Robots set sprint records in Beijing — then burst into flames

The second World Humanoid Robot Games closed out its Beijing run with viral sprints, sparks, and an uncomfortable reality check about what humanoids can actually do. Meanwhile, X reached for lawyers instead of firewalls to kill the internet's favorite ad-free reading front-end.

Humanoid robots beat human world records in the 100-meter sprint and standing high jump in Beijing this week — and then couldn't stop, slow down, or survive the aftermath. At the second World Humanoid Robot Games (August 22–26), machines went from trailing human record-holders last year to beating them, but the viral clips tell the fuller story: sprinters plowing into barriers because their software has no concept of decelerating, one robot breaking apart at the waist in a shower of sparks, another bursting into flames mid-race. Dipam Patel, a PhD candidate at Purdue University working with the US Army Research Lab, told Ars Technica the jumps prove real gains in whole-body control — yet these are single-purpose machines engineered for one finish line, many events allowed human teleoperation (with score penalties), and even the autonomous entries leaned on cloud computing over 5G links rather than onboard compute. The new practical events were more damning than any crash: only three of twelve teams completed the timed firefighting-and-rescue challenge, and the laundry-washing, hotel-housekeeping, and library-shelving heats ran well below human working speed. We covered the spectacle side when Honor's Lightning robot beats Bolt's 100m world record — the honest takeaway now is that the boring chores, not the races, are the benchmark that matters. Investors poured more than $6 billion into humanoid companies in 2025 alone, and reliability across ten runs — not one glorious heat — will decide whether that money was well placed.

X sent cease-and-desist letters to Nitter, the open-source project behind ad-free X readers, demanding its instances and code repository be taken down permanently. The letters went out August 24, citing the Texas Harmful Access by Computer Act and the Lanham Act while claiming evidence of API circumvention and scraped account data; nitter.net went dark within hours and its maintainer, the developer known as Zedeus, says development has stopped while he seeks legal advice. Unlike the 2024 crackdown, when API restrictions killed the flagship instance only for the network to return by wiring instances to real accounts, X is now pursuing the project through legal threats with a shutdown deadline of late afternoon Eastern time today. For anyone following AI, the stakes reach beyond one front-end: tools like Nitter are how researchers and agents pull public posts without logging in, and every such door closing narrows the openly readable web that models and agents depend on.

What to watch: X's August 25 deadline has now passed — watch whether the remaining Nitter instances comply quietly and whether X escalates to forcing a GitHub repository takedown.

Should platforms sell affordable read-only access to public posts instead of litigating every scraper? Tell us in the comments.

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

Amazon's new Bee devices listen all day — for $11.99 a month

Amazon turned its always-listening Bee line into a two-device family on Thursday, and NVIDIA opened a 2.8-trillion-token Indic corpus recipe. The connecting thread: both are bets that the next AI money is in memory — yours, or your model's. Amazon put two new always-listening devices on sale, and both now lead to a subscription. Bee Frontier, the next-generation wristband that also clips to clothing, ships now at $74.99 in Onyx, Lavender and Sage — up from the $50 predecessor — with seven-day

Perplexity open-sources its agent sandbox after 118 failed escapes

Perplexity handed its agent containment layer to the open-source world, and Washington's AI task force stopped reading briefings and started working. Perplexity open-sourced SPACE on Thursday — the sandbox it built to keep its own AI agents from getting out — and says nine frontier models failed to break out of it across 118 attempts. Each agent runs inside a small virtual machine with its own operating system, with credentials and network access managed outside the sandbox so an agent can re

Read more

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

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

Amazon's new Bee devices listen all day — for $11.99 a month

Perplexity open-sources its agent sandbox after 118 failed escapes