Rendered at 16:11:46 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
fhdkweig 3 hours ago [-]
That's what I thought was interesting about the Watson Jeopardy Challenge. Watson had a sense of its confidence, and it seemed to be accurate. The answers it got wrong were the ones it knew it had difficulty understanding.
I was designing a system that pulls structured data out of an unstructured transcript, and Opus initially suggested a confidence score for each value. When I asked it “do you think the model will generate a meaningful number there?” the answer was basically “ha, you caught me.”
What did work was asking the model to emit citations in the form of line numbers from the transcript. It dramatically dropped hallucination.
abound 2 hours ago [-]
> I asked it “do you think the model will generate a meaningful number there?”
This also isn't likely to give you good results, models don't have a good sense of their own capabilities.
This is also in the same camp of asking a model why it did something. It'll generate something plausible and likely totally incorrect.
bensyverson 1 hours ago [-]
> models don't have a good sense of their own capabilities.
It actually had some memorized knowledge about LLMs being bad at confidence intervals, then it did a cursory web search. But it didn't need to do any of that; I already knew they would be bad numbers.
ptx 4 hours ago [-]
Calculon: And you say you can guarantee me the Oscar?
The statement near the top of the post
> "The short version: asking an LLM to generate a score for how confident it is in its own response is, from everything I can tell, completely useless."
is definitely too strong of a claim and directly undercut by what is said near the end of the post:
> "Tian et al. found in Just Ask for Calibration that with the right prompting strategy, RLHF’d models verbalize probabilities that are better calibrated than the model’s own conditional probabilities, and that prompting plus temperature scaling can cut expected calibration error by more than half. And Anthropic’s Language Models (Mostly) Know What They Know found encouraging results asking models to estimate the probability that their own proposed answer is true."
My own experience is that stated confidence is a helpful tool and of course you need a rubric and a proper prompt, but this is clearly less work than training a classifier (as advocated by the post) and requires less data.
StellaSearch 2 hours ago [-]
Fully agree. I found for most of my work with LLMs and Finance, ~90% of the time a high or low confidence score was accurate. There's the occasional ambiguous case, and that'll happen, but the engineering work that comes with building a classifier makes it not practical for my usecases.
smohare 1 hours ago [-]
[dead]
foo12bar 6 hours ago [-]
There was a post earlier on HN where they trained a probe which could give a realistic confidence score on an LLM, they claim with 81% accuracy. They used it to interrupt and switch to a smarter model if a dumber one had low confidence: https://news.ycombinator.com/item?id=49010782
kqr 4 hours ago [-]
I find the recommendation in favour of discrete categories over numeric probability for confidence level intriguing, because it goes against the grain of all advice I've read for humans interested in forecasting.
The easiest argument in favour of numeric probability is that it's simple to evaluate. Someone claiming to be 90 % certain better be right about nine out of ten times -- no fewer, but also not too often! This is useful when you're learning to judge your own confidence, which most people are bad at. (Almost everyone can train themselves to be better at i but some people seem to be good at it with no training.)
The typical argument against categories is that people can mean very different things with the same word: https://i.ibb.co/kQT4Ymz/q.png
It seems like TFA gets around these problems by creating a translation table between a fixed set of categories and the probabilities implied by the model for those categories. That's an interesting approach!
arm32 56 minutes ago [-]
Interesting read. Dear author, side note, please increase the body font weight, it's barely readable.
dogma1138 4 hours ago [-]
You don’t ask an LLM, but certain LLMs expose internal metrics you can tell how many token candidates where there what was the score which one and which one was selected.
So there are objective ways to control hallucinations as well as figuring out how “correct” the answer is to some extent.
bigfishrunning 3 hours ago [-]
But those scores are only the likelihood of the given token being the next one in a plausible phrase, not that the underlying data is correct.
SubiculumCode 4 hours ago [-]
Who can read that text color and background combo? I had to turn on reader mode in Firefox.
fhdkweig 4 hours ago [-]
Firefox on desktop renders as black text on white background. The link colors are yellow, which is not great, but still okay for me to read.
I use the "Dark Background and Light Text" add-on. Sometimes the add-on makes a bad website worse (like IMDB), but it can be switched off with a click.
ed_mercer 4 hours ago [-]
renders fine here on firefox android, black text on slightly yellow tinted light background
csense 1 hours ago [-]
It's obvious to me that writing "Give me a confidence score from 1-5" in your prompt will have disappointing results if you use that score directly. If you want self-reported confidence scores that are at all useful, you need to think about statistics and score calibration regardless of whether those scores are produced by an LLM or a human.
Take a corpus of problems you know the answers to, but the AI sometimes gets wrong [1]. Have the AI try to solve each problem and give you a 1-5 integer confidence score. The known proportion of correct answers in each bucket gives you a mapping from scores to probabilities [2].
[1] One possible corpus creation strategy might be math problems that you really need to run a program to solve. You can run the program to generate the correct answer key, but disallow tool calling for the AI.
[2] I'm pretty sure there's some stats wizardry that will let you put error bars on each bucket's probability based on the number of problems that end up in each bucket.
bob1029 9 hours ago [-]
I agree with the author if we are trying to use this as some sort of absolute scale of confidence. It only develops meaning when we control for many other variables. Looking at confidence scores across two different models or prompts is probably not a good idea.
mexicocitinluez 4 hours ago [-]
This resonates with what I'm seeing in healthcare right now and the bad taste it's leaving.
There are a whole host of new EMRs popping up that aim to help clinicians make judgement calls about how to answer certain questions and even when particular procedures are relevant. The last one we demoed, each piece of information it retrieved from the LLM has a confidence score attached to it. Our nurses have to fill out 200+ question forms when taking on a new patient that all HAVE to be completed in a single-go meaning that you can't split it up into multiple forms and it's gotta be one cohesive unit.
Imagine being a nurse with little technical skill and almost no idea how these tools work trying to make sense of what the difference between a 90% and 70% is across 200 different questions. "We're 60% sure the patient is allergic to nuts" means jack shit to them. Granted, sometimes the scores are complimented with actual references in the underlying documentation (history and physical, referring info) but sometimes it's not.
t3hTao 4 hours ago [-]
Healthcare still uses fax. No surprise they arent using Frontier AI to answer these questions. (And if you still don't trust AI, have the report on the left side of the screen and the source pages on the right side of the screen)
acdha 4 hours ago [-]
If you work in a field which matters, not trusting even “Frontier AI” is also known as “basic professional competence”. If you’re vibe-coding a marketing site nobody dies if you miss a mistake. The same is not true in healthcare or real engineering, and because AI systems can’t be held accountable you will be.
apwheele 4 hours ago [-]
So I agree with the general geist of this, a few counter-examples though:
Some of the peer reviewed papers with the older models did show calibration was bad (not close to monotonic). This post with newer models (for one example, classifiying injuries) is not that bad, https://gmcirco.github.io/blog/posts/ai-calibration/calibrat....
I mean it just depends on the application, what level of error you can consider. But the second post shows how to recalibrate the scores as well if you need calibrated probabilities.
chrisjj 5 hours ago [-]
> the capability is highly unreliable and highly context-dependent.
Sounds like what's highly unreliable is the evidence, making the "capability" just another imagining.
empthought 4 hours ago [-]
“That’s real, and it’s interesting.”
Stop with the slop.
pamplemeese 26 minutes ago [-]
Dang, my bad. Thought I cleaned that up. To be clear, I use AI to help clean up scattered thoughts, and then I edit it further but sometimes I miss the AI-isms that it inserts.
3 videos at https://archive.org/details/Jeopardy_2011-02-14_The_IBM_Chal...
What did work was asking the model to emit citations in the form of line numbers from the transcript. It dramatically dropped hallucination.
This also isn't likely to give you good results, models don't have a good sense of their own capabilities.
This is also in the same camp of asking a model why it did something. It'll generate something plausible and likely totally incorrect.
It actually had some memorized knowledge about LLMs being bad at confidence intervals, then it did a cursory web search. But it didn't need to do any of that; I already knew they would be bad numbers.
Bender: I can guarantee anything you want.
https://www.youtube.com/watch?v=DGZ10kZ4lmE
is definitely too strong of a claim and directly undercut by what is said near the end of the post: > "Tian et al. found in Just Ask for Calibration that with the right prompting strategy, RLHF’d models verbalize probabilities that are better calibrated than the model’s own conditional probabilities, and that prompting plus temperature scaling can cut expected calibration error by more than half. And Anthropic’s Language Models (Mostly) Know What They Know found encouraging results asking models to estimate the probability that their own proposed answer is true."
My own experience is that stated confidence is a helpful tool and of course you need a rubric and a proper prompt, but this is clearly less work than training a classifier (as advocated by the post) and requires less data.
The easiest argument in favour of numeric probability is that it's simple to evaluate. Someone claiming to be 90 % certain better be right about nine out of ten times -- no fewer, but also not too often! This is useful when you're learning to judge your own confidence, which most people are bad at. (Almost everyone can train themselves to be better at i but some people seem to be good at it with no training.)
The typical argument against categories is that people can mean very different things with the same word: https://i.ibb.co/kQT4Ymz/q.png
It seems like TFA gets around these problems by creating a translation table between a fixed set of categories and the probabilities implied by the model for those categories. That's an interesting approach!
So there are objective ways to control hallucinations as well as figuring out how “correct” the answer is to some extent.
I use the "Dark Background and Light Text" add-on. Sometimes the add-on makes a bad website worse (like IMDB), but it can be switched off with a click.
Take a corpus of problems you know the answers to, but the AI sometimes gets wrong [1]. Have the AI try to solve each problem and give you a 1-5 integer confidence score. The known proportion of correct answers in each bucket gives you a mapping from scores to probabilities [2].
[1] One possible corpus creation strategy might be math problems that you really need to run a program to solve. You can run the program to generate the correct answer key, but disallow tool calling for the AI.
[2] I'm pretty sure there's some stats wizardry that will let you put error bars on each bucket's probability based on the number of problems that end up in each bucket.
There are a whole host of new EMRs popping up that aim to help clinicians make judgement calls about how to answer certain questions and even when particular procedures are relevant. The last one we demoed, each piece of information it retrieved from the LLM has a confidence score attached to it. Our nurses have to fill out 200+ question forms when taking on a new patient that all HAVE to be completed in a single-go meaning that you can't split it up into multiple forms and it's gotta be one cohesive unit.
Imagine being a nurse with little technical skill and almost no idea how these tools work trying to make sense of what the difference between a 90% and 70% is across 200 different questions. "We're 60% sure the patient is allergic to nuts" means jack shit to them. Granted, sometimes the scores are complimented with actual references in the underlying documentation (history and physical, referring info) but sometimes it's not.
If you have the raw log-probs, I show how to use conformal inference to set false-positive or recall rates, https://crimede-coder.com/blogposts/2026/ConfClassification
Some of the peer reviewed papers with the older models did show calibration was bad (not close to monotonic). This post with newer models (for one example, classifiying injuries) is not that bad, https://gmcirco.github.io/blog/posts/ai-calibration/calibrat....
I mean it just depends on the application, what level of error you can consider. But the second post shows how to recalibrate the scores as well if you need calibrated probabilities.
Sounds like what's highly unreliable is the evidence, making the "capability" just another imagining.
Stop with the slop.