Episode Details

Today's topics:

Anthropic models pulled by US - The US reportedly forced Anthropic to remove public access to Claude “Fable” and “Mythos” on short notice, raising concerns about ad hoc AI governance, export controls, and competitiveness.

Facebook AI Mode search shift - Meta launched “Facebook AI Mode” in the US, turning search into a conversational AI that synthesizes posts from Groups, Reels, and Marketplace—boosting on-platform discovery while amplifying accuracy and privacy risks.

GitHub dataset for multilingual code - GitHub released a CC0 Multilingual Repositories Dataset with language-ID metadata for READMEs, issues, and PRs, helping researchers improve non-English coverage in AI coding tools and evaluations.

Faster LLM inference with DFlash - LMSYS, Modal, and Z Lab introduced DFlash speculative decoding for Qwen models and made SGLang Spec V2 the default path, aiming to cut latency and boost throughput for large open-weight LLM serving.

Inference engineering becomes mainstream - A new explainer argues ‘inference engineering’ is now a core discipline, highlighting prefill versus decode bottlenecks and why choices like batching, caching, quantization, and speculation shape cost and UX.

Charging AI bots at the edge - AWS WAF Bot Control added AI traffic monetization, using HTTP 402 and stablecoin payments via Coinbase x402 to let publishers charge bots—addressing runaway crawler traffic and content licensing pressure.

LangChain trace analysis and judges - LangChain unveiled LangSmith Engine to cluster real production failures and propose fixes, alongside a fine-tuned Qwen ‘judge’ for perceived-error detection—making agent reliability workflows cheaper and faster.

Sovereign AI and supply chains - An analysis reframes ‘sovereign AI’ around reliable access to compute and infrastructure, emphasizing GPUs, HBM, packaging, power, and equipment across the US, Taiwan, Japan, Europe, and China.

GPU lifespan myth gets questioned - A report challenges the claim that inference GPUs die in one to three years, pointing to weak sourcing and reliability data suggesting much longer physical lifespans under proper operations.

Creators face AI advice disruption - Tim Ferriss says AI chatbots are crushing ‘how-to’ nonfiction demand, signaling an interface shift that could hit books, newsletters, podcasts, and journalism as users prefer instant tailored advice.

AGI to ASI pathways debate - A new arXiv report explores routes from AGI to ASI—scaling, paradigm shifts, self-improvement, and multi-agent collectives—arguing we may see repeated waves of breakthroughs rather than one ‘AGI moment.’

Humans shifting role in software - An essay argues AI will absorb