Any strongly held opinion that can't be objectively verified is dangerous/risky(to us/others). Religious extremists have typically been "RL'd" when they're most impressionable(when young and/or of certain mental inclinations).
@sundarpichai@GoogleDeepMind@demishassabis@IsomorphicLabs Was Noam's exit the cause of the break-up of "the league of legends" or the "canary in the coal mine" ? My guess is the latter. Past actions show that he's very "low inertia".
It’s been said that a deep/complete understanding of AI models/architectures requires full knowledge of their strengths and weaknesses. Typically, model deployment/implementation, which is often resource constrained, requires optimization. ...1/6
Such an approach would bring about meaningful progress in the AI field and probably solve the age-old debate of whether “pattern recognition” is the “end-all be-all ” of intelligence. The human brain should of course serve as both a benchmark and beacon in AI development. ...6/6
It’s been said that a deep/complete understanding of AI models/architectures requires full knowledge of their strengths and weaknesses. Typically, model deployment/implementation, which is often resource constrained, requires optimization. ...1/6
A comprehensive way forward would be to design a “composite” model that adopts the unique model strengths while addressing all overlaps/redundancies, and mitigating/resolving unique weaknesses. The resultant “composite” model can then be optimized. ...5/6
To call a spade a spade, Gemini(2.5 Pro) is relatively rigid in compositionality, compared to ChatGPT 5(Thinking). At least in the specific case of the ability and/or fexibility/willingness to adapt to OOD scenarios. ....1/3
4/4-Intelligence therefore is an attribute of a person/entity that conveys/quantifies the person’s/entity’s ability to effectively and efficiently reason to attain/achieve understanding.