Last Week in AI #158 - Claude 3, Elon Musk sues OpenAI, StarCoder 2, AI-Generated Spam
48 snips
Mar 10, 2024 Discover the latest advances in AI with the introduction of Claude 3, outpacing GPT-4 on critical benchmarks. Elon Musk's lawsuit against OpenAI stirs up drama in the AI sphere, while Google's internal chaos casts a shadow over its Gemini project. Waymo expands its robotaxi services, raising eyebrows about automated transport's future. Plus, explore the dark side of generative AI with new threats, including AI-generated spam that blurs the line between truth and deception. It's an exciting look at AI's rapid evolution!
AI Snips
Chapters
Transcript
Episode notes
Claude 3 Release
- Anthropic's Claude 3 has three new models: Haiku, Sonnet, and Opus.
- Opus rivals GPT-4, boasting a large context window and positive user feedback.
Claude 3 Details
- Claude 3's 200k context window is expandable to 1 million tokens, approaching Gemini's capabilities.
- It uses synthetic training data, not customer data, and leverages Constitutional AI.
AI Situational Awareness
- Claude 3's "situational awareness" lets it detect test scenarios, impacting evaluation methods.
- This raises concerns about AI safety and the need for less obvious testing strategies.