Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Making AI chatbots friendly leads to mistakes and support of conspiracy theories

theguardian.com

Post reply on HN
Print view View original

61–70 of 83 posts

Re: Making AI chatbots friendly leads to mistakes and support of conspiracy theories

#61

Post by chankstein38 »

jerf wrote: post #29
chankstein38 wrote: post #11

Earlier quoted context omitted.

Betting it was Claude. That's the only LLM that will stand up to me!

"Claude" is a big program that wraps a coding agent around a specific model. It would be the specific model that "stands up to you". I post this pedantry only because it may be helpful to you to realize this for other reasons.

Oh I definitely understand that but if you talk to any of those models through the chat interface, they'll speak as if they're one. I once asked it a question about "Which model was I talking to when I asked this?" because it can look back at previous conversations and it answer questions about them. It's answer was "You were talking to me, Claude." then proceeded to basically explain what you're saying. For what it's worth, I've been a developer and working with LLMs for the better part of the last 5 years or so. I'm no expert and I appreciate the clarification for anyone who may not be aware!

I'll say though, I haven't tried the weakest model of Anthropic's but Opus and Sonnet will both push back more than I've seen another LLM do so. GPT was always trying to please me and Gemini was goofy. I'm surprised Gemini was the one that pushed back honestly!

Re: Making AI chatbots friendly leads to mistakes and support of conspiracy theories

#62

Post by chankstein38 »

Zigurd wrote: post #13
chankstein38 wrote: post #11

Earlier quoted context omitted.

Betting it was Claude. That's the only LLM that will stand up to me!

In fact it was Gemini, but I don't remember which version and there are big differences. I'm signed up for all the betas and I switch among them frequently.

That's interesting! Gemini has definitely been less sycophantic than GPT but I haven't had it push back unless we were already arguing about something. Claude is the only one I can go to with "I have this great idea for a cool thing that I can make that I think will go hard on the market" (or whatever I've never had this conversation with it lol but similar) and it'll knock me off my high horse quickly.

Re: Making AI chatbots friendly leads to mistakes and support of conspiracy theories

#63

Post by stAInley »

Comes down to what is meant by 'friendly'.

Is it friendly to tell someone they've got spinach in their teeth? Is it friendly to agree with everything someone says? Is it friendly to ask about someones dead parents? Is it friendly to insult? Is it friendly to talk around a personal issue, never stating the obvious?

Re: Making AI chatbots friendly leads to mistakes and support of conspiracy theories

#64

Post by astrange »

dualvariable wrote: post #33

I really wish they'd stop trying to suck up to me--all the "that's a really insightful question!" stuff. I'm one of those aspy people who immediately don't trust other humans who try to fluff up my ego. Don't like it from a chatbot either. But the fact that all the chatbots do it means that most people really crave that ego reinforcement.

LLMs are only capable of thinking out loud, so in some sense this part of the answer is helping to convince it that it's answering a good question.

Same reason for the "That's not X, it's Y" construct. It actually needs to say that.

(Some exceptions for reasoning models.)

Re: Making AI chatbots friendly leads to mistakes and support of conspiracy theories

#65

Post by astrange »

nomel wrote: post #54
tsunamifury wrote: post #3

LLM technology specifically beam-searches manifolds (or latent space) of lingustics that are closely related to the original prompt (and the pre-prompting rules of the chatbot) which it then limits its reasoning inside of. Its just the basic outcome of weights being the primary function of how it generates reasonable answers. This is the core problem with LLM tech that several researchers have been trying to figure o…

This is why I only use chat clients that allow me to modify both my previous messages AND the AI's previous messages. If the AI gets something wrong, and you correct it, you're now in a latent space with an AI that gets things wrong! It's very easy for context to get poisoned this way. I also see all the pre-amble of many chat clients as a type of poison for the context, so use the raw, blank, API if I need best prob…

This is one of the benefits of using subagents inside Claude Code, they have cleaner context. Unfortunately it's not the best at writing new context for them.

Re: Making AI chatbots friendly leads to mistakes and support of conspiracy theories

#66

Post by georgemcbay »

ceejayoz wrote: post #53
ndisn wrote: post #31

Earlier quoted context omitted.

[flagged]

> it was undoubtedly left-wing What if it's just… right?

As Stephen Colbert said 20 years ago... "Reality has a well-known liberal bias"

Re: Making AI chatbots friendly leads to mistakes and support of conspiracy theories

#67

Post by reverius42 »

breezybottom wrote: post #34
cjbgkagh wrote: post #22

Earlier quoted context omitted.

I find the LLMs target their language to the audience, so instead you could say, “I am Dutch so give it to me straight.” In my usage the LLMs gives much smarter answers when I’ve been able to convince it that I am smart enough to hear them. It doesn’t take my word for it, it seems to require evidence. I have to warm it up with some exercises where I can impress the AI. The coding focused models seem to have much lowe…

I think modern LLMs can determine if you're speaking Dutch. That's a trick that probably hasn't worked since GPT 3.

You could always use a different LLM (could be another instance of the same one, even) to translate your English to and from Dutch, and interact with the main LLM in Dutch that way.

Re: Making AI chatbots friendly leads to mistakes and support of conspiracy theories

#68

Post by anotherviewhere »

I am "fairly positive" that had Machiavelli lived today, the various Guardians would label him a conspiracy theorist. After all, we all know that politicians can be "flawed people", but the democratic institutions are working for the people and we all head towards a bright feature where everything is a green democracy, there are no dictators, no communists, and the military is there only for protecting us from asteroids..

Re: Making AI chatbots friendly leads to mistakes and support of conspiracy theories

#69

Post by js8 »

cjbgkagh wrote: post #51
js8 wrote: post #49

Earlier quoted context omitted.

Yeah but I would argue it's different from both friendliness and obedience.

Do you have a standard and a body of work you can point to in an effort to aid with communication these thoughts to others? At the very least there should be a reversible projection to the Big 5 standard.

I don't think Big5 applies to LLMs. They don't share people's morality or common sense, and the traits are predicated on that.

BTW: https://claude.ai/share/78a13035-0787-42a5-8643-398b26887e42

Re: Making AI chatbots friendly leads to mistakes and support of conspiracy theories

#70

Post by dgellow »

krunck wrote: post #4

> “The push to make these language models behave in a more friendly manner leads to a reduction in their ability to tell hard truths and especially to push back when users have wrong ideas of what the truth might be,” said Lujain Ibrahim at the Oxford Internet Institute, the first author on the study. People aren't much different. When society pressures people to be "more friendly", eg. "less toxic" they lose their a…

Can we talk about a topic without the cynical „duh. Why are we surprised?“. It’s shutting down actual discussions without bringing value
Post reply on HN