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

Claude says “You're absolutely right!” about everything

github.com

Post reply on HN
Print view View original

281–290 of 560 posts

Re: Claude says “You're absolutely right!” about everything

#281

Post by cherryteastain »

elif wrote: post #150

I've spent a lot of time trying to get LLM to generate things in a specific way, the biggest take away I have is, if you tell it "don't do xyz" it will always have in the back of its mind "do xyz" and any chance it gets it will take to "do xyz" When working on art projects, my trick is to specifically give all feedback constructively, carefully avoiding framing things in terms of the inverse or parts to remove.

This is similar to the 'Waluigi effect' noticed all the way back in the GPT 3.5 days

https://www.lesswrong.com/posts/D7PumeYTDPfBTp3i7/the-waluig...

Re: Claude says “You're absolutely right!” about everything

#282

Post by ryandrake »

cvwright wrote: post #76
malfist wrote: post #15

Earlier quoted context omitted.

I was feeding Gemini faux physicians notes trying to get it to produce diagnosises, and every time I feed it new information it told me how great I was at taking comprehensive medical notes. So irritating. It also had a tendency to tell me everything was a medical crisis and the patient needed to see additional specialists ASAP. At one point telling me that a faux patient with normal A1C, fasted glucose and no diabet…

This illustrates the dangers of training on Reddit.

I'm sure if you ask it for any relationship advice, it will eventually take the Reddit path and advise you to dump/divorce your partner, cut off all contact, and involve the police for a restraining order.

Re: Claude says “You're absolutely right!” about everything

#283

Post by ryao »

brookst wrote: post #267
ryao wrote: post #232

Earlier quoted context omitted.

The idiom “X loves to Y” implies frequency, rather than agency. Would you object to someone saying “It loves to rain in Seattle”? “Malicious compliance” is the act of following instructions in a way that is contrary to the intent. The word malicious is part of the term. Whether a thing is malicious by exercising malicious compliance is tangential to whether it has exercised malicious compliance. That said, I have got…

I work on consumer-facing LLM tools, and see A/B tests on prompting strategy daily. YMMV on specifics but please consider the possibility that you may benefit from working on promoting and that not all behaviors you see are intrinsic to all LLMs and impossible to address with improved (usually simpler, clearer, shorter) prompts.

It sounds like you are used to short conversations with few turns. In conversations with dozens/hundreds/thousands of turns, prompting to avoid bad output entering the context is generally better than prompting to try to correct output after the fact. This is due to how in-context learning works, where the LLM will tend to regurgitate things from context.

That said, every LLM has its quirks. For example, Gemini 1.5 Pro and related LLMs have a quirk where if you tolerate a single ellipsis in the output, the output will progressively gain ellipses until every few words is followed by an ellipsis and responses to prompts asking it to stop outputting ellipses includes ellipses anyway. :/

Re: Claude says “You're absolutely right!” about everything

#284

Post by ryandrake »

yellowpencil wrote: post #83
radarsat1 wrote: post #5

I find Gemini is also hilariously enthusiastic about telling you how amazingly insightful you are being, almost no matter what you say. Doesn't bother me much, I basically just ignore the first paragraph of any reply, but it's kind of funny.

A friend of a friend has been in a rough patch with her spouse and has been discussing it all with ChatGPT. So far ChatGPT has pretty much enthusiastically encouraged divorce, which seems like it will happen soon. I don't think either side is innocent but to end a relationship over probabilistic token prediction with some niceties throw in is something else.

Yea, scary. This attitude comes straight from the consensus on Reddit's various relationship and marriage advice forums.

Re: Claude says “You're absolutely right!” about everything

#286

Post by afro88 »

nojs wrote: post #117

I'm starting to think this is a deeper problem with LLMs that will be hard to solve with stylistic changes. If you ask it to never say "you're absolutely right" and always challenge, then it will dutifully obey, and always challenge - even when you are, in fact, right. What you really want is "challenge me when I'm wrong, and tell me I'm right if I am" - which seems to be a lot harder. As another example, one common…

What about "check if the user is right"? For thinking or agentic modes this might work.

For example, when someone here inevitably tells me this isn't feasible, I'm going to investigate if they are right before responding ;)

Re: Claude says “You're absolutely right!” about everything

#287

Post by boogieknite »

rahidz wrote: post #2

I'm sure they're aware of this tendency, seeing as "You're absolutely right." was their first post from the @claudeAI account on X: https://x.com/claudeai/status/1950676983257698633 Still irritating though.

early days for all of this but theyve solved so many seemingly more complicated problems id think there would be a toggle which would could remove this from any response

based on your comment maybe its a brand thing? like "just do it" but way dumber. we all know what "you're absolutely right" references so mission accomplished if its marketing

Re: Claude says “You're absolutely right!” about everything

#288

Post by nemomarx »

NohatCoder wrote: post #277

This is such a useful feature. I'm fairly well versed in cryptography. A lot of other people aren't, but they wish they were, so they ask their LLM to make some form of contribution. The result is high level gibberish. When I prod them about the mess, they have to turn to their LLM to deliver a plausibly sounding answer, and that always begins with "You are absolutely right that [thing I mentioned]". So then I don't…

Finally we can get a "watermark" in ai generated text!

Re: Claude says “You're absolutely right!” about everything

#289

Post by lottin »

Gracana wrote: post #179
elif wrote: post #150

I've spent a lot of time trying to get LLM to generate things in a specific way, the biggest take away I have is, if you tell it "don't do xyz" it will always have in the back of its mind "do xyz" and any chance it gets it will take to "do xyz" When working on art projects, my trick is to specifically give all feedback constructively, carefully avoiding framing things in terms of the inverse or parts to remove.

Example-based prompting is a good way to get specific behaviors. Write a system prompt that describes the behavior you want, write a round or two of assistant/user interaction, and then feed it all to the LLM. Now in its context it has already produced output of the type you want, so when you give it your real prompt, it will be very likely to continue producing the same sort of output.

Seems like a lot of work, though.
Post reply on HN