beat · 183 stories
FlagOS is a BAAI-led (Beijing Academy of AI) project that ported Alibaba's 2.4-trillion-parameter Qwen3.8 to nine chip families, including eight Chinese-designed accelerators, in under 24 hours.
The world's largest PC maker reported $26.9 billion in revenue and 176% profit growth as the data-center hardware that runs AI services nearly doubled.
A new framework called SFS-DPO (Self-Fix Step-DPO) splits step-level reasoning from self-verification, with reported gains on math and code benchmarks over prior step-level training methods.
Model Context Protocol (MCP) feeds every AI agent a tool list on each call. Okta's new filter trims that list before the model sees it, dropping token cost by up to 90% in vendor modeling and turning the prompt into an access gate.
Foxconn parent Hon Hai's H1 2026 spend is already up NT$3.7B (New Taiwan dollars, roughly USD 127M) year-on-year as it prepares to mass-produce next-generation AI racks in Q3 2026 and expand US campuses in Texas, Wisconsin, Ohio, and California.
A new arXiv paper proves that cooperative AI systems, teams of agents sharing one goal, can collectively pick worse moves than any in their shared playbook.
ETIAS, the EU's pre-arrival screening system, weighs age, sex, nationality, residence, education, and occupation. Three of the six are proxies for protected characteristics the system is formally barred from using.
A Fudan University team placed a one-atom-thick carbon layer in front of a flash-memory "floating gate" so a single trapped electron can be read at room temperature, clearing a 1990s-era barrier.
Cisco joined Anthropic's Glasswing partner program, which puts a vulnerability-discovery model called Mythos in the hands of critical-software maintainers, and used its Q4 earnings to argue that AI has shortened the safe-use window for
A 4-byte floating-point load leaks stale L1 cache data into the CPU's working registers at hundreds of MB/s, and there is no software patch.
Embodied AI — putting machine-learning models inside physical robots — is usually framed as a model problem.
A 100-code-change test across Python, Java, and C++ shows why state-of-the-art tools that try to pin down where a bug entered the code still need human auditors: the code changes that introduced the flaws run about six times larger than the fixes