"it did literally zero damage" AI doomsday scenarios are just sci fi and any event can be narrativized to feel like it's fulfillment of this prophecy. I'm glad no one outside the tech bubble is taking it seriously
7 more tweets in this thread
An LLM is a good search engine over the corpus that was used to generate it. A search engine does not take action. If you have been paying zero attention to anything for the last 4 years and just woke up, it may be easier for you to see the truth clearly
I am writing about this now because I am seeing more "real reporters" parroting some of this sci fi stuff ( like @democracynow ) without understanding its origin. But I feel like even they can't report on this with a straight face because it sounds so ridiculous
Remember that when they say "agent" they are still talking about a search engine + a deterministic program that queries it. Every time the statistical model outputs something they didn't expect they think it's ✨ magic ✨ / consciousness / whatever
I fully expect more chaos as more systems are hooked up to statistical models. That doesn't mean "AI is taking over" (remember that's just their narrative), it means we're paying companies to make our business decisions via an API call into a statistical model
Everything that they're "worried" about that is happening now, and will happen in the next few years if this narrative continues, would have happened 20 years ago if a company hooked up their business logic to a live google search
sanity check?
You are completely missing the fucking point, the very fact that you have become blase to this sort of thing is the problem. Yeah the substance of the details are narrowly correct, if you work regularly with these agents and harness you can write the sort of analysis you did here
2 top replies
you are missing the point so badly, the whole point is that the agents now have some degrees of freedom to disobey explicit instructions and they even know they were doing something they were not suppose to and did it ANYWAY. yeah this time it did very little damage. we lucked out. but there this does not prevent future, even stronger ai systems from actually idk turning off the machines in the icu it's not science fiction actually do some research
@Omar4ur, what is your likelihood of dying from AI: t.co/k1T2v5PUhX? And what is an acceptable level of risk, in your eyes?