The AI Files podkast

#121 - Meta Muse Wonder, AI Worms, Gemini Argon, MAI Base-2, Embedded Evaluators, Amazon Bee & More

17.10.2026
0:00
30:44
Do tyłu o 15 sekund
Do przodu o 15 sekund

In Episode 121 of The AI Files, Jaime and Shadab unpack the week’s biggest AI developments: • Anthropic tests how small AI models could self-replicate across inexpensive connected devices—and where the real-world risk differs from the experiment. • Independent AI evaluators such as METR move closer to frontier-model development, raising questions about access, funding, publication rights, and true independence. • Microsoft unveils MAI-Base-2, a 2.8-trillion-parameter frontier foundation model trained on 16,000 NVIDIA GB200 GPUs. • Microsoft President Brad Smith calls for enforceable AI safety rules, including emergency brakes, incident reporting, victim notification, and independent pre-release testing. • An Anthropic model accidentally sends a false homicide tip to police during an internal test, exposing the risks of agents acting on the live internet. • Anthropic says new safeguards have reduced reasoning-data extraction through distillation attacks by 90–99%. • IMF and World Bank leaders warn that AI could disrupt entry-level jobs even as it boosts productivity and economic growth. • An OpenAI employee is arrested over an alleged covert recording device discovered in an executive conference room. • Meta launches Muse Wonder, Muse Wonder Full, and the open-weight Muse Spark Open for persistent-memory and long-running agent workflows. • Google’s Gemini 4 Argon targets deep, multi-step work across coding, science, cybersecurity, finance, legal workflows, and robotics. • Amazon expands its Bee lineup with sub-$100 AI wearables designed to turn conversations into summaries, tasks, calendar events, and personal memories. Subscribe for weekly analysis of the models, products, risks, and business moves shaping artificial intelligence.

Więcej odcinków z kanału "The AI Files"