> Mind you, DeepSeek-V4, GLM-5.2 or Kimi-K2.6/2.7 haven't solved a single Erdös problem so far.Ok this is enough Lisan. Mind you, not all Erdös problems are…
One of great problems for DeepSeek is that great priorities for research and great product are not aligned. They can afford ad hoc harnesses, slow (but…
for all non-Whites, Japan defeating Russia was like the last third of Independence Day where yanks disable the shields, blow up the alien flying saucer and…
"China middle income trap" is one of the most retarded ideas in circulation. Their profile is entirely different, irrespective of income *level*. The funny bit is…
On financials alone, the Dutch can trivially afford to cut Chyna off, a couple more quarters of growth in EUV will offset itthe only question is…
We all do that sometimesFlorian Brand: today is one of the days where a month of progress happensbuckle up
respect to people who bother to make precise forecastsaqui 🇧🇷🇨🇻: @teortaxesTex i think your prediction for AA is correctother predictions:44%+ on CAIS TCI5%+ in CAIS RLI64%+…
«If you have a calibration set, you can compute your own importance matrix and quantize the model yourself. All steps use stock llama.cpp tools»Tencent Hy: Can't…
another DeepGEMM updatewhat is funny is that DeepSeek's efficiency (and margin) is a moving target. GLM can optimize DSA, why can't they keep pushing their V4…
meh, imagine it if were immovable insteadthat'd be a real flexas is, that's like that "Friend" gimmickAndrew Curran: OpenAI's first hardware device will be a humanlike…