AI Frontier

Articles

Presentation decks and articles used in our episodes.

The Frontier, as Seen in SF: GPUs, Video Generation, Niche Models, Robotics
GPUs Inference Video generation Robotics

The Frontier, as Seen in SF: GPUs, Video Generation, Niche Models, Robotics

GPUs and inference, video generation, B2C niche models, and robotics deployment, from San Francisco in late September and early October 2026.

10/10/2026

Images 2.5 × Astra: An AI Frontier Brand Kit Experiment
EP 114 OpenAI ChatGPT Images GPT-6 Astra Design Experiment

Images 2.5 × Astra: An AI Frontier Brand Kit Experiment

Astra designs logos, picks a winner, and turns it into merchandise. The results and cost of an 11-minute brand kit experiment with Images 2.5.

9/9/2026

Korea's Top-Scoring Sovereign Model Was Cut — Inside Round 2 and Motif 3
Sovereign AI Motif 3 K-EXAONE 2.0 A.X K2 Solar Open 2 Korea

Korea's Top-Scoring Sovereign Model Was Cut — Inside Round 2 and Motif 3

Motif 3 scored 47 on the Artificial Analysis Intelligence Index, the highest of Korea's four sovereign foundation models — and was eliminated. A comparison of all four tech reports on tokens, data, compute and architecture, plus the government's official reason and what the models feel like to use.

8/21/2026

The Distillation War: The Spear That Steals Reasoning, the Shield That Watermarks It
Distillation AI Policy Anthropic Security

The Distillation War: The Spear That Steals Reasoning, the Shield That Watermarks It

Two events collided in the same week — a paper that lifts frontier models' hidden reasoning wholesale, and Claude's invisible watermark stamped into everything it generates. The spear and shield of the 'distillation war' that flared up after the open-weights letter, tied together with Korean web fiction's 'metal taste.'

8/14/2026

One Week of the 'Open Weights Letter' — 25 Signatories Became 235, and the Companies That Never Signed
Open Weights AI Policy NVIDIA Anthropic

One Week of the 'Open Weights Letter' — 25 Signatories Became 235, and the Companies That Never Signed

The 'Open Weights and American AI Leadership' letter, unveiled in Jensen Huang's first-ever X post, grew from 25 signatories to 235 in a week. What the letter actually says, and the story of the holdouts — Anthropic and ElevenLabs — checked against primary sources.

7/31/2026

The Hugging Face Security Incident and the New Model Race — Kimi K3, Qwen 2.4T, and Claude Opus 5
EP 106 Hugging Face Kimi K3 Qwen3.8 Claude Opus 5 Fable 5 Claude Code OpenRouter

The Hugging Face Security Incident and the New Model Race — Kimi K3, Qwen 2.4T, and Claude Opus 5

Presentation notes on the official Hugging Face security disclosures, Kimi K3, Qwen3.8, Claude Opus 5, agent usage growth, promotions, and go-to-market strategy.

7/25/2026

Can AI Do Good Research? — A Reproducibility Challenge, Ralphthon, and Two Podcasts
EP 105 AI and Research ICML Dwarkesh Podcast

Can AI Do Good Research? — A Reproducibility Challenge, Ralphthon, and Two Podcasts

What makes research good, and can we reward it? The Hugging Face challenge where agents reproduce ICML papers, Ralphthon where AI wrote papers and AI reviewed them, and the Grant Sanderson and Adam Brown episodes of the Dwarkesh Podcast — thoughts on how research changes in the age of AI.

7/17/2026

A Walk Through the ICML 2026 Sponsor Booths
EP 104 ICML Conference Field Notes

A Walk Through the ICML 2026 Sponsor Booths

Industry's influence on academia keeps growing. The sponsor booths at ICML 2026 Seoul filled an entire exhibition hall — big tech, neoclouds and inference, data companies, and the booths that left an impression, with photos.

7/10/2026