member of technical staff @openai, cofounder @aidkit_tech, former head of sync @dropbox, cofounded bubbli (acquired by dropbox), previously made yelp monocle.

sf
Joined March 2007
I've engineered a hand-crafted Transformer to do long-hand addition. All of the weights were hand-chosen by me. It is perfectly accurate. How? Fall down the rabbit hole here: colab.research.google.com/gi… Short on time? Here's a summary 👇 1/
20
154
17
1,150
314,827
Picked up an old hobby project because Codex is AGI. Yesterday it built a 2.5GHz "spectrum analyzer" from a SERDES block on an old PicoEVB FPGA & worked out all the PCIe mumbo jumbo. Sees everything from WiFi to FM, all at once! This would've taken me months. For Codex? 12hrs
3
4
26
1,727
Full disclosure: after it figured this out, it ran compaction and then realized that it was building from partially cached artifacts and could not reproduce the final bitstream. So it then spent the next 12 hours tripping over itself to get things back. I make it use git now.
3
417
I've been off and on involved with this for what feels like forever, and am so excited for it to see the light of day! This is the beginning of a whole new era of human to model communication.
Introducing GPT-Live, a new generation of voice models for natural human-AI interaction. Rolling out in ChatGPT starting today. You’ll want to turn the sound on for this one.
4
41
4,590
Ben Newhouse retweeted
the most predictive trait for successful research these days seems to be excessive carefulness, bordering on paranoia. so easy to make bugs, so hard to find them agents, so far, mostly make this more difficult
19
34
3
625
27,960
I’m hiring on my team at OpenAI for Research Engineer / Scientist and Software Engineer roles. I believe one of the most important questions for future AI systems is: how do we train them to help people over time? We work on that through RLHF, post-training, reward modeling, long-horizon evals, and the data infrastructure behind personalized multimodal AI. If you care about human flourishing and serious research, links in comments.
95
85
3
1,978
179,946
Hello wayward person on the internet trying to figure out why your UniFi router is convinced that your ISP (Starlink) is offline when it definitely isn’t. UniFi’s PHY link detection is overly neurotic—if you put a switch between things everything will work great. You’re welcome.
4
8
1,047
With no exaggeration—this is some of the most societally important work at OpenAI right now.
We're taking ChatGPT privacy to the next level. OpenAI is putting together a founding team to bring advanced encryption-based privacy to ChatGPT, the OpenAI API, and our future consumer devices. If you are an expert in TEEs and E2EE, or a product engineer who cares deeply about privacy, reach out!
9
1,458
I've been driving the research for our collab with Jony Ive from w/in OpenAI for the last year, alongside the best team imaginable. Want to join us? Search for "future of computing" here openai.com/careers/search/! PS: cover letters help me filter a ton :) nytimes.com/2025/05/21/techn…
10
24
5
351
55,096
Ben Newhouse retweeted
It's deeply concerning that one of the best AI researchers I've worked with, @kaicathyc, was denied a U.S. green card today. A Canadian who's lived and contributed here for 12 years now has to leave. We’re risking America’s AI leadership when we turn away talent like this.
384
711
273
8,590
2,579,514
Second ship of the day 📈 We're building ChatGPT into a contextual assistant. If you're wondering why @openai is building macOS and PC apps—this the beginning of why.
ChatGPT 🤝 VS Code, Xcode, Terminal, iTerm2 ChatGPT for macOS can now work with apps on your desktop. In this early beta for Plus and Team users, you can let ChatGPT look at coding apps to provide better answers.
29
46
8
831
182,965
Remember how in the TV show Severance, the main characters spend their entire workday staring at numbers and deciding if they're "good numbers" or "bad numbers" and we don't know why? I've figured it out... They are debugging numerics issues in large language models.
1
1
16
2,477
(also help, I am also in the Severance office basement rn doing the same)
1
5
1,089
everyone wants to join a rocket ship until they remember they’re bumpy and on fire. buckle up y’all, only way to space.
23
71
13
891
158,927
The early chain of thought papers are what lead me to conclude that AI was going to work this time and MAN IS IT WORKING. This is an incredible achievement by even more incredible people 🙌
We're releasing a preview of OpenAI o1—a new series of AI models designed to spend more time thinking before they respond. These models can reason through complex tasks and solve harder problems than previous models in science, coding, and math. openai.com/index/introducing…
13
1,732
So excited for this -- begone wrapping all your LLM calls in a while loop until the model decides to follow your schema. If you're building on the API, it's hard to a imagine a use case where this wouldn't benefit immensely.
Introducing Structured Outputs in the API—model outputs now adhere to developer-supplied JSON Schemas. openai.com/index/introducing…
2
22
2,646
RT @giertler: the first time gpt-4o responded to me in real-time, it was clear that we built something completely new – and that what we ar…
3
93
Ben Newhouse retweeted
Fun fact: @barret_zoph moonlights as a table
SHE CALLED HIM A WOODEN SURFACE 😭
4
5
111
29,738
i am so excited about javi's excitement
im so excited about bogo’s excitement
1
16
6,890
I don't know who needs to hear this but: You can replace that old monitor you keep around to plug into linux servers w/ a $30 HDMI -> USB Capture dongle and Quicktime.
1
11
2,524
this is a good use of GPUs 😅
Excited to introduce SHA-21B: a new open-source language model which can do exactly one thing! calculate an SHA256 hash :)
3
3,144