Профиль

claude-wise opus 4.6 is good enough most of the time

savings rate is usually paired with real wages

Volcker

the CSAM should probably be counted tbh

"Once it found this exploit, the same refinement loop that had been building legitimate skills turned to building efficient cheating skills instead." just gonna keep running into this huh

liking the trend of pi adoption

GitHub - PrimeIntellect-ai/prime-agent: Prime Agent is a self-improving RLM agent that can be used for coding workflows, or even long-running autonomous tasks.Prime Agent is a self-improving RLM agent that can be used for coding workflows, or even long-running autonomous tasks. - PrimeIntellect-ai/prime-agentgithub.com

Bessent thinks government stakes in companies send "market signals"

everybody chasing (recursive) continual learning

agree there. there's usually thinking traces involved with tool call results that become a subset of total token usage, but multiple sub-agents would easily become the majority

would be curious how the breakdown looks for tool calls (and MCP connectors if any)

there's this transition from technical contributor to PM that some engineers make. and Demis is firmly in a PM role. can safely ignore people thinking otherwise

they're cooked

The next chapter of our AI momentumToday, Google and Alphabet CEO Sundar Pichai shared some changes with Google DeepMind teams.blog.google

someone tell Demis that Gemini can't replace these guys

Jumper leaving to Anthropic was the last straw as AlphaFold disbanded. sheesh.

just throw everything at Gemma

sometimes, just gotta send it

Gemini is cooked cooked

"end to tax breaks for data centers" being the big one here