Sean
178 posts

Sean
@seanmozeik
Founder https://t.co/Prj1VdOwHE, mcp wizard
London, England Katılım Ocak 2026
548 Takip Edilen91 Takipçiler

Smashing lossless audio compression benchmarks (with some help from Fable). Full songs come out average 6.3% smaller than FLAC's best setting , decoding up to ~29x faster than realtime for playback. Not sure what to do with the codec yet 🥸 If you're spending $$$ on lossless audio egress, DM me!
English

I’m joining @OpenAI to help build the Codex app!
I’ve been using Codex all day, every day for months. It’s where I write every line of code I ship and get just about everything else done.
Can’t wait to help shape what comes next.

English

Decode is heavier than FLAC: the ratio advantage comes from adaptive prediction. So like FLAC there are presets, all bit-perfect lossless. The default keeps over 80% of the size win, decodes 20-30x realtime, and its decode path is integer-only. Phone-friendly, and small enough to put on embedded hardware someday!
English

I think about this image a lot when get asked about the state of computer use agents, specifically on the web
everyone assumes that model intelligence will improve massively (which it will), but that’s still taking the approach of intercepting DOM tree level data, taking screenshots, parsing it through an LLM that decides how to move your mouse and keyboard, all of which are human based inputs
the Internet needs to be rebuilt for agents, to complete work faster, cheaper, better and more reliably in a novel way
that’s exactly what we’re doing at AgentWeb. so if you’re a cracked engineer who wants to work on the worlds hardest problems, reach out :)

English
Sean retweetledi

Just met @seanmozeik the winner of the #LondonMaxxing hackathon today at a coworking day and we may be able to combine both our datasets. Mine from creators across Instagram and his from across London.
Thanks for bringing us together today @msefaoruc
Seema Amble@seema_amble
I used Claude to build a day to day vacation itinerary - it actually wasn’t terrible, but unsurprisingly lacked taste. It actually did save me a bunch of time by suggesting day trips, mapping where everything was, pulling train times, made sure things were open, also gave me an initial list of things like restaurants, coffee shops, stores (although it defaults to the most touristy stuff). To be able to ingest Reddit, Instagram and Google data for better recommendations would be awesome!
English

hosted solo builders LDN co-working session today, powered by @eachlabs at Tradestars Islington.
about 15 of us met for breakfast, spent a few hours working on our own things, and then caught up on what everyone was building. we wrapped up the session with baklava.
i really enjoy this morning. it's nice being around people who are all building something, even if everyone's working on completely different ideas.
big thanks to each::labs team and Tradestars Islington for supporting the meetup and helping us make these mornings happen.
see you at the next one. 🙂



English

In Codex, I wish that 'approve for me' was unsandboxed by default with the classifier standing in between the model making stupid decisions (like Claude's auto mode). Too often the model hits sandbox friction and it's really bad at knowing it needs to escalate out of the sandbox! @thsottiaux @ajambrosino
English


@Andy_AJT @OpenAIDevs You’re like Willie Wonka with your golden tickets!
English

we just sent out some tickets for the Codex Community Hackathon in London for @OpenAIDevs Build Week.
Made in Codex with GPT-5.6 Sol Ultra. See you in London.
Sol · Terra · Luna ☀️🌎🌑
#OpenAIBuildWeek #Codex
GIF
English

@Andy_AJT @OpenAIDevs Sol · Terra · Luna ☀️🌎🌑 solterraluna.dev/ticket/?name=S… #OpenAIBuildWeek #Codex via @Andy_AJT
Español

@seanmozeik @OpenAIDevs check your email, let us see if its fixed ;)
Islington, London 🇬🇧 English

Notes from LDN Cursor Cafe:
- Everyone hates Anthropic
- Cursor has great brand (+ no users)
- AI continues to suck at systems design + writing
- Grok 4.5 is great
- Do more to make your codebase work for LLMs
- People will do anything for a merch hat
- London AI energy is great
Ben Lang@benln
We're bringing Cafe Cursor back to London on July 11th Grab coffee & merch, co-work, and meet the team
English



