#202 - Qwen-32B, Anthropic's $3.5 billion, LLM Cognitive Behaviors
Summary
This episode analyzes the shift from pure pre-training to inference-time reasoning, discussing GPT-4.5 and Alibaba’s new QwQ-32B model. It covers product launches like Alexa Plus and Manus AI agents, alongside business news including Anthropic’s $61.5B valuation and CoreWeave’s IPO. The hosts also review Microsoft’s Dragon Copilot for healthcare, the UK’s antitrust ruling on Microsoft-OpenAI, and Scale AI’s defense contract.
Topics discussed
Sponsorship and Intro
Hosts Return and Episode Overview
GPT-4.5, Reasoning, and Scaling Laws
Qwen 32B and DeepSeek R1 Analysis
Amazon Alexa AI and Subscription Model
OpenAI Deep Research and Agentic AI
AI in Healthcare and Clinical Burnout
NIST OCR, Anthropic Funding, and Valuation
CoreWeave IPO and Cloud Infrastructure
Waymo-Uber Partnership and Antitrust
Microsoft Defense AI and Military Use
DeepSeek Open Source Infrastructure
Physical Intelligence and BigBench Reasoning
Research on Reasoning Strategies and RL
Turing Award and AI Safety Deterrence
GPT-4.5 Safety Evaluation and Outro
Listen ad-free on Castria