frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Tailwind Labs is joining Shopify

https://tailwindcss.com/blog/tailwind-is-joining-shopify
481•EdwinHoksberg•3h ago•187 comments

No Man's Sky Cosmos

https://www.nomanssky.com/cosmos-update/
77•Limb•54m ago•52 comments

GPT-6 Astra, Looped Transformers, and Hidden Reasoning

https://magazine.sebastianraschka.com/p/gpt-6-astra-looped-transformers-and
65•ModelForge•2h ago•8 comments

Claude, change the "Add to Cart" button to blue

https://opusfived.dev/
620•matthieu_bl•7h ago•236 comments

Desert Ant Labs: local, fast models that run on device

https://desertant.com/blog/introducing-desert-ant-labs/
264•willwhitedc•5h ago•69 comments

Anthropic Is Building a Predictive Surveillance System to Monitor Activists

https://prospect.org/2026/09/09/anthropic-artificial-intelligence-surveillance-system-monitor-act...
121•dn2k•40m ago•35 comments

GNU Radio in the Browser

https://gnuradioworld.com/
26•kristianpaul•49m ago•5 comments

Planet Labs' Open Satellite Feed

https://tech.marksblogg.com/planet-labs-open-satellite-feed.html
23•marklit•58m ago•2 comments

Rails 8 Guide: Features, Requirements and Upgrade Path (2026)

https://blog.appsignal.com/2024/10/07/whats-new-in-ruby-on-rails-8.html
19•andreigaspar•59m ago•3 comments

Defining AI Psychosis. Part 2: "Prolific AI Psychosis"

https://jeffs.blog/p/defining-ai-psychosis-part-2-prolific
13•euthymiclabs•29m ago•2 comments

We Accidentally Built a Synthetic Cell Factory

https://bnext.bio/post/we-accidentally-built-a-synthetic-cell-factory
21•rajivm•1h ago•0 comments

I advertise malicious software on Google Ads

https://xlii.space/eng/malicious-software-on-google-ads/
252•xlii•4h ago•147 comments

Investing in Mothers? The Long-Run Impact of a Universal Child Care

https://www.nber.org/papers/w35514
49•throw0101d•1h ago•49 comments

Bespoke: A Programming Language for People Who Say Please

https://blog.hofstede.it/bespoke-a-programming-language-for-people-who-say-please/
15•birdculture•3d ago•4 comments

Roame (YC S23) Is Hiring Viral Content Editor

https://www.ycombinator.com/companies/roame/jobs/KuVVqSh-content-systems-builder-editor
1•zman0225•4h ago

DeepSeek launching v4.1 flash cheaper and more capable than v4 pro

318•nickweb•5h ago•166 comments

Better AI code comment detector

https://entropicthoughts.com/better-ai-comment-classifier
13•ibobev•56m ago•8 comments

Microsoft/TracerAI withdraws copyright takedown against Luanti

https://blog.luanti.org/2026/09/08/dmca-rescinded/
21•daper•2h ago•4 comments

Muse – Meta’s personal AI agent

https://ai.meta.com/muse/
599•yks•21h ago•652 comments

Understanding the Recent DDoS Attack Against Read the Docs

https://about.readthedocs.com/blog/2026/09/2026-ddos-attack/
7•davidfischer•46m ago•0 comments

Playing whack-a-mole is losing

https://dadrian.io/blog/posts/whack-a-mole-is-losing/
33•surprisetalk•3h ago•8 comments

Coyote v. Acme (1990)

https://www.newyorker.com/magazine/1990/02/26/coyote-v-acme
111•ChrisArchitect•3d ago•51 comments

Rivian's Gambit for Full Autonomy

https://spectrum.ieee.org/rivian-self-driving
7•1970-01-01•21h ago•5 comments

Navier-Stokes – Tristan Buckmaster [pdf]

https://cims.nyu.edu/~tristanb/statement.pdf
1935•procedurecall•1d ago•789 comments

Show HN: Geiger – See every AI agent on your machine and what it can touch

https://github.com/Atomburstofficial/geiger
27•atomburst•1h ago•15 comments

Bomb Sense

https://danboland.net/2021/07/15/bomb-sense.html
6•surprisetalk•1h ago•0 comments

Searching for the best silicone USB cable

https://www.frankchiarulli.com/blog/best-silicone-usb-cable/
65•evakhoury•4d ago•41 comments

Building a Wall Lamp from Scratch

https://mbugert.de/posts/2026-09-09-bedroom-lamp-build/
64•jcklie•6h ago•24 comments

Smolts: A pedagogical IDE for a teaching language

https://eighty-twenty.org/2026/09/04/smolts
34•tonyg•5d ago•2 comments

Lotus Notes and the dangers of starting from scratch

https://buttondown.com/blog/lotus-notes-email
132•maguay•6h ago•85 comments
Open in hackernews

GPT-6 Astra, Looped Transformers, and Hidden Reasoning

https://magazine.sebastianraschka.com/p/gpt-6-astra-looped-transformers-and
57•ModelForge•2h ago

Comments

libraryofbabel•28m ago
Everyone interested in LLM internals should read Sebastian. He's great.

The tldr here is that the recent "The Information" article[0] reporting GPT 6 Astra was using “recurrent depth” or “looped transformers" made it sound like it was some special new scary thing ("secret technique!") that made train-of-thought monitoring harder to do. In fact, it's just the same as stacking more transformer layers, except that you reuse the weights and so save GPU memory. It's still just producing one token at a time, and the token sequence positions aren't interacting in any "recurrent" way that's different from a regular LLM architecture.

So, you can still monitor train of thought with these models just fine... well, if you're OpenAI, anyway. Users haven't been able to see an unsummarized trace since o1 days, because the labs are worried about distillation of their models by Chinese labs.

(There are some legitimate interpretability concerns about stacking transformer layers endlessly, but we're known about that for a long time. And the "looping" here isn't really the source of any new issues here, except insofar as it's a cheap way to add more layers.)

[0] https://www.theinformation.com/articles/secret-technique-beh...

famouswaffles•22m ago
>made it sound like it was some special new scary thing that made train-of-thought monitoring harder to do.

It's not a "scary new thing" but ultimately no-one knows exactly how OpenAI have implemented looping. You might not be aware/remember but MoE transformers perennially underperfomed their dense counterparts until GPT-4. Similarly, making reinforcement learning really work with transformers wasn't figured out until o1.

And by Open AI's own admission, Astra's CoT is significantly harder to monitor and it exhibits a significantly greater control over its own CoT than any other model released.

0c3ca83•12m ago
"Don't worry, it'll make us rich -- and that's nearly the same as everything being just fine"
libraryofbabel•5m ago
Well sure, that's the possible weak point in Sebastian's article: it could be true that there's some more sophisticated stuff going on in Astra around looping, because OpenAI haven't specified their architecture. But it's always been true that, since we don't know what's in their black box, there could be arbitrary amounts of innovations inside the models that we could speculate about. So the question is, does knowing they use "looped transformers" really add any dramatically new information that we should worry about? And what this article is saying is, not really, because the mostly likely pattern that's referring to is just, effectively, stacking layers and reusing weights.

> And by Open AI's own admission, Astra's CoT is significantly harder to monitor and it exhibits a significantly greater control over its own CoT than any other model released.

Oh sure; I don't think anyone is denying that larger issue? But does it have anything to do with looping?

aabhay•8m ago
If the agent is able to “decide” when a loop should occur vs when an output token is produced, that effectively moves the CoT inside the architecture. While that’s not what is happening here, it’s clearly a plausible way we could see CoT disappear.
throw3954•4m ago
It’s a little more complicated than that. While looped transformers can be unrolled a fixed number of times to save on memory, if loop depth is determined dynamically between tokens, a single transformer can compute any computable function between tokens.

To analogize, current transformers run a fixed-length program per step. Any program can be factored into a top-level loop with a fixed-length branching body (an interpreter). Dynamically looped transformers can run any program between tokens.

The safety argument for CoT monitoring is that in transformers information about the hidden state has to be communicated through the bottleneck of sampling a single token per forward pass. If not trained adversarially, it’s likely that a reasoning trace contains all the “bottlenecked information” we need to determine intent. But if we can compute arbitrary programs between tokens, the reasoning used is hidden.

It also opens the door to simple architectural extensions that would make the safety/monitoring side of things much more difficult.

It’s probably fine in practice at these scales though. If we keep each loop turn reasonable non-deep, we can probably recover most of the benefits by decoding “extended” CoTs from the residual stream at each loop turn between tokens. But that’s an area of active development.

cubefox•8m ago
This article is not up-to-date. There have been various benchmarks (some of which published and acknowledged by OpenAI, see the charts in this thread: https://xcancel.com/tomekkorbak/status/2095596839886274689) showing GPT-6 Astra is much less monitorable. The most recent third party benchmark I saw is showing a huge jump in capability for multi-hop reasoning without chain of thought: https://www.lesswrong.com/posts/FsCkkoGsNmPzFKRhg/gpt-6-astr...

I don't think this is explained by the model simply being more capable and therefore achieving more per token: the usage of recurrent depth (Neuralese) is exactly predicting less CoT monitorability even at equal capability.