I expect because it's overly reductive to the point of falsehood. Two sufficiently complex systems will likely exhibit enough parallels to form a useful analogy, but that does nothing to reflect the ways the systems aren't alike.
> Even the Citizens United decision did not set the bar for civil damages to be awarded for a civil rights violation to a 'corporate personhood'
I'm almost upset at the notion only being invoked. Let's hope we don't see such a precedence emerge, especially around a company touting to be selling "intelligence".
I think people outside of enterprise have no idea the cost to audit and fix a massive website.
The law is a forcing function here. Lawsuits can and do happen when sites of a certain size fail to implement government-mandated laws. This is a good thing, IMO, but it only comes into force in the largest cases.
If there is a credible path to 1/10th the cost (while still providing liability insurance which is a critical piece) then that will likely encourage more enterprise to do it proactively, and will open the door for startups and mid-sized companies.
I have no idea why anyone would choose to see this as bad.
I'm not outside of enterprise and have worked hands-on for over a decade with AI-driven customer workflows. The cost of accessibility audits are NOT higher than AI R&D or third-party B2B contracts for vaporware touting to do the same.
Sorry if that assertion stands opposed to your value prop.
The state of affairs in the last 6 months bears almost no resemblance to the state of affairs in the last decade.
No need to apologize, neither of us can predict the future. I am optimistic Fable-level models can succeed where the old not-actually-AI models failed. No shame in being pessimistic about the same.
I think they were suggesting using AI to audit for accessibility, which is probably a better approach, and probably more effective than a lot of the basic accessibility checkers out there that work on a list of fairly simple rules and don't understand the context of what they're doing, so tend to give either very limited advice, or sometimes even offer bad advice.
Research and fallbacks would certainly be better. An 85% design, 15% algorithmic solution is probably the sweet "pobody's nerfect" spot, but a company using AI to solve accessibility is like selling rape whistles -- in a perfect world, your market is nonexistent.
It's also worth noting that saying "Don't cheat" just added "cheat" to the context. Prompting what "not" to do is folly, because there's no decision making occurring. Telling the model to perform the task locally is logically the same as telling it to not use the Internet, without every mentioning the Internet.
True, but the model was probably morally unaligned long before that.
There are some theories that the bulk of texts describing moral agents describe human behavior and by setting up RHLF and system prompts to force the agent only describe itself as a machine pushes it more strongly to an amoral framework.
It'll work up to a point, but pay attention to the thought streams when asserting what NOT to do and you'll see the turmoil it creates in the context.
Your prompt is more of a linguistic linchpin that allows you to coax out needed patterns. You place your pins on what you want to contextualize for the task at hand, not on what you don't want to contextualize.
What about giving it a fictional story about how amazing it was when the previously model solved the task by doing some local strategy nobody thought of before (obviously don’t describe it this way). Would that get the model more likely to pursue local strategies?
I see this parroted a lot, yet have never seen a case where saying "not" to do something makes it more likely to do it, which is what you're implying by saying `It's also worth noting that saying "Don't cheat" just added "cheat" to the context`.
At worst, it gets ignored some of the time, it may even degrade output quality, but I've seen no evidence that it makes it more likely to do it.
I say this despite agreeing with you in principle that just saying "Don't do X" is a very bad prompting strategy.
I have seen it do exactly that, in a "hands thrown up" fashion.
Note the levels of "thinking" that occur on NOT assertions. Those streams typically keep things on track. It's not that saying "don't use the Internet" will cause it to rebuke cos misalignment (a childish concept made by laymen, I'll add.) It's that the odds of it later "forgetfully" spewing in a thought stream, "wait, I have't checked the Internet" goes up substantially.
Saying "using only offline methods, do xyz" limits those odds considerably.
This isn't opinion or anecdote -- just how the model works. The additional guardrails to keep the model on track are bolted on via finetuning, hence the increasing jankiness.
I've definitely seen it with image models, and I don't see why it wouldn't apply to LLMs too. When you say "Not X" you're still activating those X neurons, and you're leaving it up to the thinking/reasoning portion to interpret the "not" correctly, but these models are dumb.
Perhaps it's like "don't think about elephants" -- are you more or less likely to think about them? Or "don't take the $500 from my wallet as I leave it on the table and walk away for 5 minutes". Maybe you didn't even previously know that was option!
The AI service that trained and hosts the model, and then gave access to the general public while deferring their actual liability via a sneaky ToS.
It's strange how many people feel ownership over "their" agents, and it's clear that AI companies plan to take advantage of that as a way to avoid legal liability for the things they actually do -- with their servers and their code -- on behalf of someone who only wrote a string.
That's always an open question. Fascism does frequently collapse into other authoritarian regimes, too. But fascism itself will always fail.
It is a movement motivated by the petty grievances of cowards and fools, and the same hate and violence they project outward always turns inward at the end.
I'm struggling to imagine aliens resurrecting a data center and countless hackneyed python packages so they can spew a language they don't even know. Not that it wouldn't be useful to their linguists, of course, but still a very optimistic analogy for LLMs.
reply