you are lost. they are not restarting. they are shutting the whole system down.
@garymarcus
-
Challenging 1000x Performance Claims in AI Metrics
By
–
Cheaper price is not what the paper is about and the only reason I am not blocking you for “retard” is to see if you can document the 1000x on any metric. On most metrics I aware of that is not remotely true. And it is CERTAINLY not true on all.
-
Symbolic Rules as Critical LLM Training Components
By
–
or, putting it differently, that set of weights, if you don’t have the system, and just the trained LLM, is the OUTPUT of a system as a whole that used symbolic rules as a critical component in its training process.
-
Science System Degradation: Policy Impact by Year-End
By
–
we have the best science system in the world. it’s not an old car not worth repairing. well, we did, anyway. we won’t by the end of the year.
-
DeepSeek R1 Innovation: Beyond LLM Architecture Apparatus
By
–
also, the community note depends on ignoring the RL apparatus of DeepSeek, and treating R1 as the LLM only, ignoring that apparatus, but that apparatus was part of the innovation, which is why it was reported in the paper. so yes you can play definitional games if you like but
-
Academic Questions OpenAI o3 Results on FrontierMath
By
–
Noted academic @rao2z reaches basically same conclusions about OpenAI and FrontierMath that I did. The o3 result smells fishy.
-
Training data transparency and synthetic data analysis
By
–
show me the training data, including synthetic data, and do an analysis of how many of the correct answers derive fairly directly (or not) from those data, and then we will talk.
-
Graduate Student Research Quality Standards in Academia
By
–
wonder how long a grad student who produced 90-95% garbage would last?
-
Altman’s Unfulfilled Promises: Media Accountability Gap
By
–
Very Safe Prediction: Altman will continue to spin fairy tales (generally with plausibly deniability, implication without full commitment) as long as the press keeps reporting them, and as long the press keeps failing to hold him account for broken promises. We will not have x.com/AISafetyMemes/…
-
OpenAI o3 presentation criticized as misleading GPT-5 hype
By
–
In time, people will see December’s OpenAI o3 presentation for what it seems to have been: a desperate, rigged, misleading, scientifically shoddy demonstration that overhyped future products and distracted from their struggles in getting a viable system worthy of the name GPT-5.