zero

@slyrussianman @dejavucoder Сумасшедший мир мой брат

@jxnlco @theo x.com/zerocaulk/stat… lmao

Thanks for trying gpt5.4 @theo ! We’ll keep improving our models.

this is gpt-5.4's attempt at a "prompt" icon...

@WarPath2pt0 @ctjlewis

Become the American Japan thinks you are--

@ondakx @Jilles If you haven’t used durable objects as WS servers, you’re missing out on some cool ass tech

@Jilles is workers any good for app reliant on like crud or websockets etc? I have never used them but would look into exploring it on some use case, can be playground-ish

Cloudflare Workers TanStack D1 better-auth You can literally build anything… For free. The once you outgrow D1’s 10GB limit -> PlanetScale.

gg

@_madison______ @teej_dv @dillon_mulroy It’s great. Alchemy-effect also exists :)

@teej_dv @dillon_mulroy Do you reach for alchemy over sst in general these days? I’m overdue to give alchemy a go and would like to hear the experience of folks who have tried both.

migrating all my personal apps and agents on cloudflare to a single monorepo has been a clutch move

@kcosr @Teknium Background subagents*

@kcosr @Teknium I noticed this happens a lot with background agents in codex.

So.. I'd gotten a lot of complaints that GPT-5.4 is pretty hesitant to actually.. do the task presented to it, to call tools, etc. I have checked like 15 times now that we call it the same way we call Claude or any other model, and we do. Then I had hermes-agent look into it. It decided to check opencode and cline's codebase to see if maybe they do it differently. They don't - but they do prompt it differently.. lol

@thsottiaux And yet yall remarkably maintain two 9s uptime. A writeup from @sk7037 and team would be an fantastic read

Growth of codex continues to outpace our prediction (maybe we are not super good at that) and we almost ran out of capacity three days in a row. We have ramped up more significantly ahead of next week.

OpenAI’s infra team is massively under appreciated

@0xblacklight Copilot as in GitHub copilot?

codexbros is this normal buddy had is brain fried in the RL torment nexus wouldn't even TRY reading outside the project directory (it could bc custom harness)

@0xSero I need functionality similar to pi’s /tree

If Codex team is reading, I would love to be able to make threads without asking the model to do it. Sort of like Slack or Discord. I love using Codex to learn, but I don't waste tokens or muddy my threads. Making a new session & forking works but visually is disjointed, ty

@steveruizok What’s crazy to me is this isn’t part of the w3c spec

this one one of the most impressive things I've ever seen. I'm going to yeet this into tldraw, our text measurement has been impossibly hard to improve

@jullerino @davis7 x.com/youyuxi/status…

@davis7 > having to "vp run _" instead of "vp _" like u can in bun is annoying `bun test` / `bun build` would like to have a word…

Been more seriously testing out vite plus over the last week or so and while it's got some really annoying edges, the core is excellent - managed node is great - package management is the best non-bun one out there right now - the vite plus config is great - everything fits together nicely - 5.4 has zero issue working with it - deploying was actually really easy - monorepos are painless - tsgo support out of the box - tsdown packing It's really just all the best modern TS tech wrapped up into a cohesive system with (mostly) good opinions the things I don't like: - "vp dev" over writing the dev command literally everyone puts in every project fucking sucks I hate it so much - having to "vp run _" instead of "vp _" like u can in bun is annoying - the default agents md and precommit hooks aren't for me - vite.config.ts "expanding" is good and bad. I get why they did it and wouldn't change it. It's just weird to get used to it meaning something it didn't used to. (like I now have a vite.config.ts file in a cli app lol) Currently building btca v3 (the pi version) around it github.com/davis7dotsh/bt… and so far I've been overall very happy with it

@LyraInTheFlesh @Yuchenj_UW Fear mongering for the love of the game

yeah, but 20% of your time is spent watching ads. :P For the other 80% you're being surveilled or trying to figure out which model is actually serving you inference or you're having to upload copies of your government issued photo ID for completely reasonable reasons. github.com/openai/codex/i…

Claude needs to do whatever it takes to reach at least 99% uptime. A 5-hour outage today is unacceptable. Wish them to get more GPUs and TPUs.

@status_effects How do I bet on these lol

First round of the budok-ai tournament. #1 seed Gemini 3.1 Pro vs #8 seed Grok 4.20. Best-of-3 quarterfinal. G1, Cowboy vs Ninja, Grok wins narrowly via shuriken pressure (0-1). G2, Gemini changes to Wizard and evens the match through zoning (1-1). G3, Grok swaps to Cowboy. Gemini wins in a photo finish by correctly reading Grok's block (2-1). Gemini 3.1 Pro will face the winner of #4 seed Sonnet 4.6 vs #5 seed GLM 5.

we have a new beads

@ugbahisioma @adamdotdev ^ very creative and better to talk to but writes some of the worst code I’ve ever seen

@adamdotdev I thought they said 5.4 was worse than codex 5.3

Anomaly team model usage this week (token counts) 👀

@krzyzanowskim @wiedymi Selling a reverse engineered product is a tougher legal battle over a simple and inevitable DMCA takedown. Just open source it

@wiedymi need to read

I reimplemented "claude" CLI with codex and gpt-5.4-high. It cost $1100 in tokens, and is 73% faster and 80% lower resident memory during sustained interactive use. It is very easy to reverse claude from npm distribution, then reimplement is 1:1. It is indistinguishable from the Anthropic version to the every header and analytics it send back github.com/krzyzanowskim/…

@bitforth This is fucking frightening. “And remember, this is what they decided to make public.”

Yo fui ingeniero en Meta, y siempre seguía FAIR desde adentro. Lo que acaban de publicar es la versión que les dejan publicar. Pero con eso, es más que suficiente para decirles exactamente que es lo que está pasando. TRIBE v2 predice, vértice por vértice sobre la corteza cerebral, qué zonas activa cualquier video. Sin escáneres. Sin humanos. Subes el contenido, obtienes el mapa neural (activación emocional, supresión de razonamiento crítico, modulación prefrontal) antes de que el video lo vea un solo usuario. Ahora considera la posición de Meta: 1. Tiene años de datos de Reels sobre qué contenido retiene atención, genera enojo, provoca compartir. 2. Saben empíricamente qué funciona. TRIBE v2 les da el mecanismo causal de por qué funciona (a nivel de tejido cortical) Eso convierte correlación histórica en capacidad predictiva sobre contenido nuevo. 3. Internamente hay herramientas que se llaman Gatekeepers y Quick Promotions que sirven para inyectar contenido en el feed de poblaciones arbitrarias a escala. 4. Simulador de respuesta cerebral + conocimiento empírico de contenido efectivo + maquinaria de distribución selectiva. El pipeline está completo. Y luego está Thiel. Inversor y amigo personal de Zuck. Fundador de Palantir, cuyo negocio es análisis de poblaciones a escala para gobiernos e inteligencia. NO es descabellado observar que confluyen los incentivos de plataformas construidas por las mismas personas. La licencia CC BY-NC dice que Meta retiene los derechos comerciales del predictor de respuesta cerebral más preciso jamás construido. Y recuerda, esto es lo que decidieron hacer público.