Series Jev is a new decision-focused AI model announced by ChatGPT co-inventor Diogo Almeida. This post separates the announcement from social media claims, explains why the approach is fast and cheap, and shows how I reproduced the same kind of setup with NerveReflex ideas and the open model Gemma 4 26B.
Insight Running Claude Code autonomously for long sessions and chaining multiple sessions together made AI safety in the age of Recursive Self-Improvement feel concrete to me. Using the recent incidents at Anthropic and OpenAI, I look at why monitoring the chain of thought or sandboxing alone is not enough, and what a realistic layered defense looks like.
Insight Recurrent Depth is reported to be used in OpenAI's next model, Astra. I look at what it means to keep reasoning inside the model's internal representation instead of writing it out as words — for inference cost, for AI safety monitoring, and for NerveReflex, the architecture I have been building.
Insight Lessons from building HR and finance AI agents in the field as a forward deployed engineer: the 20% AI / 80% deterministic design, the limits of general-purpose assistants, and why coexistence beats full automation.
Series On Atari Bowling, TANREN beat R2D2, the human benchmark, and DQN, then hit a ceiling at 74% of the strongest RL. This is the record of establishing — by measurement — that the ceiling was structural to the class of readable reactive code, and of how that loss produced the entry gate that now guards every experiment.
Series In Breakout, evolution discovered the classic tunnel strategy without being taught it and reached about 6.1x the human benchmark. Just as important, the reason it could not reach the strongest RL agents was pinned down by measurement, not speculation — drawing the boundary of what a few dozen readable lines of code can do.
Series Cache eviction — deciding what to drop when a cache is full — has been studied for more than twenty years. On that ground, a 53-line evolved Python function beats ARC, LIRS, and W-TinyLFU on future segments of Twitter's production traces. This post covers how the first attempt lost by overfitting to one time period, the two-window training that fixed it, and the public code anyone can rerun.
Series A score of 31.9 looked like a DQN-beating result — until re-measuring under the published side's exact protocol dropped it to 28.6. This is the record of withdrawing the number, repairing the verifier, and re-running the evolution under the correct protocol to reach 31.2: past the human benchmark and DQN, at 96% of Agent57. An article about evaluation honesty.
Series On Atari Skiing — where Agent57 needed roughly 78 billion training frames to surpass the human benchmark — about 130 lines of readable Python scored −3310.7 in a held-out evaluation, beating every published RL agent and the human benchmark. Includes the entry gate that estimated the odds before running, the techniques evolution added on its own, and the cost contrast: one day and a few dollars.
