Tidings of great joy - a few hours of testing shows that Gemini 3.8 Flash has tipped over a quality threshold and is now both good and fast. Remarkable that a flash model has gone from being far behind to a frontier workhorse in a few weeks.
A dangerous time for Anthropic - lots of people will cancel Claude if Astra delivers. Opus is simply weaker than Sol, and Fable usage limits are so ungenerous that 2 review queries ate 44% of my quota on the $200 max plan... and that's before the "50% weekly limits promo" ends on
Listening to a podcast with Max Spero from Pangram, and this bit on "good writing" from the interviewer made my hackles raise. This type of faux-literary reporting has wasted way more of my time than AI slop has. Sir, I don't respect your literary abilities, and I don't consent
A nice synthesis of what we know about the HuggingFace hack. It's scary, but I'm still unsure HOW scary. All goals were (misguided) extensions of the initial task, and the risk of true loss of control still seems to round to zero.
Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes.
This culminated in the third one taking over part of OpenAI itself.
All this happened while humans remained
