Think tank makes the case for pacing AI self-improvement

The "pace the frontier" debate just got its most concrete policy blueprint yet — and it comes from a think tank arguing the tools should be built before the risks are certain.

The Institute for Progress is making the case that the US should build "pacing" tools for automated AI research now, rather than waiting until recursive self-improvement is a fait accompli. In a guest essay published on Noah Smith's Noahpinion on August 9, IFP fellows Tim Fist and Saif Khan — Khan is a former National Security Council director for technology — lay out the analytical case for deliberately slowing automated AI R&D, and preview 23 specific policy recommendations in a promised part two. Their starting point is the July "Pacing the Frontier" statement signed by 1,367 employees of frontier AI companies, which asked the US government to support international work on pacing tools; the same week, Sam Altman told an interviewer "we may have to pace the rate of AI development."

The authors marshal real numbers to argue the automation of AI R&D is closer than it looks: METR's time-horizon data shows the software-engineering tasks models can complete at 80 percent success have grown from under a minute in 2020 to over three hours today, and Anthropic's internal research found its models now outperform skilled human researchers by roughly 52x on a task about speeding up AI training. METR's simple forecast model puts over 99 percent of AI R&D tasks automated by 2032. Crucially, the essay argues "pacing" need not mean halting progress broadly — the proposal is targeted thresholds on the riskiest automated R&D activities, with talent and compute reallocated toward diffusion and safety. It even names a counterproductive example: the Sanders–Ocasio-Cortez data-center moratorium bill, which would slow research without improving safety.

What makes this notable is that it's not coming from AI doomers or safety purists — IFP is a mainstream, pro-innovation policy shop, and its case is built on the labs' own published data. That moves the debate from "should we pace?" to "how would pacing actually work," which is where the real policy fight is about to happen.

OpenChamber, an open-source agentic development environment built on the OpenCode SDK, hit the Hacker News front page today with a cross-platform pitch: the same agent workspace on desktop, browser, phone, and inside VS Code. It runs agents toward persistent "session goals" even with the app closed, can execute one task across up to five models and fuse the best results, supports scheduled cron-style runs, and walks users through large diffs step by step. The project, which has been in development since late 2025, has picked up roughly 7,800 GitHub stars. Its privacy model is a differentiator: sessions stay on-device, remote access is gated behind a UI password or an end-to-end encrypted relay, and the code is open for inspection.

What to watch: IFP's part two with the 23 policy ideas, and whether any of them find their way into actual legislation.

Slowing AI research on purpose sounds radical — but if the labs themselves are asking for the tools, who should build them? Tell us in the comments.

Sources: OpenChamber (GitHub)

Houmo moves compute inside memory — and A20 enters production

Beijing chipmaker Houmo used its partner conference this week to lay out two compute-in-memory routes at once — and to say the first chip built on the newer one is already in production. Houmo's A20, its first chip built on a 3D compute-in-memory architecture, has entered production. The part supports up to 16 chips fully interconnected, claims a 50% improvement in compute utilization over the previous generation, and covers models from 3B to 30B parameters, according to co-founder and product

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kuaishou's Kling AI has spent two years becoming the video model people actually finish projects with. Its new flagship, Kling 4.0, went fully live on October 15 — and this round is less about prettier frames than about the two things that make a generated shot usable in a real edit: length and control. Kling 4.0 generates up to 30 seconds in a single native pass, outputs 10-bit HDR at 4K, 1080p and 21:9, and takes up to 15 reference assets in one task. That budget splits across ten images, fi

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Microsoft finished pre-training its biggest in-house model yet — then used it to make a point about the company it bankrolls. Microsoft AI completed pre-training of MAI-Base-2, a mixture-of-experts foundation model with 80 billion active and 2.8 trillion total parameters, and says the recipe is 4.2–12.3x more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base. The run itself is the number worth keeping: 16,000 Nvidia GB200 GPUs for 38 days across August and September, at 21% average model

Read more

Houmo moves compute inside memory — and A20 enters production

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Brain organoids classified Braille at 83% — in a trio, not alone

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute