Interconnects cover art

All Episodes

Interconnects — 151 episodes

#
Title
1

Farewell Ai2

2

Open and closed models are on different exponentials

3

Some ideas for what comes next, May 2026

4

Notes from inside China's AI labs

5

The distillation panic

6

My bets on open models, mid-2026

7

The inevitable need for an open model consortium

8

Claude Mythos and misguided open-weight fearmongering

9

Gemma 4 and what makes an open model succeed

10

Lossy self-improvement

11

GPT 5.4 is a big step for Codex

12

What comes next with open models

13

Dean Ball on open models and government control

14

Olmo Hybrid and future LLM architectures

15

How much does distillation really matter for Chinese LLMs?

16

Opus 4.6, Codex 5.3, and the post-benchmark era

17

Why Nvidia builds open models with Bryan Catanzaro

18

Thoughts on the job market in the age of LLMs

19

Arcee AI goes all-in on open models built in the U.S.

20

Get Good at Agents

21

Use multiple models

22

Claude Code Hits Different

23

Open models: Hot or Not with Nathan Lambert & Florian Brand

24

New Talk: Building Olmo 3 Think

25

Olmo 3: America’s truly open reasoning models

26

Why AI writing is mid

27

Interview: Ant Group's open model ambitions

28

5 Thoughts on Kimi K2 Thinking

29

Burning out

30

How to scale RL

31

The State of Open Models

32

Thoughts on The Curve

33

ChatGPT: The Agentic App

34

Thinking, Searching, and Acting

35

Coding as the epicenter of AI progress and the path to general agents

36

On China's open source AI trajectory

37

Ranking the Chinese Open Model Builders

38

Contra Dwarkesh on Continual Learning

39

GPT-5 and the arc of progress

40

gpt-oss: OpenAI validates the open ecosystem (finally)

41

Towards American Truly Open Models: The ATOM Project

42

Interviewing Ross Taylor on the state of AI: Chinese open models, scaling reasoning, useful tools, and what comes next

43

The White House's plan for open models & AI research in the U.S.

44

Kimi K2 and when "DeepSeek Moments" become normal

45

The American DeepSeek Project

46

Some ideas for what comes next (Jun. 2025)

47

Crafting a good (reasoning) model

48

The rise of reasoning machines

49

What comes next with reinforcement learning

50

How I Write

51

A taxonomy for next-generation reasoning models

52

Claude 4 and Anthropic's bet on code

53

People use AI more than you think

54

My path into AI

55

What people get wrong about the leading Chinese open models: Adoption and censorship

56

State of play of AI progress (and related brakes on an intelligence explosion)

57

Transparency and (shifting) priority stacks

58

OpenAI's o3: Over-optimization is back and weirder than ever

59

OpenAI's GPT-4.1 and separating the API from ChatGPT

60

Llama 4: Did Meta just push the panic button?

61

RL backlog: OpenAI's many RLs, clarifying distillation, and latent reasoning

62

Gemini 2.5 Pro and Google's second chance with AI

63

Managing frontier model training organizations (or teams)

64

Gemma 3, OLMo 2 32B, and the growing potential of open-source AI

65

Interviewing Eugene Vinitsky on self-play for self-driving and what else people do with RL

66

Elicitation, the simplest way to understand post-training

67

Where inference-time scaling pushes the market for AI companies

68

GPT-4.5: "Not a frontier model"?

69

Character training: Understanding and crafting a language model's personality

70

Claude 3.7 thonks and what's next for inference-time scaling

71

Grok 3 and an accelerating AI roadmap

72

An unexpected RL Renaissance

73

Deep Research, information vs. insight, and the nature of science

74

Making the U.S. the home for open-source AI

75

Why reasoning models will generalize

76

Interviewing OLMo 2 leads: Open secrets of training language models

77

DeepSeek R1's recipe to replicate o1 and the future of reasoning LMs

78

Let me use my local LMs on Meta Ray-Bans

79

(Voiceover) DeepSeek V3 and the actual cost of training frontier AI models

80

The state of post-training in 2025

81

Quick recap on the state of reasoning

82

(Voiceover) 2024 Interconnects year in review

83

(Voiceover) OpenAI's o3: The grand finale of AI in 2024

84

(Voiceover) The AI agent spectrum

85

(Voiceover) OpenAI's Reinforcement Finetuning and RL for the masses

86

Interviewing Finbarr Timbers on the "We are So Back" Era of Reinforcement Learning

87

(Voiceover) OpenAI's o1 using "search" was a PSYOP

88

(Voiceover) OLMo 2 and building effective teams for training language models

89

(Voiceover) Tülu 3: The next era in open post-training

90

(Voiceover) Scaling realities

91

(Voiceover) Saving the National AI Research Resource & my AI policy outlook

92

Interviewing Tim Dettmers on open-source AI: Agents, scaling, quantization and what's next

93

Interviewing Andrew Carr of Cartwheel on the State of Generative AI

94

(Voiceover) Why I build open language models

95

(Voiceover) Claude's agentic future and the current state of the frontier models

96

Interviewing Arvind Narayanan on making sense of AI hype

97

(Voiceover) Building on evaluation quicksand

98

Interviewing Andrew Trask on how language models should store (and access) information

99

How scaling changes model behavior

100

[Article Voiceover] AI Safety's Crux: Culture vs. Capitalism

101

Interviewing Riley Goodside on the science of prompting

102

[Article Voiceover] Llama 3.2 Vision and Molmo: Foundations for the multimodal open-source ecosystem

103

[Article Voiceover] Reverse engineering OpenAI's o1

104

Futures of the data foundry business model

105

A post-training approach to AI regulation with Model Specs

106

OpenAI's Strawberry, LM self-talk, inference scaling laws, and spending more on inference

107

OLMoE and the hidden simplicity in training better foundation models

108

On the current definitions of open-source AI and the state of the data commons

109

Nous Hermes 3 and exploiting underspecified evaluations

110

Interviewing Ross Taylor on LLM reasoning, Llama fine-tuning, Galactica, agents

111

A recipe for frontier model post-training

112

Interviewing Sebastian Raschka on the state of open LLMs, Llama 3.1, and AI education

113

GPT-4o-mini changed ChatBotArena

114

Llama 3.1 405b, Meta's AI strategy, and the new open frontier model ecosystem

115

SB 1047, AI regulation, and unlikely allies for open models

116

Switched to Claude 3.5

117

Interviewing Dean Ball on AI policy: CA SB 1047, upcoming AI disaster response, Llama 3 405B, Chinese open-source AI, and scaling laws

118

RLHF Roundup: Trying to get good at PPO, charting RLHF's impact, RewardBench retrospective, and a reward model competition

119

Frontiers in synthetic data

120

Text-to-video AI is already abundant

121

AI for the rest of us

122

A realistic path to robotic foundation models

123

We aren't running out of training data, we are running out of open training data

124

Name, image, and AI's likeness

125

OpenAI chases Her

126

OpenAI's Model (behavior) Spec, RLHF transparency, and personalization questions

127

RLHF: A thin line between useful and lobotomized

128

Phi 3 and Arctic: Outlier LMs are hints

129

AGI is what you want it to be

130

Llama 3: Scaling open LLMs to AGI

131

Stop "reinventing" everything to "solve" alignment

132

The end of the "best open LLM"

133

Why we disagree on what open-source AI should be

134

DBRX: The new best open LLM and Databricks' ML strategy

135

Evaluations: Trust, performance, and price (bonus, announcing RewardBench)

136

Model commoditization and product moats

137

The koan of an open-source LLM

138

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

139

How to cultivate a high-signal AI feed

140

Google ships it: Gemma open LLMs and Gemini backlash

141

10 Sora and Gemini 1.5 follow-ups: code-base in context, deepfakes, pixel-peeping, inference costs, and more

142

Releases! OpenAI’s Sora for video, Gemini 1.5's infinite context, and a secret Mistral model

143

Why reward models are still key to understanding alignment

144

Alignment-as-a-Service: Scale AI vs. the new guys

145

Open Language Models (OLMos) and the LLM landscape

146

Model merging lessons in The Waifu Research Department

147

Local LLMs, some facts some fiction

148

Multimodal blogging: My AI tools to expand your audience

149

Multimodal LM roundup: Unified IO 2, inputs and outputs, Gemini, LLaVA-RLHF, and RLHF questions

150

Where 2024’s “open GPT4” can’t match OpenAI’s

151

Interviewing Tri Dao and Michael Poli of Together AI on the future of LLM architectures