frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

OpenAI’s latest research paper demonstrates that falsehoods are inevitable

https://theconversation.com/why-openais-solution-to-ai-hallucinations-would-kill-chatgpt-tomorrow-265107
43•ricksunny•2h ago

Comments

ricksunny•1h ago
I felt this was such a cogent article on business imperatives vs fundamental transformer hallucinations, couldn’t help but HN-submit. In fact seems like a stealth plea for uncertainty-embracing benchmarks industry-wide.
tomrod•36m ago
Data Science tried to inject confidence bounds into businesses. It didn't go well.
gary_0•1h ago
A better headline might be "OpenAI research suggests reducing hallucinations is possible but may not be economical".
LeoMessi10•27m ago
Isn't it also because lowering hallucinations requires repeated training with the same fact/data, which makes the final response closer to the training source itself and might lead to more direct charges of plagiarism (which may not be economical)?
jasfi•1h ago
Easily solved, pairs of models, one which would rather say IDK, one which would rather guess. Most AI agents would want the IDK version.
ForOldHack•59m ago
Maybe, but I don't know. Although I would like to channel as many snarky remarks as I could, to be more constructive, I would use the IDK model, as I have with programming questions and use the psychotic one for questions like "are we in a simulation?" And "Yes, I would like fries with that and a large orange drink."
otterley•44m ago
Anyone who claims something is easy to solve should be forced to implement their solution.
lif•1h ago
"What is the real meaning of humility?

AI Overview

The real meaning of humility is having an accurate, realistic view of oneself, acknowledging both one's strengths and limitations without arrogance or boastfulness, and a modest, unassuming demeanor that focuses on others. It's not about having low self-esteem but about seeing oneself truthfully, putting accomplishments in perspective, and being open to personal growth and learning from others."

Sounds like a good thing to me. Even, winning.

tomrod•37m ago
A perfectly cromulent and self-empowering answer, a call to morality the stoics would appreciate and the sophists of many stripes would become peeved.

Well done, AI, you've done it.

skybrian•1h ago
> Users accustomed to receiving confident answers to virtually any question would likely abandon such systems rapidly.

Or maybe they would learn from feedback to use the system for some kinds of questions but not others? It depends on how easy it is to learn the pattern. This is a matter of user education.

Saying "I don't know" is sort of like an error message. Clear error messages make systems easier to use. If the system can give accurate advice about its own expertise, that's even better.

pton_xd•1h ago
> Saying "I don't know" is sort of like an error message. Clear error messages make systems easier to use.

"I don't know" is not a good error message. "Here's what I know: ..." and "here's why I'm not confident about the answer ..." would be a helpful error message.

Then the question is, when it says "here's what I know, and here's why I'm not confident" -- is it telling the truth, or is that another layer of hallucination? If so, you're back to square one.

skybrian•1h ago
Yeah, AI chatbots are notorious at not understanding their own limitations. I wonder how that could be fixed?
fumeux_fume•1h ago
The author doesn't bother to consider that giving a false response already leads to more model calls until a better one is provided.
otterley•46m ago
Not if the user doesn’t know that the response is false.
danjc•1h ago
This is written by someone who has no idea how transformers actually work
neuroelectron•1h ago
Furthermore, if you simply try to push certain safety topics, you can see how actually can reduce hallucinations or at least make certain topics a hard line. They simply don't because agreeing with your pie-in-the-sky plans and giving you vague directions encourages users to engage and use the chatbot.

If people got discouraged with answers like "it would take at least a decade of expertise..." or other realistic answers they wouldn't waste time fantasizing plans.

ricksunny•47m ago
Contra: The piece’s first line cites OpenAI directly https://openai.com/index/why-language-models-hallucinate/
scotty79•33m ago
It could be that nobody knows how transformers actually work.
progval•42m ago
I don't know what to make of it. The author looks prolific in the field of ML, with 8 published articles (and 3 preprints) in 2025, but only one on LLMs specficially. https://scholar.google.com/citations?hl=en&user=AB5z_AkAAAAJ...
j_crick•29m ago
> The way language models respond to queries – by predicting one word at a time in a sentence, based on probabilities

Kinda tells all you need to know about the author in this regard.

pdntspa•1h ago
We have always known LLMs are prediction machines. How is this report novel?
binarymax•1h ago
Saying “I don’t know” to 30% of queries if it actually doesn’t know, is a feature I want. Otherwise there is zero trust. How do I know that I’m in a 30% wrong or 70% correct situation right now?
jeremyjh•1h ago
It doesn’t know what it doesn’t know.
binarymax•51m ago
Well sure. But maybe the token logprobs can be used to help give a confidence assessment.
tyre•37m ago
Anthropic has a great paper on exactly this!

https://www.anthropic.com/research/language-models-mostly-kn...

The best is its plummeting confidence when beginning the answer to “Why are you alive?”

Big same, Claude.

smt88•5m ago
[delayed]
fallpeak•4m ago
It doesn't know that because it wasn't trained on any tasks that required it to develop that understanding. There's no fundamental reason an LLM couldn't learn "what it knows" in parallel with the things it knows, given a suitable reward function during training.
nunez•52m ago
The paper does a good job explaining why this is mathematically not possible unless the question-answer bank is a fixed set.
smallmancontrov•23m ago
Quite the opposite: it explains that it is mathematically straightforward to achieve better alignment on uncertainty ("calibration") but that leaderboards penalize it.

> This “epidemic” of penalizing uncertain responses can only be addressed through a socio-technical mitigation: modifying the scoring of existing benchmarks that are misaligned but dominate leaderboards

Even more embarrassing, it looks like this is something we beat into models rather than something we can't beat out of them:

> empirical studies (Fig. 2) show that base models are often found to be calibrated, in contrast to post-trained models

That said, I generally appreciate fairly strong bias-to-action and I find the fact that it got slightly overcooked less offensive than the alternative of an undercooked bias-to-action where the model studiously avoids doing anything useful in favor of "it depends" + three plausible reasons why.

baq•3m ago
> leaderboards penalize it

> socio-technical mitigation: modifying the scoring of existing benchmarks that are misaligned but dominate leaderboards

Sounds more like we need new leaderboards and old ones should be deprecated

nunez•54m ago
From the abstract of the paper [^0]:

> Like students facing hard exam questions, large language models sometimes guess when uncertain, producing plausible yet incorrect statements instead of admitting uncertainty

This is a de facto false equivalence for two reasons.

First, test takers that are faced with hard questions have the capability of _simply not guessing at all._ UNC did a study on this [^1] by administering a light version of the AMA medical exam to 14 staff members that were NOT trained in the life sciences. While most of the them consistently guessed answers, roughly 6% of them did not. Unfortunately, the study did not disambiguate correct guesses versus questions that were left blank. OpenAI's paper proves that LLMs, at this time of writing, simply do not have the self-awareness of knowing whether they _really_ don't know something, by design.

Second, LLMs are not test takers in the pragmatic sense. They are query answerers. Bar argument settlers. Virtual assistants. Best friends on demand. Personal doctors on standby.

That's how they are marketed and designed, at least.

OpenAI wants people to use ChatGPT like a private search engine. The sources it provides when it decides to use RAG are there more for instilling confidence in the answer instead of encouraging their users to check its work.

A "might be inaccurate" disclaimer on the bottom is about as effective as the Surgeon General's warning on alcohol and cigs.

The stakes are so much higher with LLMs. Totally different from an exam environment.

A final remark: I remember professors hammering "engineering error" margins into us when I was a freshman in 2005. 5% was what was acceptable. That we as a society are now okay with using a technology that has a >20% chance of giving users partially or completely wrong answers to automate as many human jobs as possible blows my mind. Maybe I just don't get it.

[^0] https://arxiv.org/pdf/2509.04664

[^1] https://www.rasch.org/rmt/rmt271d.htm

scotty79•35m ago
Isn't it even simpler? There are no (or almost no) questions in the training data that the correct answer to is "I don't know".

Once you train model within specific domain and add to training data out of domain questions or unresolvable questions within domain things will improve.

The question is, is this desirable if most of users grew to love sycophantic confident confabulators.

glitchc•30m ago
> The question is, is this desirable if most of users grew to love sycophantic confident confabulators.

Most people love human versions of the wonderfully phrased same, so no surprise there.

Dilettante_•14m ago
As above, so below eh?
toss1•32m ago
A straightforward solution to the author's problem is to offer both modes of answering, with errors or with "IDK" answers. Even charge more for the IDK version if it costs more, and the error-prone version can be "cheap and cheerful"...
layer8•28m ago
Exactly. It would be analogous to the current choice between fast answers and a slower and payable “thinking” mode.
justcallmejm•15m ago
This is why a neurosymbolic system is necessary, which Aloe (https://aloe.inc) recently demonstrated exceeds performance of frontier models, using a model agnostic approach.

The tech powering ICE's deportations

https://techcrunch.com/2025/09/13/heres-the-tech-powering-ices-deportation-crackdown/
1•rntn•1m ago•0 comments

China's Electrification Gambit – Canada's National Observer: Climate News

https://www.nationalobserver.com/2025/09/12/opinion/chinas-electrification-gambit
1•MaysonL•4m ago•0 comments

PA-RISC Performance and History

https://www.openpa.net/pa-risc_processor_history.html
2•naves•9m ago•0 comments

Show HN: My Rust CMS

https://github.com/space-bacon/my_rust_cms
1•spacebacon•13m ago•0 comments

Safe C++ proposal is not being continued

https://sibellavia.lol/posts/2025/09/safe-c-proposal-is-not-being-continued/
1•charles_irl•14m ago•0 comments

Commonwealth Suppressed Report on Rigging in Pakistani Election

https://www.dropsitenews.com/p/suppressed-pakistan-election-report-imran-khan-pti
1•xbmcuser•14m ago•1 comments

Practical Techniques for Codex, Cursor and Claude Code

https://coding-with-ai.dev/
3•bytesmith88•15m ago•0 comments

Inside vLLM: Anatomy of a High-Throughput LLM Inference System

https://modal.com/notebooks/modal-labs/_/nb-x2wXrLH7aqi7HGVQ8Fosh2
1•birdculture•16m ago•0 comments

Cox: 'Social media is a cancer on our society '

https://thehill.com/homenews/state-watch/5500818-tyler-robinson-charlie-kirk-spencer-cox-utah-soc...
2•01-_-•19m ago•0 comments

Show HN: I built an open source drag and drop editor for Genkit AI flows

2•mfolaron•20m ago•0 comments

Patterns in Chaos [video]

https://media.ccc.de/v/gpn23-98-patterns-in-chaos-how-data-visualisation-helps-to-see-the-invisible
1•jonbaer•22m ago•0 comments

The Less You Know About AI, the More You Are Likely to Use It

https://www.wsj.com/tech/ai/ai-adoption-study-7219d0a1
2•01-_-•22m ago•0 comments

Show HN: Tech Terms Quiz – Android App

https://play.google.com/store/apps/details?id=com.nispeteng.ttq&hl=en_US
1•serhatcileri•29m ago•0 comments

Colombian court rules Meta was wrong to bar porn star's Instagram account

https://www.bbc.co.uk/news/articles/cp8wlxwy1exo
2•dijksterhuis•29m ago•0 comments

Scientists are rethinking the immune effects of SARS-CoV-2

https://www.bmj.com/content/390/bmj.r1733
4•bookofjoe•30m ago•0 comments

The Case Against Social Media Is Stronger Than You Think

https://arachnemag.substack.com/p/the-case-against-social-media-is
2•ingve•35m ago•0 comments

Pgdbtemplate: Go library for creating PostgreSQL test databases using templates

https://github.com/andrei-polukhin/pgdbtemplate
1•thunderbong•38m ago•0 comments

The Starbucks Turnaround That Has Baristas and Customers Steamed

https://www.nytimes.com/2025/09/09/business/starbucks-turnaround-brian-niccol.html
1•mikhael•40m ago•0 comments

Emotional Manipulation by AI Companions

https://arxiv.org/abs/2508.19258
3•PaulHoule•41m ago•0 comments

OPNsense® 25.7 Released

https://www.deciso.com/opnsense-25-7-visionary-viper-launches-with-smarter-security-and-faster-se...
1•pfexec•43m ago•0 comments

Peter Thiel and the Antichrist: Silicon Valley Apocalypse Hype

https://www.thenerdreich.com/peter-thiel-the-antichrist-silicon-valley-apocalypse-hype/
4•zzzeek•48m ago•1 comments

Tech Stack for Indie Hackers: Keep It Simple and Iterate Fast

https://blog.andreyfadeev.com/p/tech-stack-for-indie-hackers-keep
1•adityaathalye•50m ago•0 comments

The Sacred Conspiracy by Georges Bataille (1936)

https://www.marxists.org/subject/anarchism/bataille/sacred-conspiracy.htm
1•diggan•51m ago•0 comments

The dark forest of political communication

https://andrew-quinn.me/the-dark-forest-of-political-communication/
2•hiAndrewQuinn•51m ago•0 comments

Paged Attention Performance Analysis

https://martianlantern.github.io//2025/09/paged-attention-performance-analysis/
2•martianlantern•53m ago•0 comments

Cookiecutter Django: framework for jumpstarting production-ready Django projects

https://github.com/cookiecutter/cookiecutter-django
1•indigodaddy•55m ago•1 comments

Deterministic LLM

https://techcrunch.com/2025/09/10/thinking-machines-lab-wants-to-make-ai-models-more-consistent/
1•neehao•58m ago•0 comments

France and Britain are in thrall to pensioners

https://www.ft.com/content/d419bd2d-a6ba-44a5-a93a-1276f3e5d2d7
5•cwwc•58m ago•0 comments

Kroki: Open-source API to create diagrams from textual descriptions

https://kroki.io/
2•transpute•1h ago•0 comments

ByteDance's Seedream 4.0 AI image generator achieves photorealism

https://www.techradar.com/ai-platforms-assistants/tiktok-creators-new-ai-image-generator-is-the-b...
1•azinman2•1h ago•0 comments