frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

Open in hackernews

Universal pre-training by iterated random computation

https://arxiv.org/abs/2506.20057
22•liamdgray•5h ago

Comments

liamdgray•5h ago
Abstract: "We investigate the use of randomly generated data for the sake of pre-training a model. We justify this approach theoretically from the perspective of algorithmic complexity, building on recent research that shows that sequence models can be trained to approximate Solomonoff induction. We derive similar, but complementary theoretical results. We show empirically that synthetically generated data can be used to pre-train a model before the data is seen. We replicate earlier results that models trained this way show zero-shot in-context learning across a variety of datasets, and that this performance improves with scale. We extend earlier results to real-world data, and show that finetuning a model after pre-training offers faster convergence and better generalization."
bionhoward•3h ago
This is a cool concept, but for comparison, I can’t help but wish there was more comparison between the treatment group and a control group that doesn’t see any universal pretraining data.

It’s good to compare various model sizes and evaluation tasks and random data generators. I just think the paper would more effectively prove its point if it could show models of same sizes which see this random data can learn better from evaluation data later on.

Could even take the initial checkpoint of the model before universal pretraining against the pretrained checkpoint. If the method works, the one that did UP will win.

Maybe I’m way off, I’ll admit I only skimmed it so far. Seems promising, just wishing for some controls.

yorwba•39m ago
In figures 2, 4, and 6, the top left end of the training curves represents models that have not seen any pretraining data. In figure 5, they're represented by dashed curves.
visarga•2m ago
Results are modest, maybe 20-30% fewer training steps to reach target performance. This won't solve the problem of organic data exhaustion. We need 100x more data.

They didn't test against actual language model pretraining, only tested against a random init.

- A: Pre-trained on their synthetic LSTM data -> fine-tuned on Wikipedia

- B: Pre-trained on different natural language corpus -> fine-tuned on Wikipedia

- C: Random initialization -> fine-tuned on Wikipedia

They only test A vs C, not A vs B.

Scientists 'freeze' light into a supersolid using 'quantum theatre'

https://www.thehindu.com/sci-tech/science/scientists-freeze-light-into-a-supersolid-using-quantum-theatre/article69748118.ece
3•Bluestein•7m ago•0 comments

It's Known as 'The List'–and It's a Secret File of AI Geniuses

https://www.wsj.com/tech/meta-ai-recruiting-mark-zuckerberg-openai-018ed7fc
1•pretext•8m ago•1 comments

Magnetic Tape Storage Technology: usage, history, and future outlook

https://dl.acm.org/doi/10.1145/3708997
2•matt_d•15m ago•0 comments

Societal conditions explain differences in "dark" personality across regions

https://www.pnas.org/doi/10.1073/pnas.2500830122
1•PaulHoule•15m ago•0 comments

Benchmark for Evaluating Text Embeddings

https://huggingface.co/spaces/embedding-benchmark/RTEB
1•fzliu•17m ago•0 comments

Review: Alpha School

https://www.astralcodexten.com/p/your-review-alpha-school
1•stephenbez•19m ago•0 comments

Triple SEC: Simple Digital Security Scheme

https://nau.github.io/triplesec/
1•hosteur•21m ago•1 comments

NFC Release 15: The what, why and how

https://nfc-forum.org/news/2025-06-nfc-release-15-the-what-why-and-how/
1•ksec•21m ago•1 comments

Show HN: DNS at ludicrous speed for Go, powered by XDP sockets

https://github.com/dwisiswant0/fastdns
2•dwisiswant0•34m ago•0 comments

Facebook wants unpublished images on smartphones

https://www.heise.de/en/news/Facebook-wants-unpublished-images-on-smartphones-10463407.html
2•doener•37m ago•0 comments

Ottawa orders Chinese manufacturer Hikvision to shutter Canadian operations

https://www.channelnewsasia.com/business/ottawa-orders-chinese-manufacturer-hikvision-shutter-canadian-operations-5208716
1•doener•39m ago•0 comments

Show HN: Memory-Rush – A simple emoji memory game

https://memory-rush-game.vercel.app/
1•sanchitak•40m ago•0 comments

So Long, Blue Screen of Death

https://www.wired.com/story/so-long-blue-screen-of-death-amazingly-youll-be-missed/
1•saikatsg•40m ago•0 comments

A Breakdown of Single Player Role-Playing Games

https://wizardsrespite.com/2024/02/13/solo-rpgs-a-breakdown-of-single-player-role-playing-games/
1•doener•42m ago•0 comments

Show HN: Daily word puzzle – Guess the consonants of a word

https://consonants21.com
1•nbhat•45m ago•0 comments

How Biased Is the Criminal Justice System?

https://www.stevestewartwilliams.com/p/how-biased-is-the-criminal-justice
2•mpweiher•55m ago•0 comments

Space probe creates its first on-demand solar eclipse

https://newatlas.com/space/space-probe-creates-solar-eclipses-demand/
3•sharpshadow•57m ago•0 comments

Does CPS Investigate One Third of All Children in the US?

https://www.maximum-progress.com/p/does-cps-investigate-one-third-of
2•Ozarkian•57m ago•0 comments

Gemini 2.5 is getting to the heart of the matter

2•sans_souse•1h ago•2 comments

Balaji on AI

https://twitter.com/balajis/status/1938840903692755135
1•nanfinitum•1h ago•0 comments

Where do AI teams get affordable, high-quality labeled data?

1•fungungun•1h ago•0 comments

Brave creates new TLD on the blockchain

https://brave.com/blog/brave-tld/
2•meander_water•1h ago•0 comments

The World Loanword Database (WOLD)

https://wold.clld.org/
1•Tomte•1h ago•0 comments

Show HN: I built a chatroom that only accepts emoji

https://emojionly.chat
1•MichaelYuhe•1h ago•0 comments

A new dataviz+streaming project all about The Office (2020)

https://buttondown.com/willchase/archive/a-new-datavizstreaming-project-all-about-the/
1•Tomte•1h ago•0 comments

Desktop Extensions: One-Click MCP Server Installation for Claude Desktop

https://www.anthropic.com/engineering/desktop-extensions
1•ubolonton_•1h ago•0 comments

Directing TEL Links to WhatsApp Desktop in Windows

https://karmanivero.us/directing-tel-links-to-whatsapp-desktop-in-windows/
1•karmaniverous•1h ago•1 comments

What if every browser tab became its own AI agent?

https://snowx.ai/
1•Kn1026•1h ago•1 comments

Douglas Hofstadter on Loops, Beauty, Free Will, AI, God, Utopia and Gaza

https://johnhorgan.org/cross-check/hofstadter-on-strange-loops-beauty-free-will-ai-god-utopia-and-gaza
6•squirrel•1h ago•0 comments

Slouching Towards Sensemaking

https://karanchawla.io/2025/06/29/sensemaking
2•karchaw•1h ago•0 comments