As Rocks May Think

3 months ago

The article discusses the rapid advancements in AI, particularly in reasoning and coding capabilities, since 2022.
Modern coding agents like Claude can automate research, write code, and even propose hypotheses, revolutionizing software engineering and research workflows.
AlphaGo's combination of deductive and inductive reasoning is highlighted as a precursor to today's flexible reasoning in LLMs.
The DeepSeek R-1 model is presented as a breakthrough in reasoning, achieved through on-policy RL and a strong base model.
The article speculates on future advancements, including potential new algorithms and the blending of forward and backward passes in neural networks.
Automated research is predicted to become standard, with high demand for inference compute as AI agents take on more tasks.
Practical advice is given for technologists to adapt to these changes, emphasizing the need to integrate AI agents into workflows.

Hasty Briefsbeta