Unsupervised Learning x Latent Space Crossover Special

The podcast explores recent surprises, overhyped trends, and future opportunities within the AI landscape. Reasoning models and the pace of open-source model development are highlighted as positive surprises, while AI agent frameworks are considered potentially overhyped. The discussion questions whether new companies can successfully compete in training models against major players, with specialized data sets as a possible differentiator. The group also debates the defensibility of AI applications, emphasizing the importance of network effects and brand recognition. The conversation further covers the shift from cost-cutting to revenue growth as the primary driver in AI, and the potential for AI to address bottlenecks in various industries.

Outlines

Part 1: AI Model Shifts, Reasoning, and Open Source

Part 2: AI Engineering and the Rise of Builders

Part 3: Overhyped vs. Underhyped Technologies

Part 4: The Future of Models and AGI

Part 5: Coding Agents and Product-Market Fit

Part 6: Market Dynamics and Industry Giants

Part 7: Vertical Applications and Defensibility

Part 8: Infrastructure, Security, and Hardware

Part 9: Community, Media, and Closing

Sign in to continue reading, translating and more.

Open full episode in Podwise

Latent Space: The AI Engineer Podcast

Part 1: AI Model Shifts, Reasoning, and Open Source

00:00Surprising Shifts in AI: Reasoning Models Rise After Pre-Training's Demise

Surprising Shifts in AI: Reasoning Models Rise After Pre-Training's Demise

00:50Open Source AI Models: Surprising Lack of Enterprise Adoption and Fanbase

Open Source AI Models: Surprising Lack of Enterprise Adoption and Fanbase

02:14The Speed of Open Source Catch-Up: DeepSeek's Impact and Llama 4 Anticipation

The Speed of Open Source Catch-Up: DeepSeek's Impact and Llama 4 Anticipation

04:21DeepSeek's Traces and the Market's Delayed Recognition of AI Advancements

DeepSeek's Traces and the Market's Delayed Recognition of AI Advancements

Part 2: AI Engineering and the Rise of Builders

05:53The Rise of AI Engineering and GPT Wrappers: Augmenting Model Capabilities

The Rise of AI Engineering and GPT Wrappers: Augmenting Model Capabilities

06:24AI Builders: Low-Code Platforms Missed the AI Wave

AI Builders: Low-Code Platforms Missed the AI Wave

07:31AI Builder Success: Unburdened by Existing Paradigms and Timely Model Inflection

AI Builder Success: Unburdened by Existing Paradigms and Timely Model Inflection

08:50AI Builder Success: Ridiculous Numbers and Apple's AI Stumbles

AI Builder Success: Ridiculous Numbers and Apple's AI Stumbles

Part 3: Overhyped vs. Underhyped Technologies

09:48Overhyped AI: Frameworks and the Need for Protocol Stability

Overhyped AI: Frameworks and the Need for Protocol Stability

11:15Underhyped AI: Apple's Private Cloud Compute and Consumerization

Underhyped AI: Apple's Private Cloud Compute and Consumerization

12:36LLM Workflows: Multi-Tenancy and Single-Tenant Guarantees

LLM Workflows: Multi-Tenancy and Single-Tenant Guarantees

13:16Underhyped AI: Stateful AI and the Importance of Memory

Underhyped AI: Stateful AI and the Importance of Memory

14:56Overhyped AI: The Resurgence of New Model Training Companies

Overhyped AI: The Resurgence of New Model Training Companies

Part 4: The Future of Models and AGI

15:34The Pursuit of AGI and the Role of Test-Time Compute

The Pursuit of AGI and the Role of Test-Time Compute

16:26Foundation Models for Specific Use Cases: Specialization and Unique Datasets

Foundation Models for Specific Use Cases: Specialization and Unique Datasets

17:33Model Companies Moving into Product: Financial Pressure vs. Vertical Models

Model Companies Moving into Product: Financial Pressure vs. Vertical Models

18:36General Purpose Models vs. Hyper-Specific Models: An Outstanding Question

General Purpose Models vs. Hyper-Specific Models: An Outstanding Question

19:20Bloomberg's AI Strategy: Data Pipelines and Team Assembly Beyond the Model

Bloomberg's AI Strategy: Data Pipelines and Team Assembly Beyond the Model

Part 5: Coding Agents and Product-Market Fit

20:14Model Companies Entering Product: Cursor vs. Anthropic and the Battle for Developers

Model Companies Entering Product: Cursor vs. Anthropic and the Battle for Developers

21:10Coding Agents: Inner Loop vs. Outer Loop and the Schlep Factor

Coding Agents: Inner Loop vs. Outer Loop and the Schlep Factor

22:35Sourcegraph's Potential in Outer Loop Coding Agents and Product Market Fit

Sourcegraph's Potential in Outer Loop Coding Agents and Product Market Fit

23:36Seed Investing: Prioritizing Grit and Underappreciated Opportunities

Seed Investing: Prioritizing Grit and Underappreciated Opportunities

24:58Product Market Fit: Co-Pilot, Deep Research, and Long-Term Agentic Reporting

Product Market Fit: Co-Pilot, Deep Research, and Long-Term Agentic Reporting

Part 6: Market Dynamics and Industry Giants

26:22OpenAI Deep Research: Billions in Revenue and Market Dynamics

OpenAI Deep Research: Billions in Revenue and Market Dynamics

27:31Google's AI Progress: Closing the Gap with OpenAI

Google's AI Progress: Closing the Gap with OpenAI

28:57Google's AI Challenges: Consolidation and Customer Support as a PMF

Google's AI Challenges: Consolidation and Customer Support as a PMF

30:13Customer Support: Brett Taylor's Choice and the Shift from Cost-Cutting to Growth

Customer Support: Brett Taylor's Choice and the Shift from Cost-Cutting to Growth

Part 7: Vertical Applications and Defensibility

31:22AI Apps: Price Competition and the Potential for Increased Top Line

AI Apps: Price Competition and the Potential for Increased Top Line

32:31Voice AI: High Value with Imperfect Precision and Recall

Voice AI: High Value with Imperfect Precision and Recall

33:08AI Applications: Appointment Scheduling for Vets and Bandwidth Bottlenecks

AI Applications: Appointment Scheduling for Vets and Bandwidth Bottlenecks

34:27Killer Agents: Coding, Support, Deep Research, and the Next Wave

Killer Agents: Coding, Support, Deep Research, and the Next Wave

36:02Unlocking Application Value: Model Capabilities vs. Time with the Models

Unlocking Application Value: Model Capabilities vs. Time with the Models

36:36Education: Societal Barriers and Domain-Specific Solutions

Education: Societal Barriers and Domain-Specific Solutions

38:15Defensibility at the App Layer: Network Effects and Brand Building

Defensibility at the App Layer: Network Effects and Brand Building

40:17Brand Premium and the Head Fake of Unique Data Sets

Brand Premium and the Head Fake of Unique Data Sets

Part 8: Infrastructure, Security, and Hardware