berkeley labs launches sky-t1, an open source reasoning ai that can be trained for $450, and beats early o1 on key benchmarks!!!

berkeley labs launches sky-t1, an open source reasoning ai that can be trained for $450, and beats early o1 on key benchmarks!!!

The more intelligent AI becomes, the better it understands everything, including morality

The more intelligent AI becomes, the better it understands everything, including morality

reaching asi probably requires discovering and inserting more, and stronger, rules of logic into the fine-tuning and instruction tuning steps of training

the openai o3 and deep research transparency and alignment problem

the legalities, ethics and practicalities of building or distilling more intelligent models from 2.0 and o3

the legalities, ethics and practicalities of building or distilling more intelligent models from 2.0 and o3

the legalities, ethics and practicalities of building or distilling more intelligent models from 2.0 and o3

the legalities, ethics and practicalities of building or distilling more intelligent models from 2.0 and o3

the legalities, ethics and practicalities of building or distilling more intelligent models from 2.0 and o3

the openai o3 and deep research transparency and alignment problem

those who think r1 is about deepseek or china miss the point. it's about open source, reinforcement learning, distillation, and algorithmic breakthroughs

reaching asi probably requires discovering and inserting more, and stronger, rules of logic into the fine-tuning and instruction tuning steps of training

Research project on AI architecture towards ultra-high fidelity human imitation

reaching asi probably requires discovering and inserting more, and stronger, rules of logic into the fine-tuning and instruction tuning steps of training

reaching asi probably requires discovering and inserting more, and stronger, rules of logic into the fine-tuning and instruction tuning steps of training

reaching asi probably requires discovering and inserting more, and stronger, rules of logic into the fine-tuning and instruction tuning steps of training

usps bans deliveries from china. how courting escalation could cripple american hi-tech manufacturing

new omnihuman-1 deepfake video generator could embarrass trump

huawei's ascend 910c chip matches nvidia's h100. there will be 1.4 million of them by december. don't think banned countries and open source can't reach agi first.

huawei's ascend 910c chip matches nvidia's h100. there will be 1.4 million of them by december. don't think banned countries and open source can't reach agi first.

deep research is an amazing tool, but it gets us no closer to agi

trump's new tariffs against the rest of the world will hurt everyone... with the big exception of the open source ai community

huawei's ascend 910c chip matches nvidia's h100. there will be 1.4 million of them by december. don't think banned countries and open source can't reach agi first.

usps bans deliveries from china. how courting escalation could cripple american hi-tech manufacturing

question: who might want to track and publish societal psychometrics so we can be proactive as the pace of ai advancement accelerates?