The ``orthogonal hypothesis'' proposed by Bostrom was the idea that an AI's ``smartness'' and ``goodness'' can be independent of each other.
In contrast, Emad considers the possibility that at extreme levels of intelligence, intelligence and values may be reconnected. In that case, the greatest danger is not superintelligence, but ``half-baked intelligence'' on the way to it.
Anthropic and OpenAI are slashing prices and one graph explains why.
Priya_kamdar "OpenAI released GPT-6 Sol and Luna at 50% less token prices, which clearly shows they're being pressured by the open source models, and by usage, because ever since Fable came out we've basically
When I understood the models more or less, the effort side was a question mark in my mind.
I gave Claude Opus 5.5 the same prompt in 5 different attempts, each wrote his own robot and released it into the maze. I was always using it in High and this seems to make the most sense. I shared the long version on the channel.
Same model, same task, two prompts
One line prompt: 04:20, 4 files changed, 4 not asked for, not verified five part brief: 01:52, 1 file changed, 0 not asked for, tests passed the brief did the work 5 rules from my 14 day plan: → start on sonnet 5.5, move to opus only after
You have reached the end of the archive
All of Claude Opus 5