NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·2h ago
Data Science
Simon Willison's Blog·May 19·5 min read
The last six months in LLMs in five minutes
TL;DR
Large language models (LLMs)—AI systems trained on vast text data to generate human-like responses—have evolved rapidly, creating a need for practitioners to stay current. Simon Willison compiled annotated slides summarizing major LLM developments over six months using his custom presentation tool.
✦ Why It Matters
Engineers can quickly grasp six months of LLM progress and identify which new tools or techniques are worth investigating for their projects.
Key Takeaways
- Test new coding agents for daily software development tasks.
- Experiment with open-weight models on personal devices.
- Stay updated on advancements in LLM capabilities.
- Leverage Reinforcement Learning techniques for better coding outcomes.
Full Summary
Over the last six months, the landscape of large language models (LLMs) has evolved dramatically, particularly in coding capabilities. OpenAI and Anthropic focused on enhancing their coding agents through Reinforcement Learning from Verifiable Rewards, resulting in models that transitioned from unreliable to dependable for real-world coding tasks.
Notably, OpenAI's Codex and Anthropic's Claude Opus 4.5 emerged as leaders in this space. Meanwhile, open-weight models like Qwen3.6-35B-A3B have shown impressive performance on personal laptops, outperforming some of the leading proprietary models.
This shift indicates a growing accessibility of powerful AI tools for developers, allowing them to leverage advanced capabilities without needing high-end infrastructure. The implications are significant, as these advancements can streamline software development processes and democratize access to AI-driven coding assistance.