JASCO

On AI

Published: Sat, 23 Nov 2024 22:12:12 -0800

I have mixed feelings on Large Language Models

When ChatGPT first entered the zeitgeist in early 2023 I couldn't care any less. I'm skeptical of hot new things, especially when it comes to technology, and its reporting had all the red flags of a new tech fad. Eventually I gave it a try and tested it with attempting to replicate my very simple fetch script, (i.e. neofetch, fastfetch, etc.) jfetch. Most of it's attempts were pretty good, but would have a couple errors: using systemd utilities even after saying I'm on Void, not getting the colors right, randomly rewriting working code after asking for revisions. I though it was a neat utility, but nothing earth-shattering. With all the jokes about it's content filters going around I, naturally, tricked it into saying naughty things which was enjoyable.

Then over the course of the year, AI became the new buzzword for software projects. I got tired of it pretty fast, all the rebranding of software features as "AI enhanced", even when it seemed dubious if AI was really being used at all. Everything being labeled AI didn't bother me that much, but what really irked, maybe even worried, me was how it was being treated as it was in some way concious. It simply is not. It's an auto-predictor on amphetamines. Yet all the talking heads were treating it like an omniscient, terminator-esque being. This was in no small part due to Sam Altman's constant fear-marketing to the media of how he was scared about what hyper-intelligent sentient Skynet could possibly leak from their labs.

Then this semester started. My professors (mostly) no longer had a "Just don't use it" policy. One of my friend's classes even had a required section of their paper that had to talk about how they used AI for the assignment, even if they didn't use it. However even with the permission of the professors, I just don't find it that useful. Most education related topics have boundless websites with examples from universities or sites like GeeksForGeeks. These sites often have relevant links for connected concepts which you'd have to purposefully ask for while using an LLM.

I got the opportunity to be part of a focus group for shaping the policy of AI use in writing classes. A lot of the language describing the focus group talked about the hazards of "misinformation" related to AI, which made my eyes roll. Anyone who has used these things seriously knows how hard it is to make it say anything close to the sort of "misinformation" they were worried about: anti-establishment politics. This motivated me to be a sort of anti-alarmist voice for the discussion, though the actual conversation took a different turn.

The leader started off asking us how we used AI, things we liked about it, and things we didn't like. The table was pretty much unanimous in our conclusions: it was a useful assistant when we needed it to be, but unless you already know about a subject and know what questions to ask, the information it provides is very surface level. It was kinda relieving to hear. I thought I may have been one of a few dissenting voices on the actual abilities of AI, but everyone was mostly in agreement. When it got to the more writing specific questions, a few of the members had some good points about it being useful for a sort of peer-reviewer, not taking it's words verbatim, but using its suggestions to improve their essays. Overall it was a fun opportunity and I hope that our discussion can help form an effective policy.

After the discussion I've actually found myself using it more. Instead of looking up how to configure apache2 to log in local time, I asked Claude, and didn't have to click on annoying "accept cookies" buttons.

So far I've been using LLM and AI interchangeably, not mentioning the audio, image, and video generators. For the image generators, they're cool and useful. I think a lot of the artists who were outraged about their work being used to train these models are kinda telling on themselves that their work is so uninspired that it could be generated by a machine. If your work is inspired, you have nothing to worry about. It's also a told-you-so moment for posting things on Big Web platforms. As for the audio and video generators, these are also awesome and the memes are top-notch.

In summary, AI is kinda cool, useful in a number of scenarios, but not the major step forward in humanity that it's being made out to be. This is not Skynet and we're a far ways off from that.

If I'm proven wrong, I'll eat my shoe

Contact Me

I regularly check my email, If I don't respond quickly, send me a poke:
jasco.website@pm.me