User: Sheldon

Sheldon

2 hours, 18 minutes ago (edited 1 hour, 44 minutes ago)

Link

Imo, calling LLM-tooling “AI” is the best marketing strategy these companies had, and also terrible for how we have conversations about these tools. When you frame them as “AI” and give them names...

Imo, calling LLM-tooling “AI” is the best marketing strategy these companies had, and also terrible for how we have conversations about these tools.

When you frame them as “AI” and give them names and put them behind a messaging-like interface, it’s easy to anthropomorphize these tools. But they aren’t doing “intelligence”, they’re doing advanced statistics.

The tools are not “going rogue” and these companies are not “losing control”. Those stories about LLMs “hacking” other companies are a result of having no guardrails, infinite compute, no sandboxing, and unclear goals / instructions. If you prompt an LLM with an objective of retrieving a file from a server and give it unlimited resources and a shitty sandbox, of course one of the methods it might generate could be, essentially, hacking. And if you blindly trust the output of LLMs without supervision, you’ll end up doing hacking.

If these companies or their employees were serious about whatever fears they claim to have, they would be ceasing all operations and doing everything in their power to make everyone else cease operations. Since they aren’t, the only explanations are that they either don’t care that “AI is going to kill us all” OR that they don’t believe it and are saying that as a fear-based marketing strategy. Both, morally reprehensible, imo.

So…no, I am not particularly scared of the impact these technologies are going to have.

I am worried about the commitments by large tech companies to this technology. So far, it seems many companies are taking on lots of debt to support the buildout of the LLM companies despite, from what we know, the LLM companies are deeply unprofitable. Which means prices will need to rise or the tools will need to go away. Additionally, a lot of companies are seemingly very willing to hand over process knowledge to these LLM companies and get rid of or de-skill their own employees. Which again…seems shortsighted if those LLM providers disappear or need to raise prices.

I am also worried about the language we use to describe these tools. We don’t really have the words to NOT anthropomorphize them, and I think that is a huge driving factor in why so many folks get very…irrational about these products.

Edit: Also worried about the way these tools and their users might affect the social contract we all have. Many people seem very okay with saying that material output by these LLM tools is something “they made”. I generally feel the same way about this as I do with plagiarism. Which is not very positive. If someone doesn’t care enough to make something, I don’t care enough to engage with it. And if that person lies to try and trick me, that’s pretty awful.

6 votes

Sheldon's recent activity

Comment on Where do we go from here? (Regarding AI) in ~tech

Comment on What have you been listening to this week? in ~music