I feel this way about gpt-5-nano (EOL December 2026). It seems like the open weight models have progressed a long way since these old models were released though. Deepseek V4 Flash is even cheaper than gpt-5-nano. I'm still going to pay a cloud provider to run it for me, I'm not local inference pilled yet, but I _can_ run it myself in the future if worse comes to worst.
Objectively testable evals are one thing, but how does one judge whether a new model is adequately reproducing the subjective "writing style" of an old model that you've gotten accustomed to the feel of?
Denver has this... nominally. 3 machines (2 in circulation, one is a "Display"). 4 week checkout period. 103 current holds. 103*4/2/12 ≈ 17 year wait time.
Oh, didn't read that part of the news. That's great. Ability to run x64 docker images seminatively was one of the big reasons I jumped to the M1 platform when it came out and I was baffled that they would remove it.
Plaid wants you to enter your bank username-password into their form. If it was just routing+account it would be truly no different than other bank connection methods.
After reading this and remembering an old hobby project, I decided to switch the deploy from a systemd service to PM2, which apparently has rolling deployments without needing Docker engine (for those of us minmaxing instance RAM).
>I read the Fogbank story and recognized it immediately. Not the nuclear material. The pattern. Build capability over decades. Find a cheaper substitute. Let the human pipeline atrophy. Enjoy the savings. Then watch it all collapse when a crisis demands what you optimized away.
>In defense, the substitute was the peace dividend. In software, it’s AI.
Before it was AI, the cheaper alternative was remote contract dev teams in Eastern Europe, right?
For real work I'm already on Opus full time (plan+execute). I think there's a different space for cheap-but-good-enough options for small-to-med repo hobby projects. (Not enough usage to warrant a plan.) And Opus is only ~30 tok/s while grok-code-fast-1 is ~120. So planning with Opus and executing with grok-code-fast-1 is still on the table for me - it's good at following explicitly written plans, lightning fast.
If it's between a human or an AI copywriting SEO slop, I'm happy to see an AI take that job. SEO content marketing is so painful to read once you realize you're reading it, and I have to imagine it's as painful to write if you're a technically talented writer.
"It is the first time since 1972 that humans have travelled outside of the Earth's orbit." But they're not tho (Earth's gravitational dominance extends 4x the distance to the moon)
Objectively testable evals are one thing, but how does one judge whether a new model is adequately reproducing the subjective "writing style" of an old model that you've gotten accustomed to the feel of?