@JoshPurtell on Backscroll

5 tweets from @JoshPurtell that Backscroll ranked for substance.

Josh @JoshPurtell 2026-07-02 18:26Z

Cool mental model

Josh @JoshPurtell 2026-06-29 18:27Z

If OpenAI/Anthropic trained 20-50T parameter models, they'd almost certainly give USG and potentially allies *massive* battlefield advantages, and would give labs a peak into what deploying the next level of intelligence looks like.

Josh @JoshPurtell 2026-06-29 18:25Z

Gpt-4.5 was recently decommissioned. It was way too big to be served efficiently on the hardware of the day, so no one really used it. However, it was clearly extremely intelligent, and with decent RL probably would have been SOTA for a while. Perfect fit for military use!

Josh @JoshPurtell 2026-06-28 17:52Z

Why don't we give codex a change_effort tool? If I'm running a 12 hour /goal session, it'd be nice to get xhigh on core code, high on evals/testing, medium on running tests, and low on documentation. Surely it's not that hard for the model to make those judgements?

Josh @JoshPurtell 2026-06-28 04:05Z

How it ranks: Backscroll reads my Twitter/X timeline and ranks every tweet for substance with an LLM taste rubric — not raw engagement. Curated by Surya Dantuluri.