Jan 26, 2024 · 1h 20m · latent-space

The Four Wars of the AI Stack - Dec 2023 Recap

0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

In this episode of the Latent Space podcast, co-hosts Alessio and Swix introduce 'The Four Wars of the AI Stack' framework to dissect the most critical battlegrounds shaping artificial intelligence entering 2024. They deliver deep-dive analyses across data provenance, GPU inference pricing, multimodal commercialization, and developer tooling while reviewing key industry breakthroughs from December 2023.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →

The hosts as informed peer 6.8 Guest teaching 4.3 Guest disagreement 1.9 The hosts pushing back 2.0
05100:0020:0040:001:00:001:20:000:01–3:36 · The hosts as informed peer 6/10 Introducing the Monthly Recap: The Four Wars Framework Alessio and Swix smoothly introduce their monthly recap format. Alessio breaks down the four wars structure and upcoming guest lineup while Swix adds context on their framing. Collaborative and conversational peer dynamic.3:36–8:04 · The hosts as informed peer 6/10 The Non-Wars: AI Trends That Cooled Off in 2023 Swix explains why open source AI and agents cooled off, while Alessio provides developer perspective on why code model experimentation is fragmented across tools like VS Code and Cursor. Solid, respectful domain discussion.8:04–10:10 · The hosts as informed peer 8/10 The End of Low Background Tokens and Data Provenance Alessio takes the lead explaining the concept of low background tokens using Jeff Huber's steel analogy after nuclear testing. He details how synthetic data pollution will force data engineering and first-party datasets like Databricks Dolly.10:10–17:54 · The hosts as informed peer 7/10 War 1: Copyright Holders vs. AI Labs and Walled Gardens Alessio shares historical background on data scraping from his Rap Genius moderation days when Google scraped lyrics, while Swix analyzes fair use criteria in the New York Times OpenAI lawsuit and closed gardens of Reddit and Stack Overflow.17:54–21:41 · The hosts as informed peer 5/10 The Synthetic Data Frontier and Distribution Spiking Swix recaps key takeaways from NeurIPS on synthetic data scale versus fine-tuning synthetic data, citing conversations with Louis Castricado and Andre Karpathy's highlighted paper. Alessio engages on the risks of mode collapse in synthetic data generation.21:41–31:23 · The hosts as informed peer 7/10 War 2: The GPU Inference Price War and Unit Economics Swix details the unit economics and breakeven calculations of Mixtral token pricing while Alessio compares inference benchmarking drama and loss leaders to traditional database benchmarks and ride-sharing economics.31:23–41:33 · The hosts as informed peer 6/10 MoE Hardware Implications and Swix's 'Sour Lesson' Swix explains hardware implications of MoE models and expounds his 'sour lesson' regarding state space models and why machine intelligence doesn't mirror human cognition. Alessio offers insight on hardware memory bandwidth hurdles.41:33–53:57 · The hosts as informed peer 7/10 War 3: The Multimodal Battle and Commercial Monetization Alessio and Swix debate whether DALL-E 3 got worse due to chat integrations (Swix playfully calling it a 'skill issue') and explore go-to-market monetization paths for Midjourney and ElevenLabs voice models.53:57–1:05:07 · The hosts as informed peer 8/10 War 4: RAG Ops, Vector Databases and Platform Gravity Alessio brings deep venture and devtools knowledge analyzing how vector DBs compete with frameworks, drawing historical parallels to Jamstack, Vercel, and Gatsby. Swix provides insider Netlify context.1:05:07–1:11:55 · The hosts as informed peer 8/10 Semantic Programming and the Limits of Coding Agents Alessio outlines his framework for semantic programming and the engineering hazards of coding agents generating technical debt, while Swix pushes back with his inner versus outer developer loop paradigm.1:11:55–1:20:34 · The hosts as informed peer 7/10 December Highlights: Prompt Hacks, Gemini, and AI Hardware Alessio critiques Anthropic's needle-in-a-haystack prompt hacks and discusses hardware context appliances like Tab and Rabbit R1 alongside privacy trade-offs. Both wrap up the recap episode.0:01–3:36 · Guest teaching 3/10 Introducing the Monthly Recap: The Four Wars Framework Alessio and Swix smoothly introduce their monthly recap format. Alessio breaks down the four wars structure and upcoming guest lineup while Swix adds context on their framing. Collaborative and conversational peer dynamic.3:36–8:04 · Guest teaching 4/10 The Non-Wars: AI Trends That Cooled Off in 2023 Swix explains why open source AI and agents cooled off, while Alessio provides developer perspective on why code model experimentation is fragmented across tools like VS Code and Cursor. Solid, respectful domain discussion.8:04–10:10 · Guest teaching 2/10 The End of Low Background Tokens and Data Provenance Alessio takes the lead explaining the concept of low background tokens using Jeff Huber's steel analogy after nuclear testing. He details how synthetic data pollution will force data engineering and first-party datasets like Databricks Dolly.10:10–17:54 · Guest teaching 3/10 War 1: Copyright Holders vs. AI Labs and Walled Gardens Alessio shares historical background on data scraping from his Rap Genius moderation days when Google scraped lyrics, while Swix analyzes fair use criteria in the New York Times OpenAI lawsuit and closed gardens of Reddit and Stack Overflow.17:54–21:41 · Guest teaching 6/10 The Synthetic Data Frontier and Distribution Spiking Swix recaps key takeaways from NeurIPS on synthetic data scale versus fine-tuning synthetic data, citing conversations with Louis Castricado and Andre Karpathy's highlighted paper. Alessio engages on the risks of mode collapse in synthetic data generation.21:41–31:23 · Guest teaching 5/10 War 2: The GPU Inference Price War and Unit Economics Swix details the unit economics and breakeven calculations of Mixtral token pricing while Alessio compares inference benchmarking drama and loss leaders to traditional database benchmarks and ride-sharing economics.31:23–41:33 · Guest teaching 7/10 MoE Hardware Implications and Swix's 'Sour Lesson' Swix explains hardware implications of MoE models and expounds his 'sour lesson' regarding state space models and why machine intelligence doesn't mirror human cognition. Alessio offers insight on hardware memory bandwidth hurdles.41:33–53:57 · Guest teaching 4/10 War 3: The Multimodal Battle and Commercial Monetization Alessio and Swix debate whether DALL-E 3 got worse due to chat integrations (Swix playfully calling it a 'skill issue') and explore go-to-market monetization paths for Midjourney and ElevenLabs voice models.53:57–1:05:07 · Guest teaching 4/10 War 4: RAG Ops, Vector Databases and Platform Gravity Alessio brings deep venture and devtools knowledge analyzing how vector DBs compete with frameworks, drawing historical parallels to Jamstack, Vercel, and Gatsby. Swix provides insider Netlify context.1:05:07–1:11:55 · Guest teaching 5/10 Semantic Programming and the Limits of Coding Agents Alessio outlines his framework for semantic programming and the engineering hazards of coding agents generating technical debt, while Swix pushes back with his inner versus outer developer loop paradigm.1:11:55–1:20:34 · Guest teaching 4/10 December Highlights: Prompt Hacks, Gemini, and AI Hardware Alessio critiques Anthropic's needle-in-a-haystack prompt hacks and discusses hardware context appliances like Tab and Rabbit R1 alongside privacy trade-offs. Both wrap up the recap episode.0:01–3:36 · Guest disagreement 1/10 Introducing the Monthly Recap: The Four Wars Framework Alessio and Swix smoothly introduce their monthly recap format. Alessio breaks down the four wars structure and upcoming guest lineup while Swix adds context on their framing. Collaborative and conversational peer dynamic.3:36–8:04 · Guest disagreement 2/10 The Non-Wars: AI Trends That Cooled Off in 2023 Swix explains why open source AI and agents cooled off, while Alessio provides developer perspective on why code model experimentation is fragmented across tools like VS Code and Cursor. Solid, respectful domain discussion.8:04–10:10 · Guest disagreement 1/10 The End of Low Background Tokens and Data Provenance Alessio takes the lead explaining the concept of low background tokens using Jeff Huber's steel analogy after nuclear testing. He details how synthetic data pollution will force data engineering and first-party datasets like Databricks Dolly.10:10–17:54 · Guest disagreement 2/10 War 1: Copyright Holders vs. AI Labs and Walled Gardens Alessio shares historical background on data scraping from his Rap Genius moderation days when Google scraped lyrics, while Swix analyzes fair use criteria in the New York Times OpenAI lawsuit and closed gardens of Reddit and Stack Overflow.17:54–21:41 · Guest disagreement 2/10 The Synthetic Data Frontier and Distribution Spiking Swix recaps key takeaways from NeurIPS on synthetic data scale versus fine-tuning synthetic data, citing conversations with Louis Castricado and Andre Karpathy's highlighted paper. Alessio engages on the risks of mode collapse in synthetic data generation.21:41–31:23 · Guest disagreement 2/10 War 2: The GPU Inference Price War and Unit Economics Swix details the unit economics and breakeven calculations of Mixtral token pricing while Alessio compares inference benchmarking drama and loss leaders to traditional database benchmarks and ride-sharing economics.31:23–41:33 · Guest disagreement 2/10 MoE Hardware Implications and Swix's 'Sour Lesson' Swix explains hardware implications of MoE models and expounds his 'sour lesson' regarding state space models and why machine intelligence doesn't mirror human cognition. Alessio offers insight on hardware memory bandwidth hurdles.41:33–53:57 · Guest disagreement 3/10 War 3: The Multimodal Battle and Commercial Monetization Alessio and Swix debate whether DALL-E 3 got worse due to chat integrations (Swix playfully calling it a 'skill issue') and explore go-to-market monetization paths for Midjourney and ElevenLabs voice models.53:57–1:05:07 · Guest disagreement 2/10 War 4: RAG Ops, Vector Databases and Platform Gravity Alessio brings deep venture and devtools knowledge analyzing how vector DBs compete with frameworks, drawing historical parallels to Jamstack, Vercel, and Gatsby. Swix provides insider Netlify context.1:05:07–1:11:55 · Guest disagreement 2/10 Semantic Programming and the Limits of Coding Agents Alessio outlines his framework for semantic programming and the engineering hazards of coding agents generating technical debt, while Swix pushes back with his inner versus outer developer loop paradigm.1:11:55–1:20:34 · Guest disagreement 2/10 December Highlights: Prompt Hacks, Gemini, and AI Hardware Alessio critiques Anthropic's needle-in-a-haystack prompt hacks and discusses hardware context appliances like Tab and Rabbit R1 alongside privacy trade-offs. Both wrap up the recap episode.0:01–3:36 · The hosts pushing back 2/10 Introducing the Monthly Recap: The Four Wars Framework Alessio and Swix smoothly introduce their monthly recap format. Alessio breaks down the four wars structure and upcoming guest lineup while Swix adds context on their framing. Collaborative and conversational peer dynamic.3:36–8:04 · The hosts pushing back 2/10 The Non-Wars: AI Trends That Cooled Off in 2023 Swix explains why open source AI and agents cooled off, while Alessio provides developer perspective on why code model experimentation is fragmented across tools like VS Code and Cursor. Solid, respectful domain discussion.8:04–10:10 · The hosts pushing back 1/10 The End of Low Background Tokens and Data Provenance Alessio takes the lead explaining the concept of low background tokens using Jeff Huber's steel analogy after nuclear testing. He details how synthetic data pollution will force data engineering and first-party datasets like Databricks Dolly.10:10–17:54 · The hosts pushing back 1/10 War 1: Copyright Holders vs. AI Labs and Walled Gardens Alessio shares historical background on data scraping from his Rap Genius moderation days when Google scraped lyrics, while Swix analyzes fair use criteria in the New York Times OpenAI lawsuit and closed gardens of Reddit and Stack Overflow.17:54–21:41 · The hosts pushing back 2/10 The Synthetic Data Frontier and Distribution Spiking Swix recaps key takeaways from NeurIPS on synthetic data scale versus fine-tuning synthetic data, citing conversations with Louis Castricado and Andre Karpathy's highlighted paper. Alessio engages on the risks of mode collapse in synthetic data generation.21:41–31:23 · The hosts pushing back 2/10 War 2: The GPU Inference Price War and Unit Economics Swix details the unit economics and breakeven calculations of Mixtral token pricing while Alessio compares inference benchmarking drama and loss leaders to traditional database benchmarks and ride-sharing economics.31:23–41:33 · The hosts pushing back 2/10 MoE Hardware Implications and Swix's 'Sour Lesson' Swix explains hardware implications of MoE models and expounds his 'sour lesson' regarding state space models and why machine intelligence doesn't mirror human cognition. Alessio offers insight on hardware memory bandwidth hurdles.41:33–53:57 · The hosts pushing back 2/10 War 3: The Multimodal Battle and Commercial Monetization Alessio and Swix debate whether DALL-E 3 got worse due to chat integrations (Swix playfully calling it a 'skill issue') and explore go-to-market monetization paths for Midjourney and ElevenLabs voice models.53:57–1:05:07 · The hosts pushing back 3/10 War 4: RAG Ops, Vector Databases and Platform Gravity Alessio brings deep venture and devtools knowledge analyzing how vector DBs compete with frameworks, drawing historical parallels to Jamstack, Vercel, and Gatsby. Swix provides insider Netlify context.1:05:07–1:11:55 · The hosts pushing back 3/10 Semantic Programming and the Limits of Coding Agents Alessio outlines his framework for semantic programming and the engineering hazards of coding agents generating technical debt, while Swix pushes back with his inner versus outer developer loop paradigm.1:11:55–1:20:34 · The hosts pushing back 2/10 December Highlights: Prompt Hacks, Gemini, and AI Hardware Alessio critiques Anthropic's needle-in-a-haystack prompt hacks and discusses hardware context appliances like Tab and Rabbit R1 alongside privacy trade-offs. Both wrap up the recap episode.

speaking balance: gold is the hosts, purple is the guest (3 minute bins)

0:00 · the hosts 0% · guest 100%0:00 · the hosts 0% · guest 100%3:00 · the hosts 0% · guest 100%3:00 · the hosts 0% · guest 100%6:00 · the hosts 0% · guest 100%6:00 · the hosts 0% · guest 100%9:00 · the hosts 0% · guest 100%9:00 · the hosts 0% · guest 100%12:00 · the hosts 0% · guest 100%12:00 · the hosts 0% · guest 100%15:00 · the hosts 0% · guest 100%15:00 · the hosts 0% · guest 100%18:00 · the hosts 0% · guest 100%18:00 · the hosts 0% · guest 100%21:00 · the hosts 0% · guest 100%21:00 · the hosts 0% · guest 100%24:00 · the hosts 0% · guest 100%24:00 · the hosts 0% · guest 100%27:00 · the hosts 0% · guest 100%27:00 · the hosts 0% · guest 100%30:00 · the hosts 0% · guest 100%30:00 · the hosts 0% · guest 100%33:00 · the hosts 0% · guest 100%33:00 · the hosts 0% · guest 100%36:00 · the hosts 0% · guest 100%36:00 · the hosts 0% · guest 100%39:00 · the hosts 0% · guest 100%39:00 · the hosts 0% · guest 100%42:00 · the hosts 0% · guest 100%42:00 · the hosts 0% · guest 100%45:00 · the hosts 0% · guest 100%45:00 · the hosts 0% · guest 100%48:00 · the hosts 0% · guest 100%48:00 · the hosts 0% · guest 100%51:00 · the hosts 0% · guest 100%51:00 · the hosts 0% · guest 100%54:00 · the hosts 0% · guest 100%54:00 · the hosts 0% · guest 100%57:00 · the hosts 0% · guest 100%57:00 · the hosts 0% · guest 100%1:00:00 · the hosts 0% · guest 100%1:00:00 · the hosts 0% · guest 100%1:03:00 · the hosts 0% · guest 100%1:03:00 · the hosts 0% · guest 100%1:06:00 · the hosts 0% · guest 100%1:06:00 · the hosts 0% · guest 100%1:09:00 · the hosts 0% · guest 100%1:09:00 · the hosts 0% · guest 100%1:12:00 · the hosts 0% · guest 100%1:12:00 · the hosts 0% · guest 100%1:15:00 · the hosts 0% · guest 100%1:15:00 · the hosts 0% · guest 100%1:18:00 · the hosts 0% · guest 100%1:18:00 · the hosts 0% · guest 100%
Sharpest disagreement ▶ 43:45 Skill Issue Disagreement on DALL-E 3

When Alessio claims DALL-E 3 output quality deteriorated into boomer vibes, Swix playfully but directly dismisses his prompt crafting as a skill issue.

Hardest push from the hosts ▶ 1:01:02 Pushback on Framework Dominance Assumption

Alessio challenges the assumption that owning the framework always wins by contrasting Vercel with Gatsby, arguing that owning the framework without the monetization cloud still leads to failure.

Biggest teaching moment ▶ 37:40 Swix Explains the Sour Lesson Philosophy

Swix educates listeners and Alessio on why trying to replicate human neurological processes in AI architectures is flawed, coining the sour lesson corollary to Sutton's bitter lesson.

The host holds their own ▶ 8:14 Alessio Explains Low Background Steel and Tokens

Alessio demonstrates deep domain mastery by articulating Jeff Huber's pre-atomic steel radiation metaphor to explain the degradation of post-2023 web training data.

the scores for every segment, with the reasoning behind each
ChapterTopicThe hosts as informed peerGuest teachingGuest disagreementThe hosts pushing backWhy
Introducing the Monthly Recap: The Four Wars Framework 6312 Alessio and Swix smoothly introduce their monthly recap format. Alessio breaks down the four wars structure and upcoming guest lineup while Swix adds context on their framing. Collaborative and conversational peer dynamic.
The Non-Wars: AI Trends That Cooled Off in 2023 6422 Swix explains why open source AI and agents cooled off, while Alessio provides developer perspective on why code model experimentation is fragmented across tools like VS Code and Cursor. Solid, respectful domain discussion.
The End of Low Background Tokens and Data Provenance 8211 Alessio takes the lead explaining the concept of low background tokens using Jeff Huber's steel analogy after nuclear testing. He details how synthetic data pollution will force data engineering and first-party datasets like Databricks Dolly.
War 1: Copyright Holders vs. AI Labs and Walled Gardens 7321 Alessio shares historical background on data scraping from his Rap Genius moderation days when Google scraped lyrics, while Swix analyzes fair use criteria in the New York Times OpenAI lawsuit and closed gardens of Reddit and Stack Overflow.
The Synthetic Data Frontier and Distribution Spiking 5622 Swix recaps key takeaways from NeurIPS on synthetic data scale versus fine-tuning synthetic data, citing conversations with Louis Castricado and Andre Karpathy's highlighted paper. Alessio engages on the risks of mode collapse in synthetic data generation.
War 2: The GPU Inference Price War and Unit Economics 7522 Swix details the unit economics and breakeven calculations of Mixtral token pricing while Alessio compares inference benchmarking drama and loss leaders to traditional database benchmarks and ride-sharing economics.
MoE Hardware Implications and Swix's 'Sour Lesson' 6722 Swix explains hardware implications of MoE models and expounds his 'sour lesson' regarding state space models and why machine intelligence doesn't mirror human cognition. Alessio offers insight on hardware memory bandwidth hurdles.
War 3: The Multimodal Battle and Commercial Monetization 7432 Alessio and Swix debate whether DALL-E 3 got worse due to chat integrations (Swix playfully calling it a 'skill issue') and explore go-to-market monetization paths for Midjourney and ElevenLabs voice models.
War 4: RAG Ops, Vector Databases and Platform Gravity 8423 Alessio brings deep venture and devtools knowledge analyzing how vector DBs compete with frameworks, drawing historical parallels to Jamstack, Vercel, and Gatsby. Swix provides insider Netlify context.
Semantic Programming and the Limits of Coding Agents 8523 Alessio outlines his framework for semantic programming and the engineering hazards of coding agents generating technical debt, while Swix pushes back with his inner versus outer developer loop paradigm.
December Highlights: Prompt Hacks, Gemini, and AI Hardware 7422 Alessio critiques Anthropic's needle-in-a-haystack prompt hacks and discusses hardware context appliances like Tab and Rabbit R1 alongside privacy trade-offs. Both wrap up the recap episode.

Statements from this episode (0)

Nothing in this episode matches those filters. clear them

Made with StarZero

Turn any episode into a week of clips.

This entire site, over 200 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.