm not sure how well that one will pan out.

[HideousKojima]

I'm feeling the "we need to adopt AI because it's the hot new thing/investors want to hear it" pressure at work nonstop, including some annoying mandatory trainings on it. My brother in Christ, I'm still trying to get my teammates to use git and package management, AI is going to have to wait a little bit.

[xmcqdpt2]

ChatGPT can (probably? I haven't checked) use git and package management. Maybe you should replace your colleagues with it.

[HideousKojima]

To be completely honest ChatGPT would probably generate fewer vulnerabilities per line of code than my colleagues. Sadly that's not my call, and at least I get paid well to clean up their messes and put out their fires.

[happytiger]

This comment is hideously honest and kind of darkly hilarious.

Now I can’t help but wonder if there is an introduced vulnerability per colleague metric we should be tracking?

I feel you on this one.