frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Meta Superintelligence's surprising first paper

https://paddedinputs.substack.com/p/meta-superintelligences-surprising
93•skadamat•2h ago
https://arxiv.org/abs/2509.01092

Comments

bigyabai•2h ago
> Long awaited first paper from Meta Superintelligence Labs is not a model layer innovation. What does this mean?

It means you're reading into it too much and need to be let down, gently, from the hype train.

nine_k•2h ago
A great post, it starts with this:

TL;DR

• MSI’s first paper, REFRAG, is about a new way to do RAG.

• This slightly modified LLM converts most retrieved document chunks into compact, LLM-aligned chunk embeddings that the LLM can consume directly.

• A lightweight policy (trained with RL) decides which chunk embeddings should be expanded back into full tokens under a budget; the LLM runs normally on this mixed input.

• The net effect is far less KV cache and attention cost, much faster first-byte latency and higher throughput, while preserving perplexity and task accuracy in benchmarks.

I wish more long posts followed this model of a scientific paper.

jongjong•2h ago
Interesting. All developers I know who tinkered around with embeddings and vector similarity scoring were instantly hooked. The efficiency of computing the embeddings once and then reusing as many times as needed, comparing the vectors with a cheap <30-line function is extremely appealing. Not to mention the indexing capabilities to make it work at scale.

IMO vector embedding is the most important innovation in computing of the last decade. There's something magical about it. These people deserve some kind of prize. The idea that you can reduce almost any intricate concept including whole paragraphs to a fixed-size vector which encapsulates its meaning and proximity to other concepts across a large number of dimensions is pure genius.

_jayhack_•1h ago
Vector embedding is not an invention of the last decade. Featurization in ML goes back to the 60s - even deep learning-based featurization is decades old at a minimum. Like everything else in ML this became much more useful with data and compute scale
senderista•1h ago
Yup, when I was at MSFT 20 years ago they were already productizing vector embedding of documents and queries (LSI).
ekidd•1h ago
Vector embeddings are slightly interesting because they come pre-trained with large amounts of data.

But similar ways to reduce huge numbers of dimensions to a much smaller set of "interesting" dimensions have been known for a long time.

Examples include principal component analysis/single value decomposition, which was the first big breakthrough in face recognition (in the early 90s), and also used in latent semantic indexing, the Netflix prize, and a large pile of other things. And the underlying technique was invented in 1901.

Dimensionality reduction is cool, and vector embedding is definitely an interesting way to do it (at significant computational cost).

liampulles•1h ago
If you take the embedding for king, subtract the embedding for male, add the embedding for female, and lookup the closest embedding you get queen.

The fact that dot product addition can encode the concept of royalty and gender (among all other sorts) is kind of magic to me.

mountainriver•1h ago
This was a very obvious next step, I played around with implementing something similar at one point.

In general we need to make it simpler for LLMs to take in different forms of embeddings. At least frameworks that simplify it.

cm2012•1h ago
At first I thought the super intelligence wrote a novel scientific paper
Imnimo•1h ago
I'm curious whether this is work that was specifically begun under the "superintelligence" umbrella, or if it's just that the people who were working on it had been shifted to the Superintelligence team by the time they wrote the paper. I would guess the former?
naasking•59m ago
> the core insight here is actually: if embeddings are generated by layers within the LLM, it makes no sense to convert them back to natural language, just for another LLM to compress those tokens back to embeddings.

Doesn't this tie the two layers together in a way that they can't evolve separately?

xvector•57m ago
Working in big tech it's pretty wild to see how integral AI has become to our work internally, vs the public perception of it. People are NOT prepared.
fishmicrowaver•53m ago
Not prepared for what? Seems like the rest of the world is desperate to be shown the way to unlock something of value?
Workaccount2•38m ago
I think at this point it's software devs looking for the value unlock.

Non-software devs are actually making functional programs for themselves for the first time ever. The value is crazy.

ceejayoz•30m ago
It’s not the first time ever. People did the same with Access and HyperCard in the 90s.
fishmicrowaver•17m ago
Sure but I'm the real world do you think businesses are going to deploy piles of code into production generated this way? No, non technical people will continue to whip up MS PowerApps. AI generated code has no value to many businesses.
terminalshort•52m ago
1. Hyperbolic statement about LLM capabilities with no concrete examples

2. Wild claim that the companies that sell LLMs are actually downplaying their capabilities instead of hyping them

danielmarkbruce•17m ago
Yup, he's totally lying. Not happening. Just carry on.
BoorishBears•12m ago
Agreed, but why are they lying?
godelski•53m ago
It's kinda funny, Meta has long had some of the best in the field, but left them untapped. I really think if they just took a step back and stop being so metric focused and let their people freely explore then they'd be winning the AI race. But with this new team, I feel like meta mostly hired the people who are really good at gaming the system. The people that care more about the money than the research.

A bit of this is true at every major lab. There's tons of untapped potential. But these organizations are very risk adverse. I mean why not continue with the strategy that got us to the point we're at in the first place. Labs used to hire researchers and give them a lot of free reign. But those times ended and AI progress also slowed down. Maybe if you want to get ahead you gotta stop thinking like everyone else

Well meta... you can "hold me hostage" for a lot cheaper than those guys. I'm sure this is true for hundreds of passionate ML researchers. I'd take a huge pay cut to have autonomy and resources. I know for a fact there's many working at Mets right now that would do the same. Do maybe if you're going to throw money at the problem, diversify a bit and look back at what made SV what it is today and what made AI take leaps forward

ipsum2•48m ago
This has nothing to do with superintelligence, it's just the people that were working on the paper prior to the re-org happened to publish after the name change.

Though it is notable that contrary to many (on HN and Twitter) that Meta would stop publishing papers and be like other AI labs (e.g. OpenAI). They're continued their rapid pace of releasing papers AND open source models.

bigcat12345678•28m ago
https://docs.lamini.ai/memory_rag/ Similar approaches have been tried before already
pppoe•23m ago
I find it absurd that, compared to the past, large companies now have more abundant stock prices and cash than ever before, yet nearly every AI Lab in these companies is facing greater pressure than ever, being asked to generate short-term profits. In the midst of AI's unprecedented boom, the research environment and atmosphere in the industry seem to have worsened compared to the past.
signatoremo•14m ago
Is this Meta’s lab pressured to generate short term profits?

Which other under pressure labs are you talking about?

foldl2022•22m ago
So, show me the model weights, please.
yalogin•14m ago
I am not surprised because the culture at meta is not at all, even in the slightest, to focus on science for the sake of it. It’s actively actively purged out of you. The focus is on metrics and how the bottom line is impacted. So this is in line with that

Labels in HTML

https://pureooze.com/blog/posts/2025-10-07-memory-bank-labels-in-html/
1•pureooze•33s ago•0 comments

Israeli Spyware Vendor NSO Group Reportedly Sells to U.S. Company

https://gizmodo.com/creepy-israeli-spyware-vendor-nso-group-reportedly-sells-to-u-s-company-at-th...
1•diogenes_atx•1m ago•0 comments

Africa's oldest leader isn't ready to retire; he's not the only one defying age

https://text.npr.org/nx-s1-5570875
1•mooreds•3m ago•0 comments

Ed Davey urges regulator to go after Musk over X 'crimes'

https://www.bbc.com/news/articles/c5y558r2r1qo
1•donsupreme•4m ago•0 comments

Why Can't Fashion See What It Does to Women?

https://www.nytimes.com/2025/10/10/style/women-designers-paris-fashion-week.html
1•mooreds•5m ago•0 comments

I've Gone to Look for America

https://magazine.atavist.com/2025/america-i95-conversations-trump
1•mooreds•5m ago•0 comments

A GenZ billionaire:Shayne Coplan figured out a society that gambles on everythin

https://fortune.com/2025/10/11/shayne-coplan-polymarket-youngest-self-made-billionaire-nyse-inves...
1•thelastgallon•6m ago•1 comments

Before Haskell, there was the Orwell programming language [pdf]

https://homepages.inf.ed.ac.uk/wadler/papers/orwell/orwell2.pdf
1•KnuthIsGod•9m ago•0 comments

BillionToOne IPO

https://aseq.substack.com/p/billiontoone-ipo
1•new299•11m ago•0 comments

China's New Rare Earth and Magnet Restrictions Threaten US Defense Supply Chains

https://www.csis.org/analysis/chinas-new-rare-earth-and-magnet-restrictions-threaten-us-defense-s...
6•stopbulying•11m ago•0 comments

Trump Administration Lays Off CDC Officials

https://www.nytimes.com/2025/10/11/us/politics/trump-administration-cdc-layoffs.html
3•JKCalhoun•12m ago•0 comments

Gitid: Nvm for Git identities, now supports per-repo identity

https://github.com/nathabonfim59/gitid
1•nathabonfim59•12m ago•0 comments

FOXP3 and scurfy: how it all began

https://www.nature.com/articles/nri3650
1•antimora•17m ago•0 comments

Bilingual Hebrew Liturgical Songs Sung Beautifully

https://klappn.com/
1•KnuthIsGod•21m ago•1 comments

Power-over-Fiber

https://en.wikipedia.org/wiki/Power-over-fiber
1•thelastgallon•23m ago•0 comments

People on the far-right and far-left exhibit strikingly similar brain responses

https://www.psypost.org/people-on-the-far-right-and-far-left-exhibit-strikingly-similar-brain-res...
1•nis0s•24m ago•0 comments

Terminal Lucidity: Envisioning the Future of the Terminal

https://arxiv.org/abs/2504.13994
1•felineflock•25m ago•0 comments

How Google Is Walking the AI Tightrope

https://www.wsj.com/tech/ai/google-search-ai-business-48e13735
1•Brajeshwar•31m ago•0 comments

Susam Pal: My Lobsters Interview

https://susam.net/my-lobsters-interview.html
2•signa11•32m ago•0 comments

Study Traces Autism's Origin to the Rise of Human Intelligence

https://www.sciencealert.com/study-traces-autisms-origin-to-the-rise-of-human-intelligence
1•nis0s•43m ago•0 comments

The 'profound' global impact of China's rise as an electrostate

https://www.ft.com/content/013e8a27-ade5-48ed-8f2e-ffbf70cc508c
5•bookofjoe•45m ago•1 comments

What do coyotes think?

https://andys.blog/what-do-coyotes-think/
2•andytratt•47m ago•0 comments

ROSA+: RWKV's ROSA implementation with fallback statistical predictor

https://github.com/bcml-ai/rosa-plus
2•dedicateddev•53m ago•0 comments

Marc Benioff Says Trump Should Deploy National Guard to San Francisco

https://www.latimes.com/california/story/2025-10-11/tech-billionaire-marc-benioff-says-trump-shou...
7•bryan0•54m ago•1 comments

Putin Has a New Tool to Monitor Russians

https://www.theatlantic.com/international/archive/2025/10/russia-super-app-max/684524/
3•JumpCrisscross•1h ago•1 comments

1990s Millport CNC Vertical Mill Revival

https://salvagedcircuitry.com/90s-cnc-revival.html
3•beckthompson•1h ago•0 comments

Chinese EV giant BYD sees UK sales soar by 880%

https://www.bbc.com/news/articles/c3w5jl2jgqwo
13•JumpCrisscross•1h ago•1 comments

The Rise of 'Conspiracy Physics'

https://www.wsj.com/science/physics/the-rise-of-conspiracy-physics-dd79fe36
6•joak•1h ago•1 comments

The uranium plant at the center of U.S. plans to expand nuclear power

https://www.cnbc.com/2025/10/04/urenco-centrus-orano-enriched-uranium-nuclear-russia-ai-data-cent...
2•JumpCrisscross•1h ago•0 comments

OpenRouter drops fees in response to Vercel's AI Gateway

https://www.coplay.dev/blog/openrouter-drops-fees-in-response-to-vercel-s-ai-gateway
2•josvdwest•1h ago•0 comments