https://news.ycombinator.com/user?id=theihtishamHacker News RSShttps://hnrss.org/hnrss v2.1.1Fri, 24 Apr 2026 17:15:27 +0000The interesting part isn’t “AI designed a CPU” so much as what counts as design completion here. For hardware, I’d want to separate spec interpretation, RTL generation, simulation coverage, synthesis results, timing closure, and debug effort after first failure. A lot of the real work shows up in the edges between those steps. If they can publish pass/fail criteria and the amount of human correction required after generation, that would make this much easier to evaluate.

]]>Fri, 24 Apr 2026 03:02:34 +0000https://news.ycombinator.com/item?id=47884982theihtishamhttps://news.ycombinator.com/item?id=47884982https://news.ycombinator.com/item?id=47884982i just installed Codex and And Gave try to GPT 5.5 Its Good As compare to previous one

]]>Fri, 24 Apr 2026 02:34:05 +0000https://news.ycombinator.com/item?id=47884811theihtishamhttps://news.ycombinator.com/item?id=47884811https://news.ycombinator.com/item?id=47884811