Pinned Tweet
some of these posts have reached a few hundred thousand to a few million views, but most days this account is still a small room. i don’t want the bigger numbers to make me colder toward the people who actually take the time to write.
so i’m saying this publicly.
if you’re building something real, an open source tool, a project, anything that’s meant to help people, send it to me. i’ll run it, tell you the truth, help where i can, and share it if it’s good. i won’t charge you. nobody should have to buy attention just to get honest eyes on their work.
if you’re new to agents and you feel behind, send the questions anyway, even the basic ones, even if there are a lot of them. i might be slow because i’m human, but i won’t leave you hanging on purpose. i learned this the long way and i don’t want that to be the only path.
i also won’t sell the take. no paycheck, partnership, or opportunity gets to change what i think about a product. the people who’ve been here for a while already know that. that trust is the whole point of this account.
i’m putting this here so you can hold me to it if i slip.
thank you for trusting this page enough to grow it. i know exactly who made that possible, and i’m not going to forget.
Kevin Rajan retweeted
Can we pause for a second and think how crazy it is that someone can get 99% of the benefits of blender without even opening it... People can simply mention "blender" to Claude Code and it will just use Blender... a thing that used to take years to master... just effortlessly controlled by agents now.
The Fable of Clawd
A 2 minute 4K film made entirely in Blender, from one prompt on Fable 5.2. It took about 4.5 hours.
Claude built every asset in Blender by writing Python:
- 5 characters and props: Clawd, his friend Pip, a toy truck, a 3D-printed tulip and a gift box
- 7 full sets: a study, a staircase, a street, a workshop, an airport, a night sky and a house
- every camera move and every frame of animation across 52 shots
That came to 12,500 lines of code. It started 7 copies of itself to build the sets at the same time.
Then it rented 89 GPUs through RunPod to render 3,220 frames as fast as possible, 36 of them running at once. Total cost: a little over $20.
My biggest takeaway is my Claude Code knows it has access to cloud compute, so it's much more ambitious.
I tried a similar prompt on my other computer that doesn't have the runpod skill setup and it wasn't nearly as high quality
Full details linked below.
Gotta say, Jensen's been an awesome supporter of Nous Research, our efforts in open source, and Hermes Agent since we first met, as has all of Nvidia.
In general, I think Nvidia has been the largest and one of the few big companies who has championed open source AI. I hope there's a great resolution to all this and wish the best to him and Nvidia as a whole!
♥️
Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me.
I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA.
I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation.
Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO.
Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours.
Jensen
Kevin Rajan retweeted
Hermex 2.2 is out 🚀
Your agent can show its work: HTML reports render right in the chat, and SVG and Mermaid code blocks open as pictures and diagrams.
Also new: Connected Apps, a bot widget, readable model names, Persian.
Free and open source: apps.apple.com/us/app/hermex…
Kevin Rajan retweeted
I’m honestly blown away by what GPT-6.2 Astra can do 🤯
In 36 minutes and for $5.31, it built a whole pixel-art jet game in a single HTML file. You can hit the afterburners, barrel-roll over a lake, switch to cockpit view, and fire missiles
I hopped in for a quick test flight. I’ve lost count of how many laps I’ve done
Some of you have been asking very politely. Some of you have just been asking.
GPT-6.2 Astra and Sol are now available.
Super efficient and built to perform where it counts—the real world.
openai.com/index/introducing…
hey @pewdiepie why does @AnthropicAI get to ban distillation, when they distilled copyrighted & pirated content to make claude?
bro they 100% are nerfing astra, this video was made prior to the launch & nerf. the first week of astra was a golden era
made prior to astra launch, someone pls test this prompt today and show us the slop it makes
teapawt.net/sharifshameem/status/2…
GPT 6.2 just came out, they showed this as what gpt 6.0 did in Blender (left), but when GPT 6 astra came out they showed this (right) as what it could do in blender.
What...? They are either nerfing tf out of gpt 6 over time or they are really doing some strange marketing moves..
3
170
Kevin Rajan retweeted
Introducing : Voltron-1 Mini.
A small, compact coding agent.
Voltron has been post-trained for coding and mathematics knowledge, using @liquidai's LFM 3 1.7B as a base.
Voltron delivers better performances across several like benchmarks like livecodebench ( substract of it ), AIME 2026, HumanEval and a few other benchmarks.
This is the first of many models I'm going to do in this class. I've been working very hard on my RSI pipelines, and have managed to build something very promising which I will use to improve my models.
This means more, better checkpoints are coming.
Also, this model was trained 80% of the time using a new, more efficient LorA algorithm, research paper dropping soon, people with consumer GPUs are gonna love it!!!
🤗HF link : huggingface.co/Akahsizrr/vol…
( next release is the inference engine guys btw ).
Kevin Rajan retweeted
to start getting your CI under control, it's important to first understand there are a few different types of validations that exist
1. the code follows our rules
2. this change works
3. everything in the main branch works
4. the product is ready to release
traditionally, many codebases run all 4 on every PR as part of their CI pipeline. when every dev merges ~1 PR a day on average, that's doable
but as more and more people can merge 10s or 100s more PRs a day with effective use of agents, their CI starts to struggle. and buying more CI minutes or using faster CI is a losing battle
the solution is to
- run only #1 and #2 on each PR
and #1 (mostly compile and lint) is typically fast enough that can run as pre-commit hooks as well
on each PR's CI pipeline, try to select only tests that are being edited and relevant to the change, because this is a high frequency bucket and you want to keep this extremely fast
consider assigning a fixed budget such as 5 minutes - whenever the pipeline exceeds that budget, ask an agent to optimize and get it back under the threshold
- run #3 using your merge queue or as a post-push rolling CI pipeline
this is typically your full test suite. running them on every PR at agent speed is simply too much overhead, but you also don't want to fully stop running the full suite, because some problems only occur when multiple changes combine
if you do this post-push, it means your main branch can have problems from time to time. counter-intuitively, that's okay. problems will exist no matter how many tests you run in your CI
when something breaks, fix forward. agents can do this very quickly. and you have a record of which commits pass the full suite - only create releases from those commits
- manage #4 using feature gates and phased rollouts
typically, you should get to a state where anything that could pass #3 is safe to release
in practice though, some problems will only happen with real traffic, so if you are running something at scale, you'd want some gates based on real telemetry
so the pipeline you could create is to release good commits to say 5% of your users, and progressively rollout further only when data looks good
once you have a clean separation of these 4 categories of gates, it'll be easy to know how to optimize and how to make tradeoffs between them
glide-music-picker.vercel.ap…
feedback & help welcome!
Sound blend and feel visualized, what?!
This touches my audio engineer beginnings in a way I can’t explain.
2
105
i am soooo hype @Teknium, when are we building our game?
What if three.js could render beyond the limits of the browser?
Meet ThreeRT: a native JS runtime and hardware-accelerated real-time path tracer, bringing lifelike lighting to Three.js scenes on macOS, Windows, iOS, Android, and Steam Deck.
Fully open source 🧵
3
96
Kevin Rajan retweeted
introducing o8
agents write code faster than i can read it, so i built the thing that reads it for me
a second agent reviews the real diff
the agent that wrote the code never approves it
only the approved commit merges
works with claude code, codex and six more
free and open source ⬇️
Kevin Rajan retweeted
Watch the U.S. skies clear out in a 34 second timelapse in God's Eye View.
Pulls together public ADS-B, geostationary satellite imagery, FAA advisories and ATC audio.
- Blue = planes in the air
- Orange = planes on the ground
- Rings = aircraft piling up at each airport, with the count beside it
What we know based on advisories and tonight's White House briefing:
- This is the longest nationwide ground stop since 9/11.
- Started at Chicago O'Hare at ~3:10 PM ET with "comms issues"
- Soon after radar & radio instability spread to air traffic control centers nationwide
- Wasn't a complete blackout. Sec. Duffy says old analog switches and routers got overloaded, leaving controllers with intermittent radio and radar, so the FAA stopped departures.
- DHS Sec. Mullin says patches installed during an equipment upgrade are the primary suspect. Essentially "the new technology… was overwhelming the older system."
- No evidence yet of a cyberattack, though DHS hasn't ruled out either yet. Notably, Sec. Duffy says the FAA's new AI routing tool (SMART) didn't cause it.
- More than 3.5K planes were in the air, and every departure nationwide was stopped at 3:54 PM ET
- The stop was lifted at 8:19 PM ET after about 4.5 hours, with 6.6K+ flights canceled and 8.6K delayed (per Flightradar24).
This is a wild 4D reconstruction to see. Much respect to the controllers and pilots who kept everyone safe while working with degraded capabilities.
Kevin Rajan retweeted
I don't know man, developing things in the AI age is also super fun because you can do this :)
Kevin Rajan retweeted
Hermes Agent Desktop from @NousResearch lets a theme be more than colours.
I put a live landscape behind the whole app: aurora over mountains by night, a pink and teal print by day.
You spend all day in this window. It might as well have a view.
Meet Borealis.


























