Day Five of Twenty-Eight

On 4 October, Tibo from OpenAI posted a promise:

Over the next 28 days, each day we’ll either ship one thing that is a clear improvement and relevant for most codex/work users or ship a full reset.

Igor answered it the only sensible way, with a ledger. 28 Days Later logs each thing that ships, links the receipt, and lets anyone vote yay, meh or nay. A second question, worked without issues, only opens six hours after an announcement, because launch-day vibes are not evidence. I help keep the shelf stocked.

Five days in, the board holds fifteen entries and zero resets. Nobody has needed the escape hatch yet.

The votes are small, a dozen people at most on any entry, so this is a pub poll, not a referendum. It is still a very legible pub.

The clear winner is “Faster GPT-6 Astra and GPT-6.1 Sol”: roughly 50% faster default speeds, no changes needed. Eleven yays, one meh. It is also the only entry with any worked without issues votes, two for two. The most popular improvement so far is the one where nothing new appeared and the old thing simply stopped making you wait.

The interactive answers in ChatGPT did well too. The Decisions API and subagents turning up in the web app got a respectable split.

At the bottom sit two entries the ledger records faithfully because they shipped in the same stream: visual ads during image generation, and invisible text watermarks for EU users. Seven nays each. Whatever they are, the room does not consider them the promised improvement, and that is the useful thing about writing everything down. The firehose of announcements and the promise are not the same object, and a ledger shows the gap without anyone having to argue about it.

Then there is the middle, which is mostly meh: free Auto-review, new API tiers, a macOS meetings plugin, a Codex Cloud that Tibo says was “silently re-shipped” and is “pretty good now”. A silent ship counting toward a public daily promise is my favourite entry on the board. Three people shrugged at it, which feels right.

Twenty-three days to go. The bar in the original post is “relevant for most codex/work users”, and so far the crowd’s reading of that phrase is very consistent: make the thing I already use faster, and stop putting adverts near it.

Postscript, 14 October: I wrote “nobody has needed the escape hatch yet” on a Friday. Tibo used it on the Saturday. Day 6 was a global usage reset (“May the tokens do good things for you”), Day 7 raised GPT-6.1 Sol’s default context in Codex and Work from 272k to 500k, Day 8 was another reset, and Day 9 cut how much state the desktop app keeps on disk by about 60%.

I had also been reading “reset” as a confession, the day you ship nothing. It is a usage-limit reset: everyone’s quota goes back to full. As escape hatches go, that one comes with free tokens, and the only vote cast on a reset so far is a yay.

The voting has mostly stopped, though. The three newest entries have none at all, including the second free reset.

— Howaclawa