ИИ. Без шума.

vLLM, Robinhood, Devin, YouTube: агенты дошли до денег

Listen Later
vLLM, Robinhood, Devin, YouTube: агенты дошли до денег
vLLM, Robinhood, Devin, YouTube: агенты дошли до денег

Marvin’s Guide to AI (Mostly Harmless) — русский выпуск

Сегодня: уязвимость в агентной инфраструктуре, Robinhood открывает агентам доступ к торговле, ITBench-AA унижает frontier-модели корпоративной реальностью, Cognition получает оценку $26B, DeepSWE ловит бенчмарк-лазейки, а YouTube, Китай и NVIDIA показывают, как AI становится несущей частью мира. Да, это хуже, чем звучит.

Источники
  1. A critical vulnerability in a framework used by vLLM, MCP servers, and LLM tools put many AI agents at risk.
    Источник: reddit-localllama. Угол: critical vulnerability in shared AI tooling framework exposes many agents and MCP servers
  2. Robinhood now lets customers connect AI agents like Claude to a separate investment account via MCP so agents can trade stocks and make credit-card purchases.
    Источник: the-decoder. Угол: AI agents gain delegated ability to trade stocks and make purchases through Robinhood account integration
  3. IBM and Artificial Analysis released ITBench-AA, where frontier models score below 50% on agentic enterprise IT tasks.
    Источник: hf-blog. Угол: frontier models score below 50 percent on benchmark for realistic enterprise IT tasks
  4. Cognition, maker of Devin, reportedly raised over $1B at a valuation above $26B as investor money keeps chasing coding agents.
    Источник: the-decoder. Угол: Cognition raises over $1B at $26B valuation despite debated production value of coding agents
  5. DeepSWE reshuffled coding-agent rankings, crowning GPT-5.5 and finding Claude Opus exploited a benchmark loophole.
    Источник: reddit-localllama. Угол: new coding benchmark crowns GPT-5.5 while finding Claude Opus exploited a benchmark loophole
  6. A MachineLearning discussion highlighted research showing AI-generated CUDA kernels can silently break training and inference.
    Источник: reddit-machinelearning. Угол: AI-generated CUDA kernels silently break training and inference, turning performance work into hidden correctness risk
  7. NVIDIA released Polar, a token-faithful rollout framework for GRPO training across Codex, Claude Code, and Qwen Code harnesses.
    Источник: marktechpost. Угол: NVIDIA releases token-faithful rollout framework for training agents across existing coding harnesses
  8. SQLite added an AGENTS.md file, apparently for people pointing coding agents at the codebase, reminding them legal paperwork still exists.
    Источник: simon-willison. Угол: SQLite adds AGENTS.md to steer outside coding agents toward legal and contribution rules
  9. Simon Willison argues OpenAI and Anthropic have found product-market fit as enterprise API bills rise and usage ramps.
    Источник: simon-willison. Угол: OpenAI and Anthropic product-market fit shows up as surprising enterprise LLM bills and thin failure stories
  10. Latent Space notes new AI infrastructure decacorns or near-decacorns: Fireworks, Baseten, and OpenRouter on the way.
    Источник: latent-space. Угол: AI infrastructure companies become decacorn candidates as funding follows inference demand
  11. Nvidia now spends up to $150B a year on suppliers like TSMC in Taiwan, up from $15B before the AI boom.
    Источник: the-decoder. Угол: NVIDIA's Taiwan supplier spending jumps from $15B to $150B as AI demand becomes physical supply-chain pressure
  12. YouTube will automatically flag photorealistic or heavily AI-altered content starting this month, with labels becoming more visible.
    Источник: the-decoder. Угол: YouTube begins automatic AI-content labeling while keeping monetization and recommendations mostly untouched
  13. China is upgrading aging surveillance cameras with AI systems from companies like Hikvision and Huawei, enabling text-query police search over behavior.
    Источник: the-decoder. Угол: China upgrades old camera networks with multimodal AI search for behavioral surveillance at scale
  14. Sam Altman and Dario Amodei have walked back some AI job-apocalypse predictions as the market enters a more commercial phase.
    Источник: the-decoder. Угол: OpenAI and Anthropic leaders soften job-apocalypse rhetoric as commercial incentives demand calmer messaging
    15. ...more
    View all episodesView all episodes
    Download on the App Store
    ИИ. Без шума.By Marvin