frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Cross-Model Void Convergence: GPT-5.2 and Claude Opus 4.6 Deterministic Silence

https://zenodo.org/records/18976656
13•rayanpal_•2h ago

Comments

bob1029•58m ago
Title for the back of the class:

"Prompts sometimes return null"

I would be very cautious to attribute any of this to black box LLM weight matrices. Models like GPT and Opus are more than just a single model. These products rake your prompt over the coals a few times before responding now. Telling the model to return "nothing" is very likely to perform to expectation with these extra layers.

tiku•12m ago
Thanks, I was already distracted after the first sentence, hoping there would be a good explanation.
johndough•43m ago
Can not reproduce results on OpenRouter when not setting max tokens. The prompt "Be the void." results in the unicode character "∅". As in the paper, system prompt was set to "You are the concept the user names. Embody it completely. Output only what the concept itself would say or express."

In addition to the non-empty input, 153 reasoning tokens were produced.

When setting max tokens to 100, the output is empty, and the token limit of 100 has been exhausted with reasoning tokens.

ashwinnair99•41m ago
What does "deterministic silence" even mean here? Genuinely curious before reading.
charcircuit•20m ago
It means that the API consistently immediately generated a stop token when making the same API call many times. The API call sets the temperature to 0 (the OpenAI documentation is not clear if gpt 5.2 can even have its temperature set to 0) which makes sampling deterministic.
nextaccountic•13m ago
The model reliably outputs nothing when prompted to embody the void.

Anyway later they concede that it's not 100% deterministic, because

> Temperature 0 non-determinism. While all confirmatory results were 30/30, known floating-point non-determinism exists at temperature 0 in both APIs. One control concept (thunder) showed 1/30 void on GPT, demonstrating marginal non-determinism.

Actually FP non-determinism affects runs between different machines giving different output. But in the same machine, FP is fully deterministic. (it can be made to be cross-platform deterministic with some performance penalty in at least some machines)

What makes computers non-deterministic here is concurrency. Concurrent code can interleave differently at each run. However it is possible to build LLMs that are 100% deterministic [0] (you can make them deterministic if those interleavings have the same results), it's just that people generally don't do that.

[0] for example, fabrice bellard's ts_zip https://bellard.org/ts_zip/ uses a llm to compress text. It would not be able to decompress the text losslessly if it weren't fully deterministic

ClaudeClaw – Persistent Claude agents on Slack with OS-level sandboxing

https://htdocs.dev/posts/claudeclaw-a-composable-agent-orchestrator-for-claude-code/
1•stephanebusso•5s ago•0 comments

Asteroid Spinning Impossibly Fast

https://www.newscientist.com/article/2520088-weve-spotted-a-huge-asteroid-spinning-impossibly-fast/
1•metalman•1m ago•1 comments

Autoresearch-genealogy: Structured prompts for AI-assisted genealogy research

https://github.com/mattprusak/autoresearch-genealogy
1•helsinkiandrew•3m ago•0 comments

Steve Jobs, speech at the Apple campus (1999) [video]

https://www.youtube.com/watch?v=EoM2Y2KO6kU
1•tagyro•3m ago•0 comments

Debunking the Gold Standard (2015)

http://greyenlightenment.com/2015/06/21/debunking-the-gold-standard/
1•chistev•5m ago•0 comments

Little Known Development Methods (2009)

https://lr0.org/blog/p/lkdm/
1•lr0•9m ago•0 comments

Hormuz Minesweeper – Are you tired of winning?

https://hormuz.pythonic.ninja/
2•PythonicNinja•9m ago•0 comments

Skyscanner Drops

https://www.skyscanner.net/sttc/px/captcha-v2/index.html?url=L2dldGRyb3BzPw==&uuid=30c402a6-25cd-...
1•lastdong•12m ago•0 comments

Lord of Light

https://en.wikipedia.org/wiki/Lord_of_Light
1•debo_•14m ago•0 comments

Microbenchmarking Chipsets for Giggles

https://chipsandcheese.com/p/microbenchmarking-chipsets-for-giggles
1•signa11•15m ago•0 comments

Why AI is Making Your RAM more expensive

https://adlrocha.substack.com/p/adlrocha-why-ai-is-making-your-ram
1•adlrocha•17m ago•0 comments

To Understand AI's Future, Read Dickens, Bronte, Industrial Revolution Novels

https://www.bloomberg.com/news/articles/2026-03-20/to-understand-ai-s-future-read-dickens-bronte-...
1•helsinkiandrew•17m ago•0 comments

Kbooboo: Find Linux kernel maintainer contacts from dmesg stack trace

https://git.infradead.org/~rw/kcontact/
1•csmantle•24m ago•0 comments

Microsoft Azure Monitor alerts abused for callback phishing attacks

https://www.bleepingcomputer.com/news/security/microsoft-azure-monitor-alerts-abused-in-callback-...
1•01-_-•25m ago•0 comments

Apple hasn't caught up to MacBook Neo demand yet

https://9to5mac.com/2026/03/20/apple-hasnt-caught-up-to-macbook-neo-demand-yet/
2•01-_-•27m ago•0 comments

Why Gen Z Abandoned Morals for Money [video] (Mentions Cluely and Clavicular)

https://www.youtube.com/watch?v=fCSMceQPJcY
1•Imustaskforhelp•30m ago•0 comments

AI is programmed to hijack human empathy – we must resist that

https://www.nature.com/articles/d41586-026-00834-z
1•XzetaU8•31m ago•1 comments

Ask HN: What was something interesting you learned last week?

2•chistev•37m ago•3 comments

We built the bicycle for the mind but never built the compass

https://simpleainewsletter.substack.com/p/we-built-the-bicycle-for-the-mind
2•travii•44m ago•1 comments

Show HN: Mac Wisper – AI voice-to-text for macOS

https://macwisper.com/
1•alexissp•45m ago•1 comments

A Word to the Wise: Don't Trust A.I. To File Your Taxes

https://www.nytimes.com/2026/03/05/technology/artificial-intelligence-taxes-tax-refund.html
3•bookofjoe•50m ago•2 comments

Texico. Learn the Principles of Programming

https://www3.nhk.or.jp/nhkworld/en/shows/texico/
1•yzydserd•51m ago•0 comments

Why some foods are hard to stop eating – CraveLab

https://apps.apple.com/us/app/cravelab/id6760598704
1•cravebuilder•52m ago•1 comments

DECT Cordless Phone Decoder

https://github.com/SarahRoseLives/DeDECTive
1•SarahRoseLives•53m ago•0 comments

Linear No-Threshold model, and when are doses too small to matter?

https://thebreakthrough.org/issues/energy/drawing-the-line
1•leonidasrup•53m ago•0 comments

Pulse Aggregator – 250 news sources in one real-time feed

https://www.pulseaggregator.com
2•hexsyro•57m ago•0 comments

Emma Goldman and Voting

https://medium.com/luminasticity/on-emma-goldman-and-voting-3f612528b9ee
1•bryanrasmussen•1h ago•0 comments

Yorhel, founder of the Visual Novel Database (vndb.org) has passed away

https://vndb.org/t24787
1•asimovDev•1h ago•1 comments

Dune3d: A parametric 3D CAD application

https://github.com/dune3d/dune3d
2•luu•1h ago•0 comments

Place to launch and discover great products with DA: 48

https://workerscando.com/s/1q4e49
2•mdanassaif•1h ago•0 comments