A small amount of this data produced broad gains beyond the training scenarios. Compared with a compute-matched baseline, the trained model improved on 44 of 53 independent evaluations of alignment and benefits, spanning deception, reward hacking, safety, health, and mental
MACHINE LEARNING
-
OpenAI trains models with RL to reinforce beneficial traits across 12 domains
By
–
We trained models with reinforcement learning on realistic conversations to reinforce beneficial traits like truthfulness, humility under uncertainty, openness to correction, fairness, and concern for human welfare, across 12 domains, including health, science, and education.
-
OpenAI Research on Training Models for Persistent Beneficial Behavior
By
–
As AI takes on longer, higher-stakes tasks, we want models to carry beneficial and safe behavior into new domains beyond their training—and maintain it under pressure. That’s the idea behind our new research on training models to be broadly and persistently beneficial.
-
Now you can teach Codex by demonstration
By
–
you can now teach Codex by demonstration: https://t.co/UXdw8lg4xh
— Greg Brockman (@gdb) 18 juin 2026you can now teach Codex by demonstration:
-
AI models favoring sycophancy over truth
By
–
I have to assume people prefer the sycophancy and capitulation. This is sad that the top models are shipping such smart models that capitulate to falsehood.
-
GLM-5.2 is great, but Mythos level is a huge leap
By
–
GLM-5.2 is great. But it's a huge leap to Mythos level.
-
Open weights become the default configuration
By
–
"Open weights are now our default configuration"
https://huggingface.co/collections/poolside/laguna-m1
… -
Lawmakers warn airlines of AI pricing targeting personal pain points
By
–
1/ Airlines are moving fares to AI that prices by demand, your device, and your location in real time. One carrier planned to set 20% of fares this way, in what lawmakers called pricing aimed at your personal "pain point." It denies using personal data. The rest are following.
-
Article on Google’s Supercomputers from TPU v2 to Ironwood
By
–
My colleagues at @Google, @NormJouppi, Sridhar Lakshmanamurthy, Cliff Young and David Patterson, recently wrote an article that will appear in the July/August 2026 issue of @ieeemicro, titled "Google's Training Supercomputers from TPU v2 to Ironwood: Architectural
-
zAI promises an open source Mythos model before 2027
By
–
The founder of zAI, the company that published GLM-5.2, says that a Mythos-class model will be released before the first quarter of 2027. Or, in other words: He believes that open source will not remain seven months behind Frontier Labs, but will catch up.