Last Week in AI #231 - Claude Cowork, Anthropic $10B, Deep Delta Learning | Last Week in AI

The Last Week in AI podcast covers recent developments in artificial intelligence, focusing on tools, business deals, open-source projects, and research advancements. Anthropic's new Co-work tool, an integration of cloud code into the Claude desktop app, is highlighted for its potential to streamline tasks, though security considerations are noted. Google's Gemini introduces Personal Intelligence, connecting to user data across Google services, while also facing scrutiny for inaccuracies in AI-generated health summaries. Financings for Anthropic and XAI are discussed, alongside Nvidia's supply chain challenges with H200 AI chips for China. Open-source projects like Nemetron Cascade and Falcon H1R-7B are reviewed, emphasizing advancements in reinforcement learning and hybrid model architectures. The podcast also explores research papers on deep delta learning and recursive language models, examining new approaches to memory and positional embeddings in LLMs.

Outlines

Part 1: Anthropic’s New Tools and Security

Part 2: Google and Gemini Ecosystem

Part 3: Enterprise Agents and Market Valuations

Part 4: Hardware Supply Chain and Infrastructure

Part 5: Training Methods and Model Architectures

Part 6: Advanced Research and Context Scaling

Part 7: Safety, Alignment, and Geopolitics

Part 8: Conclusion

Sign in to continue reading, translating and more.

Continue

Last Week in AI #231 - Claude Cowork, Anthropic $10B, Deep Delta Learning

Last Week in AI

Part 1: Anthropic’s New Tools and Security

00:10Anthropic's Co-work: Integrating Cloud Code for Desktop Automation

Anthropic's Co-work: Integrating Cloud Code for Desktop Automation

03:36Desktop Agent Security: Anthropic's Sandboxed Virtual Machines

Desktop Agent Security: Anthropic's Sandboxed Virtual Machines

06:14Anthropic's Pricing Shift: Selling Labor with Co-work

Anthropic's Pricing Shift: Selling Labor with Co-work

Part 2: Google and Gemini Ecosystem

09:48Gemini's Personal Intelligence: Integrating with Google Services

Gemini's Personal Intelligence: Integrating with Google Services

12:45Google Removes AI Health Summaries Due to Dangerous Flaws

Google Removes AI Health Summaries Due to Dangerous Flaws

14:46AI Overviews Become Google: Overtaking ChatGPT Use Cases

AI Overviews Become Google: Overtaking ChatGPT Use Cases

Part 3: Enterprise Agents and Market Valuations

17:23Slackbot's AI Upgrade: An Agentic Assistant for Business Apps

Slackbot's AI Upgrade: An Agentic Assistant for Business Apps

20:11Anthropic's $10 Billion Raise: Valued at $350 Billion

Anthropic's $10 Billion Raise: Valued at $350 Billion

22:25XAI Raises $20 Billion: Valued at $230 Billion

XAI Raises $20 Billion: Valued at $230 Billion

Part 4: Hardware Supply Chain and Infrastructure

24:12Nvidia's Supply Chain Miracle: Meeting China's H200 Demand

Nvidia's Supply Chain Miracle: Meeting China's H200 Demand

27:50China's Rush for Nvidia Chips: Export Control Uncertainties

China's Rush for Nvidia Chips: Export Control Uncertainties

31:12CoreWeave Amends Credit Agreements: Addressing Hardware Delays

CoreWeave Amends Credit Agreements: Addressing Hardware Delays

Part 5: Training Methods and Model Architectures

34:30LLM Arena Valued at $1.7 Billion: Securing Series A Funding

LLM Arena Valued at $1.7 Billion: Securing Series A Funding

35:54Nemetron Cascade: Scaling Reinforcement Learning to Prevent Catastrophic Forgetting

Nemetron Cascade: Scaling Reinforcement Learning to Prevent Catastrophic Forgetting

40:04Nvidia's Open Source Push: Promoting Hardware Sales

Nvidia's Open Source Push: Promoting Hardware Sales

43:32DeepSeek's MHC: Manifold Constraint Hyperconnections for Training Stability

DeepSeek's MHC: Manifold Constraint Hyperconnections for Training Stability

49:53iQuest Coder V1: A Coding Model with a Multi-Pass Workflow

iQuest Coder V1: A Coding Model with a Multi-Pass Workflow

54:04Falcon H1R-7B: A Hybrid Transformer-Mamba Model for Reasoning

Falcon H1R-7B: A Hybrid Transformer-Mamba Model for Reasoning

57:23Mamba 2 and Transformer Mergers: Efficiency and Optimization

Mamba 2 and Transformer Mergers: Efficiency and Optimization

Part 6: Advanced Research and Context Scaling

1:01:42Deep Delta Learning: A More Efficient Way of Designing Data Flow

Deep Delta Learning: A More Efficient Way of Designing Data Flow

1:04:47Deep Delta Learning: Mitigating Vanishing Gradients

Deep Delta Learning: Mitigating Vanishing Gradients

1:09:45Recursive Language Models: Scaling to Arbitrarily Long Prompts

Recursive Language Models: Scaling to Arbitrarily Long Prompts

1:13:00Conditional Memory: Splitting Computation and Memory

Conditional Memory: Splitting Computation and Memory

1:15:37Conditional Memory: N-grams and Multi-Head Hashing

Conditional Memory: N-grams and Multi-Head Hashing

1:18:34Dropping Positional Embeddings: Extending Context of Pre-trained LLMs

Dropping Positional Embeddings: Extending Context of Pre-trained LLMs

1:20:17Removing Positional Embeddings: Recalibrating for Longer Sequences

Removing Positional Embeddings: Recalibrating for Longer Sequences

Part 7: Safety, Alignment, and Geopolitics

1:25:34Anthropic's Constitutional Classifiers: Defenses Against Jailbreaks

Anthropic's Constitutional Classifiers: Defenses Against Jailbreaks

1:30:24Red Teaming and Bounties: Testing Model Defenses

Red Teaming and Bounties: Testing Model Defenses

1:31:30Nvidia CEO: Purchase Orders Will Signal Chinese Approval of H200

Nvidia CEO: Purchase Orders Will Signal Chinese Approval of H200

1:32:31China AI Leaders Warn of Widening Gap with U.S.

China AI Leaders Warn of Widening Gap with U.S.

1:34:54Export Controls: Crippling Chinese AI Efforts

Export Controls: Crippling Chinese AI Efforts

1:37:21Jake Sullivan: Trump Removed Biden's AI Chip Export Controls

Jake Sullivan: Trump Removed Biden's AI Chip Export Controls

Part 8: Conclusion

1:40:51Podcast Outro

Podcast Outro