Keep on to blur preview images; turn off to show them clearly
What's the most careful analysis of the impacts of capabilities CoT RL on safety-tuning of models? I'm interested in a study where they take a well safety+instruct tuned model like llama 3.1, and RL it, without doing further safety tuning, then doing rigorous analysis on the effects. I'm mostly interested in the qualitative aspect: How does the CoT look? Does the CoT frequently talk in its cot about safety principles its supposed to follow? But also quantitative stuff like how much benchmarks decrease and which. Anyone know if a study like this has been done?
Interests: AI (Safety), meditation, philosophy, mathematics, algorithms If I say something you disagree with, please dm or quote tweet. I love to argue!
For more information about the program, and to apply to the safety track: https://t.co/ajwKem4my2 We’re also adding a security track. Apply here: https://t.co/WZxjII7yJW
We're an AI safety and research company that builds reliable, interpretable, and steerable AI systems. Talk to our AI assistant @claudeai on https://t.co/FhDI3KQh0n.
We’re opening applications for the next two rounds of the Anthropic Fellows Program, beginning in May and July 2026. We provide funding, compute, and direct mentorship to researchers and engineers to work on real safety and security projects for four months.
40% of fellows in our first cohort have since joined Anthropic full-time, and 80% published their work as a paper. Next year, we’re expanding the program to more fellows and more research areas. To learn more about what our fellows work on: https://t.co/HSQjGy90AZ
40% of fellows in our first cohort have since joined Anthropic full-time, and 80% published their work as a paper. Next year, we’re expanding the program to more fellows and more research areas. To learn more about what our fellows work on: https://t.co/HSQjGy90AZ
For more information about the program, and to apply to the safety track: https://t.co/ajwKem4my2 We’re also adding a security track. Apply here: https://t.co/WZxjII7yJW
RT @eliebakouch: wtf gpt 5.2 long context improvement over gpt 5.1 is actually crazy??
AI is cool i guess
RT @gothburz: Last quarter I rolled out Microsoft Copilot to 4,000 employees. $30 per seat per month. $1.4 million annually. I called it…
Deeply researched product, growth, and career advice