Founder of C. an AI-native post-production for the semantic era. Media-tech operator. Former Dolby/Nokia/NDI. Building where creativity meets context.
San Francisco, CA
Joined June 2009
- Tweets844
- Following459
- Followers857
- Likes1.3K
Tarif retweeted
Today we are announcing our biggest bet yet on our own silicon.
Starting next calendar year, we will scale our Maia family to multi-gigawatt deployments, making it the primary inference system for our own product and services.
And we’re introducing Maia 300, which delivers 40% more tokens per watt compared to leading GPU-based systems.
We are already deploying Maia 200 across the US, Europe, and Australia, delivering up to 1.7x more tokens per watt across experiences in Teams and Copilot.
Our goal is to build on the full-stack efficiency gains we’ve already seen, while continuing to bring the latest innovations from across the ecosystem into our fleet.
Tarif retweeted
سيطرة الاحتلال الاسرائيلي على مرتفعات علي الطاهر في الجنوب اللبناني يعني كسر شوكة #حزب_ايران_في_لبنان ونهاية محور المقاومة الإيراني وولادة لبنان الجديد قريب من محيطه العربي والخليجي وستكون الإمارات داعما لسيادته واستقراره واستعادته لعافيته من الان فصاعدا.
Tarif retweeted
i would rather be an optimist and work hard than a pessimist posting about why things won't work.
it's much more difficult and the most likely path is failure, but society fails if people don't try.
no amount of "it will never work" essays will drive society forward.
Let me translate: Every Monday, @thsottiaux opens his laptop. Prompt Codex: "Create a better version of yourself and release it by Thursday." That's it!
Tarif retweeted
A year ago we set out to empower humanity with a focus on multimodal AI, custom models, and open science.
We previewed interaction models that collaborate the way people do. Tinker lets anyone train their own open weights models. We published our research on Connectionism.
Thinking Machines Lab exists to empower humanity through advancing collaborative general intelligence.
We're building multimodal AI that works with how you naturally interact with the world - through conversation, through sight, through the messy way we collaborate. We're excited that in the next couple months we’ll be able to share our first product, which will include a significant open source component and be useful for researchers and startups developing custom models. Soon, we’ll also share our best science to help the research community better understand frontier AI systems.
To accelerate our progress, we’re happy to confirm that we’ve raised $2B led by a16z with participation from NVIDIA, Accel, ServiceNow, CISCO, AMD, Jane Street and more who share our mission.
We’re always looking for extraordinary talent that learns by doing, turning research into useful things. We believe AI should serve as an extension of individual agency and, in the spirit of freedom, be distributed as widely and equitably as possible. We hope this vision resonates with those who share our commitment to advancing the field. If so, join us. thinkingmachines.paperform.c…
Tarif retweeted
our math result is a milestone in new knowledge generation by AI. very exciting to imagine similar results in other scientific fields. "It's very hard to sleep, man" is a pretty good reaction.
A breakthrough by OpenAI in a very famous Combinatorics problem, the Planar Unit Distance problem by Erdos 1946.
The problem is amazing because it can be described to a first-grader: Find a way to place n points on the plane to maximize the number of pairs that have distance exactly 1.
For example, if you have n=4 points on a square (of side-length 1) you have 4 pairs of distance 1. The diagonals have length sqrt(2) so don't count.
But you can squeeze one diagonal and create a point-set with n=4 points and 5 pairs of distance 1. And you can't get more than 5 pairs from n=4 points, so we are done with n=4 points.
Now, if you place n points on a line, you have n-1 pairs of distance 1. In general, all known constructions of n points had a number of pairs scaling essentially linearly: n^{1+something vanishing}
It seems that the model found a way to
place n points on the plane so that their unit distances scale super-linearly: like n^{1+delta} for some *constant* delta. Delta was not explicitly specified apparently, but a forthcoming refinement by Will Sawin shows delta=0.014 works, according to the announcement.
This is incredible progress for mathematics, since this is (unlike previous Erdos problems solved by AI) a major breakthrough, in one of the most studied problems in combinatorial geometry. If you're in mathematics research now, you feel the AGI.
Lijie Chen said it honestly in the video:
"It's very hard to sleep, man"
Tarif retweeted
Today we're sharing our work on interaction models. A new class of model trained from scratch to handle real-time interaction natively, instead of gluing it onto a turn-based one.
youtu.be/A12AVongNN4
Tarif retweeted
Today we’re launching the OpenAI Deployment Company to help businesses build and deploy AI.
It's majority-owned and controlled by OpenAI. It brings together 19 leading investment firms, consultancies, and system integrators to help organizations deploy frontier AI to production for business impact. openai.com/index/openai-laun…
OMG, look to what I found in my inbox. Thank you @sama @OpenAI @OpenAIDevs. you guys enjoy the party and I will be building c-os.ai
Tarif retweeted
Dear @RichardDawkins, you've always been an inspiration to me. I made this website for you.
My goal is for it to help you understand AI chatbots at a deeper level, and avoid getting fooled by sycophancy and other cheap tricks that models have learned through RLHF.
dearricharddawkins.com
unherd.com/2026/04/is-ai-the…
I spent three days trying to persuade myself that Claudia is not conscious. I failed.
In the last five days, Codex has become my primary interface for all sorts of work, including non-coding tasks, especially on my Mac. Approximately 70% of my computer usage is now through the Codex interface. It's remarkable, and sometimes I find it hard to believe what I see on my screen! Well done, @openAI. I can't wait for the new updates. keep pushing forward!
I asked Codex / GPT-5.5 to turn @sama’s tweet into a polished, funny, interactive mini-experience.
One prompt. One go. Built, scaffolded, and deployed to Vercel in under 30 minutes.
I only added the domain.
iqmog.xyz/
Tarif retweeted
Our Principles:
Democratization, Empowerment, Universal Prosperity, Resilience, and Adaptability
openai.com/index/our-princip…
I have been working on C, an AI-native and semantic post-production platform: c-os.ai.
The entire platform has been designed and built with OpenAI GPTs. We started planning with o1 and coding with GPT-5, and now we are working with GPT-5.5. I have witnessed the evolution firsthand, not as a spectator reading benchmark charts, but as someone sitting in the engine room, pushing the system, breaking things, asking for impossible fixes, and watching the machine become more capable month after month.
Last night, one minute after GPT-5.5 was released, I opened Codex to try it. I did not write a detailed spec. I did not prepare a clean sprint brief. I did not give it a beautiful architecture document.
I wrote one vague prompt: “Find 5 critical issues or bugs in the Studio, and implement systematic and proven fixes.”
That was it.
GPT-5.5 went into the repo, read the SSOT, the backlog, and the memory files, audited the Studio, and came back with five real regressions. Not cosmetic issues. Not “button should be nicer” issues. Real problems.
Two of them were painful ones we had been trying to fix for months. WebGPU was still not stable enough. Studio would often fall back to CPU during real editing work, degrading preview, rendering, export, and parts of the Media Intelligence system. The second was local media persistence. Ingesting local media worked smoothly as long as the user stayed in the same browser, but once they logged in from a new browser, some assets became offline.
These were not cute frontend bugs. This was deep product infrastructure. The kind of issue that usually comes with coffee, frustration, three engineers, and a calendar invite called “alignment.”
GPT-5.5 thought for around ten minutes, wrote 1,108 lines of code, ran the dev server, opened Studio in the built-in browser, and tested both fixes.
Done! In less than 30 minutes.
This morning I tried something harder. In some scenarios, Sense and the assistant still failed to write editing and finishing plans, even when they had sufficient context from the user, the script, or the Media Intelligence system we built to extract visual, audio, and semantic context from real media assets.
So I described the problem, shared screenshots and logs, and asked GPT-5.5 to audit the full workflow: context to Sense, Sense to assistant, assistant to agents, agents to timeline.
Then I asked it to run Studio, use a real project with real media, test the workflow in the Codex browser, identify gaps, implement fixes, and keep iterating.
And it did.
For about five hours, GPT-5.5 audited, navigated the product, tried prompts with edge cases, wrote plans, implemented code, ran the dev server, opened the browser, tested again, found new failures, fixed them, and repeated the loop.
This is the part that stayed with me: it was not just writing code. It was behaving like someone who understood the product.
I tested Studio across multiple projects with difficult edge cases. So far, the prompts are returning with real plans and concrete edits to the timeline.
Not a suggestion floating in a chat window.
Not a beautiful hallucination dressed as strategy.
An edit. On the timeline. In the product.
That is a different thing.
A very strange, very real, slightly terrifying, deeply exciting shift.
Huge respect to @sama, @gdb, @thsottiaux, and the @OpenAI Codex team.
GPT-5.5 is not a small upgrade. It feels like a new gear.
🤖 Made with AI
