Philip Kung

unfortunately i'm back to opus 4.7 -- no idea if i'm getting gaslit by the codex TUI but there seems to be no improvement between 5.4 and 5.5 for coding related tasks.

4.6 -> 4.7 feels like a larger change than 5.4 -> 5.5

@kwindla had a ton of fun!

Sub-agents in (latent) space! We’ve been working on a side project. As far as I know, this is the first massively multiplayer, completely LLM-driven game. Come play Gradient Bang with us. See if you can catch me on the leaderboard. This whole thing started because I wanted to explore a bunch of things I’m currently obsessed with, in an application of non-trivial size, that felt both new and old at the same time. So … a retro-style space trading game built entirely around interacting with and managing multiple LLMs. Factorio, but instead of clicking, you cajole your ship AI into tasking other AIs to do things for you. Some of the things we’ve been thinking about as we hack on Gradient Bang: - Sub-agent orchestration - Partial context sharing between multiple LLM inference loops - Managing very long contexts, and episodic memory across user sessions - World events and large volumes of structured data input as part of human/agent conversations - Dynamic user interfaces, driven/created on the fly by LLMs - And, of course, voice as primary input If you’ve been building coding harnesses, or writing Open Claw agents, or doing pretty much anything that pushes the boundaries of AI-native development these days, you’re probably thinking about these things too! This is all built with @pipecat_ai, the back end is @supabase, the React front end is deployed to @vercel, and all the code is open source.

@NielsRogge what's the storage tradeoff with late-interaction models? from my light research a couple of months ago, it seemed like they required at least an order of magnitude more storage than single-vector dense embedding models

Very cool to see MixedBread include our newly introduced MADQA, an agentic RAG benchmark, in its results Shows that late-interaction models are a a lot better than bi-encoders, but it will take time for the industry to adopt these

very impressed with composer 2.0, great work @cursor_ai team

I’m increasingly convinced that only Gen Z and boomers have taste - millennials should be banned from touching the UI

Puffin ❤️❤️❤️

@Ishaank1999 HUGE! Congrats Ishaan!!!

By the way this happened!

@OpenAIDevs spark has been down for me for the past 3 weeks, plz fix

Proportionately speaking, a handful of factories in China can manufacture as many Shaheds in a day as the entire country of Iran can in a month.

@thsottiaux hey @thsottiaux, have you seen any errors related to 'reasoning.summary' is not supported with the 'gpt-5.3-codex-spark' model? i've been unable to use the spark model through the codex cli since friday evening :(

Codex is back up. We introduced an issue that caused requests to be blocked and rejected as high cyber risk. This was fixed in ~8 mins and we are now back online and fully operational. Team will reset rate limits, that’s been a while. Enjoy one on us.

supabase support: 'did you know you have to be on enterprise plans to have a working DB?'

supabase has been fraudulent for years btw

thanks opus

@chongdashu It already exists, run /statusbar

Claude Code users -- do yourselves a favour an add the remaining context to your status line. Codex CLI has it. Gemini CLI has it. Cursor has it. No reason you shouldn't have it. Here's mine ``` npx @chongdashu/cc-statusline@latest init ``` Or ask Claude to vibe code one for you

An incredible year for turbopuffer is closing. We now manage trillions of vectors and tens of petabytes. We’ve puffed up revenue 10x and headcount 5x. We’ve helped Anthropic, Cursor, Notion and others connect every petabyte to AI to ship the most ambitious version of their product. This week we doubled down with Lachy Groom and added Thrive Capital to the team. They have both proven themselves repeatedly to us over the past 2 years. In 2025, we’ve proved the puffer architecture is capable of state-of-the-art vector and text search. Next year, we’ll continue to optimize the core and widen our surface area, to make it effortless to search every byte.

@iamgingertrash if anthropic or openai existed in a vacuum, people would pay a cent per token.

If AI demand is overwhelming, Why don’t AI companies increase prices? This maybe a simple question, But upon thinking deeper You’ll realize the truth; AI demand doesn’t exist at the current cost of compute It only exists when subsidized You’re being told a Lie

why is git down? are we unable to build reliable software nowadays 🤣

i'm going to guess anthropic's API revenue is down at least 8% post the release of composer