Jason Liu

RT Boris J’adore: - the actual film grain - meeting OpenAI taste 🐐 @jxnlco - Startup grinds with @maxjendrall - Life journey talks with @rhiannon_io DevDay trough the lens of Charlie

RT David Walter (Shiny Gen) No cables! @modretro Chromatic running wireless local multiplayer Tetris. This is a sneak peek for everything coming Monday, when you'll be able to try this out and more.

Not all the wood of a temple needs to be unknotted. You can use the best pillows at the entrance and the walls. As they worse. You move them to the back. And worse you use as shingles and roofing. Knowing next season they may be used as firewood. This is true of software and slop as well.

This happens even at labs I think

when is premier pro mcp?

RT Keyan Zhang we made a simple thing, people internally *really* like it, and we hope you like it too the little composer that could, from planning the launch with @sharifshameem and @andhudhow:

RT Tibo Day 5/ Composer predictions in the desktop app. Often leads to a double take with how on point they are. Included in the Pro plans without consuming usage.

RT OpenAI Developers Now in beta: composer predictions in Codex for Pro users. Codex can now suggest your next message based on your conversation and how you talk to it. One of the most loved new features we've ever tested internally.

RT DHH My friends at @modretro just sent me the new Codex edition of the Chromatic Gameboy that's ready for vibe coding games. How freaking cool this looks 🤩

RT OpenAI Developers The attention to detail is a little concerning.

Ready player one

https://chat.com

anyone a freediver in Monterey area? I'm trying to do a few kelp forest dives this year

RT Rabi Shanker Guha ChatGPT launched their version of Generative UI yesterday. We took it apart tldr; the agent never write raw code. It emits DIL (a new language), the server compiles and streams it, the client renders it using a component set so that it feels native https://x.com/i/article/2108220775568404480

i lacked the spells and incantations these past couple of days, i’ve been trying to do a little more cad design, making things i eventually want to 3d print. i didn’t lack the ambition. i didn’t lack the desire to make something and will it into existence. i just didn’t have the spells and incantations i needed to make it at the level i wanted to. i lacked the vocabulary. and for the first time in a while, i really had to learn. i had to learn about what makes 3d printing possible, how to make parts more efficient and durable, and how to think about the real world. i had to develop the words to describe curvature and shape, and what i wanted those things to feel like. it’s made me appreciate things so much more. i’m starting to understand why the curves on apple devices are so beautiful, and how much work has gone into them. i appreciate the physical space around me more, and what it takes to build the things in it. i keep coming back to this idea that ai really is magic. we’re all magicians and mages now, but we still have to learn the spells and incantations. the more i work with it, the less i believe that the advent of ai will cause us to stop learning. for me, it’s been the opposite. having the ability to make something has made me want to understand it more deeply. language isn’t unique to humanity, of course. but i think our ability to find the words for what we want is going to carry us further now. if you have the courage and the ambition to want something, you can start learning the spells and the incantation to bring it into the world.

Intelligent UI slaps. 1) scheduled task finds a workout slot for me 2) shows me the workout with a visualization. 3 saves it to a space.

devday 2026: from single player to multiplayer agents

RT Shiva Kintali Here is a Short Proof of the Quasi-Riemann Hypothesis. My goal here is to substantially simplify OpenAI’s proof and focus only on proving the Quasi-Riemann Hypothesis for some 1-c, without worrying about maximizing the value of the c. I obtained a much shorter proof for c=1/48. This proof is easy to understand and it can be taught in math courses in one lecture. It is self-contained. It is only 21 pages including abstract, proof overview, all proof details and references. Paper is on my homepage: https://shivakintali.github.io PDF file: https://shivakintali.github.io/papers/QRH.pdf CC: @AlexKontorovich @__alpoge__ @jdlichtman @cremieuxrecueil @danielmarinq @stevenstrogatz @kenono691 @littmath @nasqret @PI010101 @nechita_ion @Isaac__kim @FryRsquared @wtgowers @MarcusduSautoy @DrEugeniaCheng @JSEllenberg @JohnDCook @JDHamkins @AxlerLinear @emilyriehl @hollykrieger @Kit_Yates_Maths @PoShenLoh @robertghrist @AlexKontorovich @kenono691 @ch_nira @rrwilliams @lreyzin @ChrisPeikert @fortnow @RadishHarmers @thomasfbloom @abhatt2 @mahdi_tcs_ @octonion @prz_chojecki @thegautamkamath

my dev day talk is live enjoy! I make peace signs cause people were taking my photo, not cause im super awkward on stage let me know how it was, i have not watched this yet. https://www.youtube.com/watch?v=aDTPTwnrRyA

RT Doug Colkitt We are publishing an update to OpenAI problem #109 Integer multiplication) with another substantial further tightening: κ = 2⁻⁵⁹ (from OpenAI’s original κ = 2⁻¹⁸²) Approximately 500 thousand fold improvement over our previous result and a 2¹²³ fold improvement over the original OAI result. The latest redesigned the finite network to share intermediate computations and scratch space, then tightened the recursion and Gaussian estimates.

RT Andrew Chen it turns out the hardest part of generative UI isn't generating the UI. models have been able to generate good-looking HTML for a while. the more interesting challenge is getting it to feel like it actually belongs in chatgpt. one of our big bets with Intelligent UI was that you didn't have to choose -- that you could keep ~all of the power of HTML and still have it feel fast and native, like it was always part of chat the other half of the problem is model judgement. how do we post-train gpt-6 to know when an interface would actually help? what should it show or let you do? how do we make it useful and beautiful without taking over the conversation? none of this was obvious when we started, and there's plenty we're still figuring out. luckily, one of my favorite things about openai is that if you believe in an idea, you can just go build it, even if it takes a lot of prototypes to get there really excited this is finally out, and incredibly proud of the team. onwards!

RT OpenAI GPT-6 and Intelligent UI, now rolling out in ChatGPT for everyone. Intelligent UI in ChatGPT delivers fast, interactive answers that make everyday questions more visual, complex topics easier to grasp, and interactive tools for your task available on the spot.

gf lost her bike and then found it again i designed a discrete AirTag mount for her bike that also acts as a reflector can you tell how it works?

So embarrassing that someone asked me to demo dots, and I basically asked it what I had to do this week. It told me I needed to book 10 plane tickets for the next 50 days of the year. And rent three cars. Give me my travel schedule and call times for two production shoots in two different cities, then two talks in two other cities.

When codex fast

Anyone using send cut send with Codex would love to talk to you

RT Tibo Completely unrelated to the rest of the ships today, but our models did solve some important math problems and the team documented it in the following blogpost. Remarkable new era of scientific progress. https://openai.com/index/sharing-ai-progress-in-mathematics/

RT OpenAI Developers We’re making it easier to qualify for higher OpenAI API rate limits. Five paid usage tiers become three: Build, Launch, and Grow. You can qualify for Grow, our new highest tier, with $500 in total API payments—down from $1,000 for the previous highest tier.

RT Tibo Day 2.1/ We have made Auto-review free for all users signed in through a ChatGPT account. You can enable it in settings > permissions > auto-review. Auto-review improves upon the default sandbox setting that requires you to approve everything, which is prone to decision fatigue unless you spend a lot of time configuring specific rules. It allows you to run long tasks while having a second agent review all actions taken by the primary agent. Its only goal is to prevent high-risk actions from being taken and to protect against unwanted actions that are not aligned with the original user intent. This Auto-review feature is now free and does not draw usage from your plan.

RT Tibo Day 1/ We have optimized the default speed to be ~50% faster across GPT-6 Astra and GPT-6.1 Sol through the subscription across all our products and partners using Sign in With ChatGPT (including OpenCode, Pi, Amp, Devin, ...). No changes needed on your end and this should be felt within the next two hours.

RT Peter Gostev I know nobody cared about this at OpenAI dev day, but I think this is one of the biggest potential future product lines. So far all agents assume there's a human who is taking responsibility & prompting it. One clear way forward is to have an agent that takes full responsibility for something - e.g. accounting or email marketing, same way like an employee would. The reason why you hire people is not just they are good at what they do, but they'll take responsibility and make sure things happen. That's why you are ok hiring people even with zero experience in that function - the important part is that someone is thinking about it and it's not all on you. What happens now with agents is kind of a disaster - since you can do so much more, you end up going into more and more areas. But guess what, the agents aren't responsible for anything, so everything new that you are doing is 100% your responsibility. So yes agents are helping, that amount that you need to worry about grows enormously. It is also not realistic to expect everyone to learn to use agents effectively, Excel has been around for 30+ years and what's the % of people who have any clue how to use SUMIFS or something simple like that. AI is even harder to grasp, since model capabilities & products are changing every month. If AI adoption relies on people learning how to use agents en masse, we are screwed, it is just not going to happen. So instead what we need is agents that just work. Imagine you are a starup founder, you need an accountant. Instead of spending 1/5th of your headcount on hiring an accountant, you pay $30k a year and get an AI accountant that would just take care of things - it would monitor your work, proactively tell you what's wrong and have weekly meetings with you to discuss issues, like you would with a human. I have no idea how well Specialist dots will work, but it is the first mainstream agent that you don't need to actually prompt. It is possible that it ...

RT Joanne Jang A group from Anthropic was at our neighborhood bar last night and tipped $0 on a $500 tab. Apparently Claude told them that tipping would be endorsing the alcohol industry and the bartenders' choice to participate in it. The bartenders were so pissed they made Claude name every shrimp in the kitchen, then sent it photos of them deep-fried

RT Adit Your life as a founder will be much better once you internalize that (almost) everything that isn’t going well is ultimately your fault

Autistic Chinese dad as been achieved internally.

RT rajan agarwal reply to this tweet with some sloppy code you’ve seen our models produce! what’s your biggest pain point with code quality?

RT Tibo Fortunately our future models will be much better at code deletion and simplification. Just in time

“These demos are unrealistic” Yeah cause you spend every waking hour on Twitter.

RT cova i was speaking to a musician friend who had the most interesting reframe of ai i’ve ever heard he said: “ai is you, ai is me, it’s the culture of the internet and everything we were and know. by definition, ai is who we are.” is there a way where its not just 'machine art' and 'human art' but... just art.

RT Matthew Berman I've had 4 GPT 6.1 Sol Extra High threads running for 20+ hours straight and I still have 85% of my quota left. I'll never get to use my resets....

I think they said this about coding too

who is going to ai engineering new york this month???

Saw a guy with a Thinking cap in Mission Bay and started shouting, "What hood you from?"

How do I get access to the X credit card?

Is this a vague post.

COVAAAAA

RT cova me and @jxnlco in a group dm thats grown to 50 people

RT OpenAI Developers dots demo. Now... with better WiFi.

should dots be a seperate app?

Your reminder that the codex app came out in February

RT Thomas Ricouard We had so much fun building the ultrafast game with @codestantine. Once again, blown away by Astra’s capabilities. People could make their cars live by just prompting them using GPT-Live and getting custom 3D models in less than 10 seconds.

How did the Dev Day Game Boy idea start? There’s something pretty special about making a game and then holding it in your hands. I’m glad we got to share that feeling with everyone there and wanted to recognize @ko_kuramoto and his team as part of how this started. The first inspiration came from seeing Ko Kuramoto’s inventions and exploring ideas with him and the nanu team. I loved the tinkering: taking familiar hardware and asking what else it could do. We talked about connecting Codex to a Game Boy through a remote-control bridge. https://x.com/ko_kuramoto/status/2042959996041023792?s=20 That part never came together, but I really enjoyed those early conversations. @dkundel had the idea to do something for DevDay. @DylanAbruscato pushed for every single person attending to get one and @thsottiaux needed little convincing. That took us from tinkering with a concept to figuring out how to make thousands. We ultimately worked with @modretro to bring it to DevDay, just as these models were getting really good at making games!

How did the Dev Day Game Boy idea start? There’s something pretty special about making a game and then holding it in your hands. I’m glad we got to share that feeling with everyone there and wanted to recognize @ko_kuramoto and his team as part of how this started. The first inspiration came from seeing Ko Kuramoto’s inventions and exploring ideas with him and the nanu team. I loved the tinkering: taking familiar hardware and asking what else it could do. We talked about connecting Codex to a Game Boy through a remote-control bridge. That part never came together, but I really enjoyed those early conversations. https://x.com/ko_kuramoto/status/2039679565086462247 @dkundel had the idea to do something for DevDay. @DylanAbruscato pushed for every single person attending to get one and @thsottiaux needed little convincing. That took us from tinkering with a concept to figuring out how to make thousands. We ultimately worked with @modretro to bring it to DevDay, just as these models were getting really good at making games!

Love this idea that I can just suggest my favorite restaurant and DDoS it with 100,000 agents for bookings and never be able to go to that restaurant again.

she straight up looked at me and said, "I looked sick." i go "thank you. It's cashmere" She then looked dead in my eyes and said "No, you look ill." i go "thank you its dev day"

to be clear you can do almost everything in dot/space with vanilla codex for example before those tools i just had * 15 pinned tasks each with a heart beat automation * Github repo for a memory vault based with agents.md files everywhere so all your agents can contribute * make sure both your technical and non technically team know about Github and can clone the repo and contribute based on my rules I just had 4-5 different mini apps served to render artifacts and then do a bunch of 1:1s to set up the rest of your technical and non technical teams to use this workflow to collaborate and make sure their automations are set up on their computers or you can be like 'yo dot whats new, lmk'

Can someone make a calendar app for codex. I’m off today

My dot

A year ago I was in the audience of cowell theater thinking about whether I could join OpenAI Now I’m on stage in that same theatre. Thanks to everyone who made a dev day such a great experience. I want a world where AI does my laundry and does my dishes so I can make art And all these tools have made it possible Or at least the work that feels like laundry in the work that feels like art.

LMA-O

RT Gabriel Chua I worked on the dots live demo. Holly did a phenomenal job on stage today. If you’ve ever been there, you know the pressure is immense. She handled the recovery SO well. I flew to SF from Singapore for my 4th DevDay (and first an OAI employee). It’s been a joy to get a small glimpse of what the dots team has been cooking. Truth be told, I was slightly skeptical of using dots when it was first introduced. I was already really comfortable using Codex and setting up plenty of automations. I wasn’t sure how much more I’d get out of it. But a few personal examples that I can share publicly changed my mind. 1. En route to SF last week, my rideshare cancelled on me. There was the generic automated refund email. My dot immediately picked it up, knew I had a flight, and asked if I was okay. 2. During this trip, after days of DevDay prep, I woke up and told it, “Fuck my life, I’m so tired - what’s on my cal today.” It walked me through my day, then added one last thing: “Please take some time to eat and drink. Take one step at a time - you got it”. Alone in a hotel room, that felt really reassuring. 3. On the building side, I’ve been using dot to spawn Codex threads that use CUA and ultrafast, and to proactively scan PRs for changes that might affect the demo. It’s been useful for getting work done, and increasingly insightful and high-signal in ways I hadn’t expected. I could delegate more confidently without repeating the same context for the umpteenth time. 4. Even this post came together with my dot, as I spoke to it throughout the day to investigate what happened and emotionally process it. Hope you give it a shot and see what I mean :)

Moved to SF in March. Can’t wait to experience SF tmr.

Dots and space and multiplayer.

RT Pauline P. Narvas I’ve done a lot of cool things in my career, but moderating the closing DevDay Q&A with these legends is definitely up there. Thank you, @sama, @thsottiaux, @tejalpatwardhan, @CoryDecareaux, @romainhuet and our community for the trust 💙

15 minutes until my talk at cowell at 3pm!

devday you can find me after my talk before i black out 3pm at cowell

RT Tibo You can now use your ChatGPT subscription directly in over 16 partners products. No little rules, you can just use all your included usage right there. This includes things such as Devin, OpenCode, Notion, and many more. Sign in with ChatGPT and go.

RT Sherwin Wu Use your ChatGPT subscription usage in all your favorite tools! People have been asking about this ~basically since ChatGPT subs launched, now it's finally out!!

If you are at DevDay and want to meet up, come to my talk its the last one at cowell theatre! I'll be talking about my transition to a multiplayer experience within chatgpt, dot, and space. and i'll be around that pavillion to meet folks afterwards!

my dms instantly went from devday ticket requests to modretro chromatic requests

RT cova what's in the box?

RT Iheanyi Ekechukwu It's easy to shit on people for live demos going wrong, but I got mad respect for OpenAI doing live demos vs. faking it. The ultrafast demos have been crazy. Imagine when this speed is normalized?

RT ChatGPT ChatGPT Space is your new home for creating and collaborating with your team and AI. In Space, you can work with pages—a new kind of interactive document with charts, images, checklists, and dashboards. ChatGPT can build pages for you based on the work you need to get done or the context of your conversation. Edit pages directly with your team, or tag ChatGPT, Codex, or your dot to help with action items and keep pages updated from your connected sources. Rolling out today to Pro, Business, and Enterprise plans. Create and edit pages on web and desktop; view on mobile. Collaborative slides and other formats are coming soon. https://chatgpt.com/features/space/

RT Thariq Re 4/ "In order to have taste you need to eat" - credit @jxnlco

RT OpenAI Your dot is ready to meet you.

just signed my openai speaker release form, see you at dev day

RT Ed Bayes Re @benjitaylor codex, may 2025

RT OpenAI Get ready.

https://github.com/567-labs/instructor/discussions/2706

do people want to use jev with instructor?

can jev play overcooked?

RT Atty Eleti 🇯🇵 Grandmas go global! 日本語でChatGPT Voiceを試してみて！ https://youtu.be/htpj3ujwsvc

RT cova wip

Yes, we are aware that Codex is down.

RT Atty Eleti Do you use live video in ChatGPT? What's your favorite use-case?

anyone got a lighter?

RT Gabe Greenberg Just had my 8th surgery..... i've been having nerve pain when sitting for 4 years... and after ten different doctors (who missed the diagnosis) we finally traced it back to hamstring tendinosis. the tendon was 40-50% diseased because of hip surgeries starting 17 yrs ago. @ChatGPT helped me figure out the issue myself and get to the right MD. i've had 4 hip surgeries (freak ice hockey accident from 20 yrs ago) and a few surgeries to try and fix the nerve pain. I'm very grateful to have this "solved" i was sick from 2012 to 2019 with mercury and mold toxicity.... it's been one thing after another with my health. but I'm going to keep sharing out loud about the recovery process. I have a wonderful family..... the tech community here on twitter has continued to support me. and God has been there for me in some very dark moments. hope to get back to rock climbing one day.... but for now time to rest and recover.

RT Daniel Fradin my best friends

Whereas I mostly blame feature flags

musicians: how do you feel about AI transcribing songs into sheet music or tabs to help people learn? curious if you’ve used it, whether it’s helped, and how you think it affects learning to play by ear.

been using astra to transcribe music, then to review the spectrogram to do error correction and then review the music theory resulting in very robust transcriptions for me to learn... >The spectrogram shows a few clear problems: some wide vibrato became extra chromatic notes, and around 3:22 the tracker briefly jumped to a lower piano note while the clarinet continued above it. I’m correcting those and simplifying isolated sixteenth rests. The quiet ending is still ambiguous.

so cool i asked dewalt 6 to make me a table

RT notch I will admit, I am enjoying vibe coding. I may have been a teensy tiny bit wrong.

I was taught this lesson at 21 and one of the best lessons in both disruption and sales.

RT Anish Acharya Re @WillManidis 100p my personal favorite is "the best way way to compete with an incumbent is to build something at the intersection of two VPs" they hate each other more than they could ever love the customer courtesy of @stevesi

RT cova day trip @1x_tech

Excess knowledge without action can poison the mind just as excess food can poison the body without movement.

Finally. Summer is over and I can wear my layers

The Waterloo water is white hard. Ruins a water boiler every year and a few shower heads a semester.