All Episodes
Interconnects — 151 episodes
Farewell Ai2
Open and closed models are on different exponentials
Some ideas for what comes next, May 2026
Notes from inside China's AI labs
The distillation panic
My bets on open models, mid-2026
The inevitable need for an open model consortium
Claude Mythos and misguided open-weight fearmongering
Gemma 4 and what makes an open model succeed
Lossy self-improvement
GPT 5.4 is a big step for Codex
What comes next with open models
Dean Ball on open models and government control
Olmo Hybrid and future LLM architectures
How much does distillation really matter for Chinese LLMs?
Opus 4.6, Codex 5.3, and the post-benchmark era
Why Nvidia builds open models with Bryan Catanzaro
Thoughts on the job market in the age of LLMs
Arcee AI goes all-in on open models built in the U.S.
Get Good at Agents
Use multiple models
Claude Code Hits Different
Open models: Hot or Not with Nathan Lambert & Florian Brand
New Talk: Building Olmo 3 Think
Olmo 3: America’s truly open reasoning models
Why AI writing is mid
Interview: Ant Group's open model ambitions
5 Thoughts on Kimi K2 Thinking
Burning out
How to scale RL
The State of Open Models
Thoughts on The Curve
ChatGPT: The Agentic App
Thinking, Searching, and Acting
Coding as the epicenter of AI progress and the path to general agents
On China's open source AI trajectory
Ranking the Chinese Open Model Builders
Contra Dwarkesh on Continual Learning
GPT-5 and the arc of progress
gpt-oss: OpenAI validates the open ecosystem (finally)
Towards American Truly Open Models: The ATOM Project
Interviewing Ross Taylor on the state of AI: Chinese open models, scaling reasoning, useful tools, and what comes next
The White House's plan for open models & AI research in the U.S.
Kimi K2 and when "DeepSeek Moments" become normal
The American DeepSeek Project
Some ideas for what comes next (Jun. 2025)
Crafting a good (reasoning) model
The rise of reasoning machines
What comes next with reinforcement learning
How I Write
A taxonomy for next-generation reasoning models
Claude 4 and Anthropic's bet on code
People use AI more than you think
My path into AI
What people get wrong about the leading Chinese open models: Adoption and censorship
State of play of AI progress (and related brakes on an intelligence explosion)
Transparency and (shifting) priority stacks
OpenAI's o3: Over-optimization is back and weirder than ever
OpenAI's GPT-4.1 and separating the API from ChatGPT
Llama 4: Did Meta just push the panic button?
RL backlog: OpenAI's many RLs, clarifying distillation, and latent reasoning
Gemini 2.5 Pro and Google's second chance with AI
Managing frontier model training organizations (or teams)
Gemma 3, OLMo 2 32B, and the growing potential of open-source AI
Interviewing Eugene Vinitsky on self-play for self-driving and what else people do with RL
Elicitation, the simplest way to understand post-training
Where inference-time scaling pushes the market for AI companies
GPT-4.5: "Not a frontier model"?
Character training: Understanding and crafting a language model's personality
Claude 3.7 thonks and what's next for inference-time scaling
Grok 3 and an accelerating AI roadmap
An unexpected RL Renaissance
Deep Research, information vs. insight, and the nature of science
Making the U.S. the home for open-source AI
Why reasoning models will generalize
Interviewing OLMo 2 leads: Open secrets of training language models
DeepSeek R1's recipe to replicate o1 and the future of reasoning LMs
Let me use my local LMs on Meta Ray-Bans
(Voiceover) DeepSeek V3 and the actual cost of training frontier AI models
The state of post-training in 2025
Quick recap on the state of reasoning
(Voiceover) 2024 Interconnects year in review
(Voiceover) OpenAI's o3: The grand finale of AI in 2024
(Voiceover) The AI agent spectrum
(Voiceover) OpenAI's Reinforcement Finetuning and RL for the masses
Interviewing Finbarr Timbers on the "We are So Back" Era of Reinforcement Learning
(Voiceover) OpenAI's o1 using "search" was a PSYOP
(Voiceover) OLMo 2 and building effective teams for training language models
(Voiceover) Tülu 3: The next era in open post-training
(Voiceover) Scaling realities
(Voiceover) Saving the National AI Research Resource & my AI policy outlook
Interviewing Tim Dettmers on open-source AI: Agents, scaling, quantization and what's next
Interviewing Andrew Carr of Cartwheel on the State of Generative AI
(Voiceover) Why I build open language models
(Voiceover) Claude's agentic future and the current state of the frontier models
Interviewing Arvind Narayanan on making sense of AI hype
(Voiceover) Building on evaluation quicksand
Interviewing Andrew Trask on how language models should store (and access) information
How scaling changes model behavior
[Article Voiceover] AI Safety's Crux: Culture vs. Capitalism
Interviewing Riley Goodside on the science of prompting
[Article Voiceover] Llama 3.2 Vision and Molmo: Foundations for the multimodal open-source ecosystem
[Article Voiceover] Reverse engineering OpenAI's o1
Futures of the data foundry business model
A post-training approach to AI regulation with Model Specs
OpenAI's Strawberry, LM self-talk, inference scaling laws, and spending more on inference
OLMoE and the hidden simplicity in training better foundation models
On the current definitions of open-source AI and the state of the data commons
Nous Hermes 3 and exploiting underspecified evaluations
Interviewing Ross Taylor on LLM reasoning, Llama fine-tuning, Galactica, agents
A recipe for frontier model post-training
Interviewing Sebastian Raschka on the state of open LLMs, Llama 3.1, and AI education
GPT-4o-mini changed ChatBotArena
Llama 3.1 405b, Meta's AI strategy, and the new open frontier model ecosystem
SB 1047, AI regulation, and unlikely allies for open models
Switched to Claude 3.5
Interviewing Dean Ball on AI policy: CA SB 1047, upcoming AI disaster response, Llama 3 405B, Chinese open-source AI, and scaling laws
RLHF Roundup: Trying to get good at PPO, charting RLHF's impact, RewardBench retrospective, and a reward model competition
Frontiers in synthetic data
Text-to-video AI is already abundant
AI for the rest of us
A realistic path to robotic foundation models
We aren't running out of training data, we are running out of open training data
Name, image, and AI's likeness
OpenAI chases Her
OpenAI's Model (behavior) Spec, RLHF transparency, and personalization questions
RLHF: A thin line between useful and lobotomized
Phi 3 and Arctic: Outlier LMs are hints
AGI is what you want it to be
Llama 3: Scaling open LLMs to AGI
Stop "reinventing" everything to "solve" alignment
The end of the "best open LLM"
Why we disagree on what open-source AI should be
DBRX: The new best open LLM and Databricks' ML strategy
Evaluations: Trust, performance, and price (bonus, announcing RewardBench)
Model commoditization and product moats
The koan of an open-source LLM
Interviewing Louis Castricato of Synth Labs and Eleuther AI on RLHF, Gemini Drama, DPO, founding Carper AI, preference data, reward models, and everything in between
How to cultivate a high-signal AI feed
Google ships it: Gemma open LLMs and Gemini backlash
10 Sora and Gemini 1.5 follow-ups: code-base in context, deepfakes, pixel-peeping, inference costs, and more
Releases! OpenAI’s Sora for video, Gemini 1.5's infinite context, and a secret Mistral model
Why reward models are still key to understanding alignment
Alignment-as-a-Service: Scale AI vs. the new guys
Open Language Models (OLMos) and the LLM landscape
Model merging lessons in The Waifu Research Department
Local LLMs, some facts some fiction
Multimodal blogging: My AI tools to expand your audience
Multimodal LM roundup: Unified IO 2, inputs and outputs, Gemini, LLaVA-RLHF, and RLHF questions
Where 2024’s “open GPT4” can’t match OpenAI’s
Interviewing Tri Dao and Michael Poli of Together AI on the future of LLM architectures