Keep pulling the thread on Sholto Douglas.
The AI compute supercycle is properly beginning in 2024 due to increased chip availability from foundries like TSMC.
Anthropic's latest Sonnet model is smarter than its current top-tier Opus model.
A new inference stack developed for the Gemini program at Google saved several hundred million dollars in its first six months of operation.
Anthropic's strategic focus is on long-term AI alignment and near-term economic impact, particularly in coding and computer use.
Anthropic believes AGI is achievable within the next couple of years using paradigms similar to current ones.
An Anthropic AI agent was able to work autonomously for 30 hours to build a chat application similar to Slack or Microsoft Teams.
OpenAI's O-1 was the first major release to successfully combine reinforcement learning with large language models.
The AI industry is on track to create AI that is at least as capable as most humans on most computer-facing tasks within the next 2-3 years.
In the next one to two years, individuals will be able to manage a team of AI coding agents that work 24/7.
Anthropic's latest model improved its SWE-Bench score from approximately 72% to 78%.
The AI field's performance on the SWE-Bench benchmark was under 20% as recently as one year ago.
Cognition rebuilt its architecture around Anthropic's latest model because it was so useful.