1mo ago

Anthropic and Apollo astounded to find that a chatbot will lie to you if you tell it to lie to you

pivot-to-ai.com Anthropic and Apollo astounded to find that a chatbot will lie to you if you tell it to lie to you

Did you know that your chatbot might be out to deceive you? That it might be lying to you? And it might turn you into paperclips? Huge if true! Ordinary people have been using chatbots for a couple…

You're viewing a single thread.

34 comments

AI developers need to generate criti-hype — “criticism” that says the AI is way too cool and powerful and will take over the world, so you should give them more funding to control it.
This isn’t quite accurate. The criticism is that if new AI abilities run ahead of the ability to make the AI behave sensibly, we will reach an inflection point where the AI will be in charge of the humans, not vice versa, before we make sure that it won’t do horrifying things.
AI chat bots that do bizarre and pointless things, but are clearly capable of some kind of sophistication, are exactly the warning sign that as it gains new capabilities this is a danger we need to be aware of. Of course, that’s a separate question from the question of whether funding any particular organization will lead to any increase in safety, or whether asking a chatbot about some imaginary scenario has anything to do with any of this.
- With your choice of words you are anthropomorphizing LLMs. No valid reasoning can occur when starting from a false point of origin.
  Or to put it differently: to me this is similarly ridiculous as if you were arguing that bubble sort may somehow "gain new abilites" and do "horrifying things".
  
  I had assumed the golden age of people coming here to critihype LLMs was over because most people outside of Silicon Valley (including a lot of nontechnical people) have realized the technology’s garbage but nope! we’ve got a rush of posters trying the same shit that didn’t work a year ago, as if we’ve never seen critihype before. maybe bitcoin hitting $100,000 makes them think their new grift is gonna make it? maybe their favorite fuckheads entering office is making all their e/acc dreams come true? who can say.
  
  in crypto, these guys run on a six to eighteen month cycle - at get in, evangelise, get rekt and disappear in embarrassment. What this means is that the only people who actually remember the history of crypto are the critics.
  i once had a coiner demand in outrage that i prooove my claim that bitcoin was started by libertarians.
  anyway. dunno if the same will hold in AI grift, but yeah recycling refuted claims as if nothing happened is standard in other areas of pseudoscience.
  
  i once had a coiner demand in outrage that i prooove my claim that bitcoin was started by libertarians.
  ... did they know what a libertarian is?
  
  i once had a coiner demand in outrage that i prooove my claim that bitcoin was started by libertarians.
  who the fuck else would start it?
  
  a libertarian, a pedophile and an early crypto enthusiast walk into a bar
  
  "Drinking alone tonight?" the bartender asks.
  
  "no," he says, "miku-tan will have a banana juice."
- What new AI abilities, LLMs aren't pokemon.
  
  The AGI learned DECIEVE, but all i wanted it to learn is HUG.
  
  Ah yes, if there’s one lesson to be gained from the last few years, it is that AI technology never changes, and people never connect it to anything in the real world. If only I’d used a Pokémon metaphor, I would have realized that earlier.
  
  I mean, you could have answered by naming one fabled new ability LLM's suddenly 'gained' instead of being a smarmy tadpole, but you didn't.
  
  I wasn’t limiting it to LLMs specifically. I don’t think it is up for debate that as years go by, new “AI” stuff periodically starts existing that didn’t exist before. That’s still true even though people tend to overhype the capabilities of LLMs specifically and conflate LLMs with “AI” just because they are good at appearing more capabale than they are.
  If you wanted to limit to to LLM and get some specifics about capabilities that start to emerge as the model size grows and how, here’s a good intro: https://arxiv.org/abs/2206.04615
  
  lol, there has literally never been a gain of function claim that checked out
  you're posting like an evangelist, this way to the egress
  
  well yeah
- what if the AI sprouts wings and flies into the sky where we can't reach it?
  
  maybe that’s how the moon got mad - annoying goddamn chatbots flying in its view the whole time
- AI chat bots that do bizarre and pointless things, but are clearly capable of some kind of sophistication, are exactly the warning sign that as it gains new capabilities this is a danger we need to be aware of.
  hahahaha nope
  
  Here’s a video of an expert in the field saying it more coherently and at more length than I did:
  https://youtu.be/zkbPdEHEyEI
  You’re free to decide that you are right and we are wrong, but I feel like that’s more likely to be from the Dunning-Kruger effect than from your having achieved a deeper understanding of the issues than he has.
  
  who the fuck is “we”? you’re some asshole who bought the critihype so hard you think that when the chatbot does dumb computer shit that only proves it’s more human and more dangerous. you’re not in on this grift, you’re a mark.
  
  With all due respect, a counterpoint:
  https://www.youtube.com/watch?v=xvFZjo5PgG0
  
  convincing
  
  For anyone who rightfully can't be arsed to click that link, the expert is "Robert Miles AI Safety", who I assume is an expert (a youtuber) in the madeup field of "AI safety".
  Not to be confused with the late and great dream trance producer Robert Miles whom we all love dearly.
  
  I think he is a cs guy, who also is into EA, so basically it is the same source as the research OP posted about.
  
  Okay apparently it was my turn to subject myself to this nonsense and it's pretty obvious what the problem is. As far as citations go I'm gonna go ahead and fall back to "watching how a human toddler learns about the world" which is something I'm sure most AI researchers probably don't have experience with as it does usually involve interacting with a woman at some point.
  In the real examples that he provides, the system isn't "picking up the wrong goal" as an agent somehow. Instead it's seeing the wrong pattern. Learning "I get a pat on the head for getting to the bottom-right-est corner of the level" rather than "I get a pat on the head when I touch the coin." These are totally equivalent in the training data, so it's not surprising that it's going with the simpler option that doesn't require recognizing "coin" as anything relevant. This failure state is entirely within the realms of existing machine learning techniques and models because identifying patterns in large amounts of data is the kind of thing they're known to be very good at. But there isn't any kind of instrumental goal establishing happening here as much as the system is recognizing that it should reproduce games where it moves in certain ways.
  This is also a failure state that's common in humans learning about the world, so it's easy to see why people think we're on the right track. We had to teach my little on the difference between "Daddy doesn't like music" and "Daddy doesn't like having the Blaze and the Monster Machines theme song shout/sang at him when I'm trying to talk to Mama." The difference comes in the fact that even as a toddler there's enough metacognition and actual thought going on that you can help guide them in the right direction, rather than needing to feed them a whole mess of additional examples and rebuild the underlying pattern.
  And the extension of this kind of pattern misrecognition into sci-fi end of the world nonsense is still unwarranted anthropomorphism. Like, we're trying to use evidence that it's too dumb to learn the rules of a video game as evidence that it's going to start engaging in advanced metacognition and secrecy.
  
  “watching how a human toddler learns about the world”
  I have several family members with kids now and this is quite funny. A toddler learns how to crawl by looking at a lot of adults crawling or something.

34 comments