Evoke keeps a generated world's memory outside the video model
Alaya Lab released Evoke, a 14-billion-parameter open-weights world model that stores scene geometry in an external bank indexed by camera position instead of inside the image generator, letting an interactive generated world stay consistent for hours.
A robot that learns a new task from a twelve-second demo
Generalist AI's GEN-1.5 robot foundation model can attempt an unfamiliar physical task after watching a single 3-to-12-second demonstration, with no fine-tuning at all, succeeding 59 percent of the time across ten tasks.
Businesses are not buying Anthropic's best model
Ramp's spend data shows Claude Fable 5 -- the highest-scoring model on the market -- accounted for just 6 percent of the tokens businesses bought from Anthropic in July and 11.4 percent of the dollars, while Anthropic's overall business adoption kept growing.
OpenAI says open models will enable persistent cyber-attacks
OpenAI's chief global affairs officer Chris Lehane told the Guardian that freely downloadable models only months behind frontier systems will let attackers run continuous automated campaigns, and called for a U.S. law making pre-release safety proof mandatory.
A humanoid robot ran 100 meters in 9.39 seconds in Beijing
At the second World Humanoid Robot Games on August 22, a robot from the Tianzhuo team won a fully autonomous 100-meter preliminary in 9.39 seconds, a clock time below Usain Bolt's 9.58-second human world record.
Gallup is testing AI agents that answer surveys for real people
Gallup has built AI agents from in-depth interviews with about 1,000 of its panel members and is independently validating whether their simulated answers can stand in for human ones, while its partner Simile raised over $200 million at a $2 billion valuation.
Ornith-1.5 writes its own training problems and grades them
Ornith released an open-weight model family whose training loop generates its own tasks, builds its own scoring harnesses, and feeds the reward back into all three stages -- with the 397-billion-parameter flagship matching Claude Opus 4.8 on agentic coding benchmarks.
Self-hosting a frontier open model now takes a whole node
Engineers at the Belgian research institute imec benchmarked open-weight coding models against commercial APIs and found the best open model needs an eight-GPU B300 node to run at all, serving only 16 concurrent sessions at roughly eight times slower than a hosted frontier agent.
Iran-linked hackers took a UK power plant offline for four days
The UK government confirmed that a cyber-attack blamed on hackers linked to Iran shut down a small-scale energy generator for four days last month, the first publicly acknowledged British power generation outage caused by an intrusion.
Alibaba is raising 80 billion Hong Kong dollars purely for AI
Alibaba proposed placing newly issued shares with non-U.S. investors for HK$80 billion, saying it will put 100 percent of the net proceeds into its full-stack AI capabilities including AI infrastructure.
SenseNova generates images in pixels with no VAE in the middle
SenseTime released SenseNova-U1.5-8B-MoT under Apache 2.0, an open-weight model that understands and generates images without the vision encoder and latent autoencoder nearly every other system relies on, working directly on pixels instead.
Models can train each other without a single correct answer
A method called Co-RL trains language models with no labels at all by rewarding each model for agreeing with a different model's majority vote, matching and sometimes beating the same recipe trained with ground-truth answers.