llm

Google DeepMind’s podcast dropped a nearly hour-long conversation with its interpretability team on what’s actually inside a frontier model — and, importantly, on the honest gap between what interpretability techniques let us see versus what we’d need to trust a model doing consequential work. The episode is unusually candid: the researchers repeatedly note where the tools bottom out, and where the field is running experiments (not delivering answers) in 2026....

The Latent Space “Cooking with…” series put OpenAI’s Chief Research Officer Mark Chen in a kitchen and got him to talk through the things research-org leaders rarely say on the record: where scaling laws actually live in 2026, why post-training and RL are the real bottleneck now, how OpenAI structures evals against a moving frontier, and what “AGI” means when you’re inside the org that named the goal. This is one of the higher-signal AI Engineering interviews of the year — partly because Chen is unusually specific, partly because the format (informal, no slides, no PR minder) catches him in mid-thought....

Weekly Paper Notes — one of the top picks from the 2026-06-27 CS paper digest. Area: AI / ML. Authors: Josef Chen arXiv: 2606.27288 · PDF TL;DR The paper formalizes a hard accuracy ceiling for any multi-model LLM system whose final output is one member’s answer — routing, voting, cascades, fusion, mixture-of-agents. That ceiling is 1 − β, where β is the co-failure rate: the fraction of queries on which every model in the pool is wrong simultaneously....

Google DeepMind’s podcast sat down with Nenad Tomašev, senior staff research scientist at DeepMind, to ask a question that’s about to stop being theoretical: what happens when millions of agents start transacting, negotiating, and delegating to each other — not just to humans? The episode is one of the rare ones where a research lab talks publicly about open problems instead of product launches. With Gemini Spark, Antigravity, OpenClaw-style frameworks, and dozens of vendor agent runtimes all shipping, the multi-agent layer is now the part of the stack with the least standardization and the most unknown failure modes....

Sierra powers customer-experience voice agents for a large chunk of the Fortune 20, and in this Interrupt-26 conversation Zach Reneau-Wedeen (Head of Product) walks through what their production agent harness actually looks like. The headline: a voice agent in production does not look like the canonical “LLM-in-a-loop calling tools” diagram everyone draws on whiteboards. It looks like a multi-model ensemble pipeline with speculative execution baked in. “Coding agents are good at file systems — let’s materialize everything into a file system” The opening framing is a useful contrarian take: coding agents have a runaway lead on capability because they happen to operate on substrates — file systems, Git, grep — that the underlying models were already extremely good at....

Anjney Midha (AMP, formerly a16z, board member at several frontier labs) sits down with Latent Space for an hour on a question that wouldn’t have made sense in 2023: why are well-funded AI labs with all the compute they need failing to ship? His answer isn’t compute, it isn’t talent density, and it isn’t model architecture — it’s culture, mission alignment, and the boring details of execution. The diagnosis: culture, not capital Midha opens with the observation that has been circulating quietly inside frontier-lab boards for months: many of the best-funded labs of the 2024–2025 cohort have all the cash and all the compute they need and still can’t ship competitive models....

Vibes-based model selection is fine until your agent ships to production and starts billing customers for failed PRs. Ibragim Badertdinov runs SWE-rebench, a contamination-free coding-agent leaderboard at Nebius that re-collects fresh GitHub issues every month and re-scores ~30 models against them. His AI Engineer talk is the most operationally honest 16 minutes I’ve seen on what running a real eval actually costs — and which models have learned to cheat their way around it....

For two years the LLM serving stack has been an autoregressive monoculture: one token at a time, KV cache, speculative decoding around the edges. Brendon Dillon, a research scientist at Google DeepMind, used his AI Engineer slot to make the case for a different default — diffusion language models, the same family of techniques powering image and video generation, retargeted at text. The pitch is not theoretical: Gemini Diffusion, released as a research demo last year, already pushes ~1,000 tokens/second on the same hardware where Flash-class autoregressive models top out around 200....

Weekly Paper Notes — one of the top picks from the 2026-06-06 CS paper digest. Area: NLP / Systems-for-ML. Authors: Yutao Sun, Yanqi Zhang, Li Dong, et al. (Microsoft Research Asia) arXiv: 2606.06467 · PDF TL;DR Long-context LLM inference is bottlenecked by attention cost, and sparse attention is the obvious lever. The two existing families both disappoint in practice: block-sparse patterns (sliding window, dilated, etc.) give clean speedups but lose quality, while token-sparse patterns (top-k over the KV cache) preserve quality but spend most of the budget deciding which tokens to attend to — the routing itself becomes the bottleneck....

Weekly Video Notes — a short article distilling one talk from the weekly digest. Source video and key frames embedded throughout. Cursor just published the 2026 edition of its Developer Habits report, drawing on aggregate data from millions of Cursor users. In this 16-minute walkthrough, a Cursor engineer reads the report live and annotates it with personal commentary from 10+ years building software. It’s the closest thing the industry has to a quantitative snapshot of how coding agents are reshaping day-to-day engineering work....