Last Week in AI #207 - GPT 4.1, Gemini 2.5 Flash, Ironwood, Claude Max
760 snips
Apr 18, 2025 OpenAI's latest GPT-4.1 model brings enhanced coding capabilities and a massive million-token context window. However, the company faces criticism for scaling back safety testing amid these advancements. Meanwhile, Meta encounters serious allegations regarding collaboration with China in AI development, stirring public scrutiny. Additionally, XAI launches its Grok 3 API, showcasing competitive features, while Crusoe's $3.5 billion data center investment raises significant implications for the industry and national security.
AI Snips
Chapters
Transcript
Episode notes
GPT-4.1 Coding Focus
- OpenAI's GPT-4.1 models focus on coding with up to 1 million token context windows and are cheaper variants with strong performance.
- These models advance real-world software engineering tasks and include multimodal abilities like video analysis.
ChatGPT Gains Memory
- ChatGPT now remembers and references all past conversations to enable personalized experiences.
- Users control memory use and can opt out, with rollouts starting on paid tiers except in some strict regions.
Google's Gemini Efficient Models
- Google's Gemini 2.5 Flash offers cheaper, faster distilled versions of their prior larger model for market segmentation.
- This reflects industry trends of mining large pre-trained models for lightweight, efficient inference variants.