Security guard, teen and father-son duo beat 37,000 at TRAE's AI contest

A former security guard who says he can't speak English, a high-school junior, and a father-son pair took three of the top prizes at ByteDance's TRAE AI creativity contest in Shanghai — beating roughly 37,000 entrants with apps built almost entirely by describing what they wanted. The grand prize went to "Token Genesis" (词元开物), a team of first-time hardware founders who built an AI workbench plus a desktop circuit-board mill: type "make me a music keyboard," and the system drafts the spec, picks components, writes and flashes the code, routes the PCB, then mills the physical board.

The other winners read like a rebuttal to the idea that you need a CS degree to ship. Runner-up Kyle ("重构者"), a second-year high school student from Zhongshan, Guangdong, built an AI badminton coach after pricing lessons at 360 yuan an hour — his MotionFrame app analyzes swing video on iPhone, pulls motion data from Apple Watch, calls out corrections through AirPods between drills — and it cleared App Store review, reaching No. 22 on the charts ten days after launch. Third place went to Tan Tan, 40, a tech-industry veteran laid off this year, and his middle-school son: their Open Space satellite tracker renders live orbital data on a 3D globe and simulates task planning and debris-avoidance maneuvers. The most viral story belonged to Peng Mingyu, 24, a former security guard whose neck-pain web game made the top 20; singer Hu Yanbin and filmmaker Tim (of 影视飓风) handed him the award in person.

Why it matters: vibe-coding contests are multiplying across China, but this one is a clean snapshot of where the industry is actually heading — the bottleneck has moved from writing code to knowing what to build. The TRAE tool itself evolved from an AI IDE into an agentic assistant that plans tasks, runs tools, edits files and executes commands toward a stated goal, which is exactly why non-programmers could finish real products instead of demos. The prize money is real ($650K total pool), the products are real — an App Store top-30 app, a working satellite mission simulator — and the pattern echoes what we saw when Cloudflare open-sourced a vibe-coding platform for people who aren't coders: software creation is quietly becoming a consumer skill.

What separates the winners is no longer syntax fluency but judgment: Kyle spent days at a badminton facility filming coaches so his model had ground truth; Token Genesis pushed through nearly five straight 14-hour days and 361 commits after real users broke their demo. The hard part is deciding what's worth building and verifying that it works — and that, notably, is still a human job.

What to watch: whether TRAE opens an international season of the contest, and whether any contest-born product holds its App Store ranking past the novelty window.

If AI can take someone from security guard to award-winning developer in two months, what's your excuse for that app idea you keep postponing? 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