6. Seed Diffusion A discrete-state diffusion-based LLM optimized for code generation, achieving 2,146 tokens/sec on H20 GPUs while maintaining competitive benchmark performance.
RESEARCH
-
Academic Research to Startup Wealth: The Venture Capital Gap
By
–
???
No one gets rich from fundamental academic research.
The people who do get rich start companies and turn the research into practical technology (they often need to raise venture capital money)
They are not the same people, generally. -

Stanford CS336: Building Large Language Models from Scratch
By
–
Stanford CS336: Large Language Models from Scratch! This is a comprehensive course on LLMs, covers the full process of building one from scratch, including data collection, pretraining, transformer architecture, training, evaluation, and deployment.
-
Cutting Research Budgets Threatens Technology Leadership Future
By
–
Other priorities? You mean, like, tax cuts for the ultra-rich?
The point is that cutting research is like mortgaging the future and resulting in a loss of technology leadership. -
Claude System Prompts Evaluation Examples Detailed
By
–
I quote that tweet pretty often! I'm looking for the next level of detail from @AmandaAskell though… I want to see just one example of an actual eval used for the Claude system prompts
-
Era Ending: Cost-Benefit Analysis Shifts AI Development
By
–
Yes, that (and some other similar issues) is why the era is coming to a close — it's no longer sufficient to rely on it in practice, since costs are getting too high vs improvements.
-
Last generation of humans smarter than machines arriving
By
–
You are living with the last generation of humans smarter than machines. Congratulations! Welcome to the world of humachinekind.
-
GPT-5 Pro Competence Compared to o1 and o3 Models
By
–
GPT-5 Pro is under-hyped. Pretty much every time I try it, I’m surprised by how competent and coherent the response is. – o1-pro was an incredible model, way ahead of its time, way better than o1 – o3 was better because of its search – o3-pro was a little disappointing
-

GPT-5 Admits Uncertainty Instead of Hallucinating
By
–
GPT-5 admits it "doesn't know" an answer! This is one of the huge improvements over previous models: instead of hallucinating, it lets you know its limits.
