frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

AMD acquires Taalas to boost inference performance by etching models in silicon

https://www.theregister.com/systems/2026/08/06/amd-acquires-ai-chip-startup-taalas-to-boost-infer...
296•itvision•3h ago•230 comments

Scientists discover Kelvin-Helmholtz Instability on the surface of the Sun

https://nso.edu/press-release/nsf-inouye-solar-telescope-enables-major-discovery-of-a-hidden-sola...
115•neversaydie•1d ago•19 comments

Welcoming the Nepalese Government to Have I Been Pwned

https://www.troyhunt.com/welcoming-the-nepalese-government-to-have-i-been-pwned/
51•gnabgib•2h ago•12 comments

Mario Meets Pareto

https://www.mayerowitz.io/blog/mario-meets-pareto
843•theanonymousone•12h ago•147 comments

Herdr is joining Y Combinator. The runtime stays open

https://herdr.dev/blog/herdr-is-joining-y-combinator/
122•collinmanderson•4h ago•81 comments

Taste Is All That's Left

https://notashelf.dev/posts/taste-is-all-thats-left
162•tsak•6h ago•121 comments

Inside vLLM: Anatomy of a High-Throughput LLM Inference System (2025)

https://www.aleksagordic.com/blog/vllm
36•sebg•2h ago•2 comments

I stopped trusting USB-C cable labels and started testing them

https://www.makeuseof.com/i-stopped-trusting-usb-c-cable-labels-started-testing-with-meter-instead/
40•baranul•3d ago•26 comments

Show HN: Pokémon Emerald Ported to Raspberry Pi Pico 2

https://github.com/mattdeeds/pokeemerald-rp2350
33•mdeeds•2h ago•13 comments

Launch HN: ProvenMetal (YC S26) delivers circuit boards in days instead of weeks

https://provenmetal.com
180•willcarkner•8h ago•128 comments

Almost no skill required to cook a steak

https://blog.sydorets.com/en/posts/almost-no-skill-required-to-cook-a-steak/
259•yusyd•8h ago•301 comments

GitHub Actions and Pages are experiencing degraded availability

https://www.githubstatus.com/incidents/qcvjkzcs7j74
300•Footkerchief•8h ago•254 comments

Improving GPT‑5.6 Sol in ChatGPT, expanding GPT‑5.6 Luna access for free users

https://openai.com/index/improving-gpt-5-6-sol-in-chatgpt/
132•tedsanders•6h ago•89 comments

My phone detects going on a run as “someone snatching my phone and running off”

https://mastodon.gamedev.place/@rygorous/117047697255584965
74•luu•5h ago•132 comments

STV: A full-motion video codec for the Atari ST

https://medium.com/@jonas.eschenburg/stv-a-video-codec-for-the-atari-st-6e46355c50e4
11•indyjo•1w ago•0 comments

Quake – 30th Anniversary Update

https://slayersclub.bethesda.net/en-US/news/quake-30th-anniversary-update
189•dsubburam•3h ago•87 comments

Can you reverse engineer an ASIC?

https://blog.janestreet.com/can-you-reverse-engineer-an-asic/
50•bschne•4h ago•29 comments

Humans missed 1 in 3 threats approving AI agent commands across 40k game runs

https://scalex.dev/blog/ai-agent-permissions-stats/
240•Wirbelwind•12h ago•186 comments

Learn how chips are made with this Rollercoaster Tycoon-inspired animation

https://laurentiugabriel.github.io/ChipTycoon/
101•laurentiurad•6d ago•17 comments

Show HN: The Channels SDK – Bring Any Agent to Any Channel (Slack, MS Teams)

https://github.com/CopilotKit/channels-sdk
82•davidmckayv•7h ago•20 comments

Crime Pays but Botany Doesn't

https://www.crimepaysbutbotanydoesnt.com/reading-list
644•DarkContinent•19h ago•200 comments

How to Make a Nintendo 64 Game in 2026

https://phoboslab.org/log/2026/08/xibalba64-making-of
453•atan2•2d ago•231 comments

The simple elegance of the integrated timing belt loopback fastener

https://danielmangum.com/posts/integrated-timing-belt-loopback-fastener/
91•hasheddan•4d ago•19 comments

Qwen3.8 Max now ranked as the best overall model by agentic index

https://artificialanalysis.ai/?intelligence=agentic-index
400•apitman•5h ago•258 comments

Spin audit of SQD/QSCI quantum-chemistry benchmarks on iron–sulfur clusters

https://zenodo.org/records/21359923
3•purestatelabs•1h ago•0 comments

Building Progressively Enhanced Forms Using htmx

https://www.rafa.ee/articles/progressive-enhanced-forms-htmx/
50•mpweiher•1w ago•9 comments

Bioengineered chewing gum may offer a way to fight microbes linked to cancers

https://www.sciencedaily.com/releases/2026/08/260803080917.htm
35•Audiophilip•2h ago•1 comments

Pareto Front

https://en.wikipedia.org/wiki/Pareto_front
213•binyu•1w ago•92 comments

Towards Bottom-Up Enumeration in miniKanren via Pruning and Memoization

https://arxiv.org/abs/2607.25373
8•Jimmc414•3d ago•1 comments

Federal Communications Commission scraps limit on broadcast TV ownership

https://www.nbcnews.com/business/media/federal-communications-commission-scraps-limit-broadcast-t...
123•pseudolus•5h ago•91 comments
Open in hackernews

Spin audit of SQD/QSCI quantum-chemistry benchmarks on iron–sulfur clusters

https://zenodo.org/records/21359923
3•purestatelabs•1h ago
Author here.

Background, for anyone who hasn't followed this fight: IBM's iron–sulfur SQD results (Sci. Adv. 2025) are one of the flagship "quantum computers are useful for chemistry now" claims, and a published critique (arXiv:2501.07231) argues the quantum samples never beat classical selected-CI at matched cost. That argument is still live.

Both sides have been arguing about energies. Neither measured which electronic state these calculations actually converge to.

So I measured it. ⟨S²⟩ comes out between 4.7 and 7.0 depending on the system and the subspace size, where the target these papers name is a singlet at ⟨S²⟩ = 0. Every starting guess I tried lands in the same place, and the error at convergence is about the size of the spin-state ladder itself, which is the physics under dispute.

IBM ships a mitigation for this. Run exactly as shipped, using their driver, their recovery loop and their own spin_square() diagnostic, it moves the ground energy by under a nanohartree while making the subspace 4.00x bigger: 194,481 determinants against 48,600. It makes a singlet representable. It never produces one. Their solver also takes a spin_sq argument that would target the singlet directly, and the default pipeline never sets it.

I filed that narrow part on their tracker yesterday, before posting this: https://github.com/Qiskit/qiskit-addon-sqd/issues/337. A maintainer answered and closed it the same day, and his answer is the useful part. The flag, in his words, augments the sampled subspace by using alpha CI strings as beta strings and vice versa. That is a statement about which determinants span the space, not about what spin the returned state comes out in, which is exactly what the measurement says. He did not contest the numbers. The second question, whether the default path is meant to reach spin_sq at all, is still unanswered.

The obvious objection is that this is all my own reimplementation, so here's the part that isn't. IBM's data-availability archive for the flagship paper contains the raw hardware measurement records: 2,457,600 shots on [2Fe-2S] from December 2023, 3,163,742 outcomes on [4Fe-4S] from April 2024, plus the integrals and the optimized circuit's parameters. Running their shots through their own pipeline, [2Fe-2S] reaches low ⟨S²⟩ but sits 248 mHa off their own reference, and [4Fe-4S] converges to a spin-pure triplet, Var(S²) = 3e-6, a genuine S=1 eigenstate: a clean state, and the wrong one, 1,438 mHa from their reference.

Two things in that archive need no analysis from me at all. Their uniform-random null control matches or beats the hardware samples in every published [2Fe-2S] comparison. And their largest [4Fe-4S] runs, at subspace dimension 10⁸, trail their own classical HCI file by 149 mHa.

On the AI angle, since that's half of why this is on HN: Claude did the initial audit end to end in about 72 hours under my direction, and it's been through many rounds of adversarial review since. The part I'd defend as actually interesting isn't the speed. It caught five defects in its own work through pre-registered validation gates, one of them by cross-checking against IBM's own published energy tables, and it retracted its own strongest pro-quantum finding when the new instrument showed that result was a spin-sector artifact. AUDIT_TRAIL.md has the timeline. REPRO_MAP.md maps every claim to a file and a command that regenerates it.

If you want to kill this, and I mean that, here's how. Exhibit any state in a spin-completed ground manifold of these benchmarks with ⟨S²⟩ < 1. Or show me a quantum-sampled subspace at matched determinant count whose spin-identified energy beats HCI or CIPSI. Or get any shipped-pipeline run on IBM's archived samples to land ⟨S²⟩ < 1 and error under 50 mHa at once, with no spin penalty. The harness is in the archive and I'll publish whatever comes back, including if it's me who's wrong.

Preprint: https://doi.org/10.26434/chemrxiv.15006382/v1

The limitations section is real: comparison dimensions are fixed, the [4Fe-4S] reference is approximate, and the 58.6M "samples" file is deduplicated, so the shot-level distribution of their optimal-circuit numerics isn't publicly auditable. It's short. Worth reading before the hot take.