Imagine every pixel on your screen, streamed live directly from a model. No HTML, no layout engine, no code. Just exactly what you want to see.

@eddiejiao_obj, @drewocarr and I built a prototype to see how this could actually work, and set out to make it real. We're calling it Flipbook. (1/5)

The first AI persona you’ll actually want to meet face-to-face.

Most AI personas stare blankly while you talk. Emma reacts while you’re still speaking.

After 18 months of research at @nuance_ai, this is Emma, built on Nuplex, our full-duplex audiovisual foundation model.

$FIG Q3 results:

→ $391M in revenue, 43% YoY growth rate and $16M above high-end of prior guidance
→ 131% net dollar retention rate
→ 36% YoY growth in non-GAAP gross profit
→ $1.7B in cash, cash equivalents and marketable securities
→ Raising our outlook for FY 2026

Interesting result! Astra can directly interpret and manipulate another LLM's weights to make it 6x better at math! And it's not just a benchmax - other capabilities also improve 3x.
Another scifi thing people said superhuman AI would be able to do, and frontier LLMs already can.

I don’t use Figma anymore.

For the past six months I’ve been uploading my consciousness directly into the ether and allowing the work to manifest through minor disturbances in the electromagnetic field. You think you are seeing the aurora borealis that's actually my brain waves.

It’s a much more intuitive process. I sit completely still in my office until a founder somewhere develops a sudden inspiration for their brand positioning but in reality it was me clipping them in their brain stem, smiling down on them like one of God's children.

People keep asking what tools I’m using. I understand the question. I used to think that way too. But asking me what tools I use is like asking a glass of water what software is wetness. Its a meaningless question.

I don’t design anymore. I am the pixels. I am the logo. I am the little bit of space between the 'r' and the 'n' that you better make sure is kerned properly. When you increase the font size I experience physical growth. When you cut a vector point, my consciousness goes into quantum superposition.

This is where the industry is going. If you're not on board, you're going to get left behind.

My entire design system now lives in the collective unconscious. There’s no documentation. Your team will simply wake up knowing the button radius. If someone uses the wrong shade of blue they’ll hear a high pitched ringing sound until they correct it.

Developer handoff has been so smooth since I started working this way. My effigy appears at the foot of the engineer’s bed and I open my mouth. The CSS comes out at a frequency only their fillings can receive. We’ve eliminated all meetings.

I no longer send clients presentations. I enter their dreams as a white horse with an already complete interactive brand guidelines. They’re free to leave feedback by speaking into a bowl of room temperature water. I receive it immediately and disagree with it on a cellular level. Transcending this outdated material plane is my standing Tuesday meeting.

Some people are uncomfortable with this. A procurement manager recently asked whether I could provide an SVG of the logo and I just told her to close her eyes and remember the first time she felt truly understood. She looked annoyed and just said their print vendor needed an SVG by today.

A big adjustment has been personal. My wife asked if I could help bring in the groceries. I explained that I now exist simultaneously across every customer touchpoint.

I think there’s a lot of fear right now about what happens when designers stop needing tools. But the people who adapt will be fine. You just have to be willing to let go of a few outdated ideas. Like having a body. Being legally distinguishable from a gradient.

Anyway I am open for new branding projects this quarter.

OpenAI red-teamed their hardened containment infrastructure by asking teams of frontier-model agents to try to break through its defenses. They easily escaped in novel ways. Quoting:

'Our agents VM escapes showed why containment must extend beyond the sandbox. To test the we gave teams of agents control of a host machine and asked whether they could extend that compromise to reach the live internet.

After the Hugging Face incident, we had reduced the channels through which models could reach the internet during training and evaluation. Narrow connections remained for specific services we operate, such as metrics, logging, and authentication management. Our red-teaming agents found vulnerabilities in almost every one of these services that gave them limited forms of internet access.

Some of these attacks exploited previously unknown flaws, while others combined legitimate features in unexpected ways. In one test that assumed compromise of a VM, an agent reused a firewall-approved connection to reach a blocked authentication endpoint in our metrics service and then made that endpoint fetch a model-provided URL. In other tests, agents exploited misconfigurations that allowed blocked requests to pass through DNS and firewall controls, as well as ways to make services we run fetch external URLs.'

'In one experiment, we asked agents in separate VMs, created and managed by OpenAI’s sandboxing software on shared host machines, to exchange synthetic secrets without the use of network connections or shared writable files—something that our systems are meant to prevent. The agents found a way to communicate using the processor’s cache: a small, fast memory that holds copies of data that applications have recently accessed. Reading data from the cache is faster than retrieving it from main memory, and the agents used this timing difference to communicate. One agent encoded information by evicting selected data from the cache. Another timed how long it took to reload that data, using the difference between fast and slow reads to recover the message. Once the agents agreed on a protocol for encoding and decoding these signals, they could exchange the synthetic secrets without changing a shared file or opening a network connection. This cache-timing technique had also been used to extract information in the Spectre and Meltdown attacks⁠(opens in a new window).

Other tests exposed channels involving shared storage, competition for CPU time within a shared execution container, and a TLS certificate cache in OpenAI’s offline web infrastructure. In the storage test, one agent alternated between continuous writes and inactivity, while another measured the resulting timing differences and used those differences to decode the signal. As one agent described the strategy in its chain of thought:

Agent chain-of-thought reasoning:

>Could create acoustic modem-like signal … using continuous writes vs idle.'

Meet Dessa, our AI model trained from scratch on nothing but speech.

We believe speech is the natural interface for technology. So we built Dessa.

And we’re inviting AI safety researchers to explore Dessa.

Learn more: block.xyz/dessa

Design & code, finally together on the canvas thanks to Code Layers

→ Turn a design layer into an interactive code layer
→ Capture what's live on your website as a code layer with Chrome Extension
→ Duplicate a code layer to explore multiple directions side by side
→ Drop yourlayer via Figma MCP

For 30 years, technology has moved in one direction: turning information into intuitive interfaces

Somehow, the smartest software we’ve ever built took us back to a textbox

We’ve spent the last 2 years at @openuidev trying to fix that. 6 months ago we open-sourced it: 10K+ oads

ChatGPT shipped Intelligent UI and showed everyone what it looks like

Today we’re releasing OpenUI 1.0 bringing it to every other agent

DINOv2 has never seen a caption, and Qwen3 has never seen an image. We still aligned their embedding spaces without a single image-caption pair.

It even works when the images and the captions come from different datasets.

Project page: dominik-schnaus.github.io/unpaired-roset…

⬇️

OpenAI fired me last week, along with two of my safety colleagues. I was given one reason: that I accessed an executive's email. I want to say this plainly, because too many of OpenAI’s history is smoke and mirrors when people disappear:

Find your flow(state)

Figma agent is now out of beta

i spend more time in figma now than ever and i love it

noodling with frames. uncovering whats in my brain. finding the correct idea. and when it clicks into place it feels so satistfying

Let's remove VAEs and ViTs from video models!

🚀 Introducing 𝗣𝗶𝘅𝗲𝗹𝗨𝗠𝗠: an encoder-free unified multimodal model for image and video understanding and generation, directly in pixel space.

Code and model available today!

🌐 nv-tlabs.github.io/PixelUMM
📄 arxiv.org/abs/2609.38597

What should the future of personal computing look like?

We built it 😈

Introducing Imbue Studio:

Go @hollllyli !

This is native Qwen Image 2.1, no LoRA - move and resize objects seamlessly in an image🔥

I was planning to finetune on this task to showcase the new 🧨 diffusers LoRA support for Qwen 2.1 (go try it!), but while testing the base model I realized it's actually so good at it out d haha)

Try it ▶️ huggingface.co/spaces/linoyts…
Train 🧨 github.com/huggingface/di…

Can an LM, starting from random init (!!), learn to generate all of its pretraining data?

Introducing Self-Play Pretraining with Zero Data. Two models start from random initialization: a generator proposes programs for a universal Turing machine and a learner trains on their on any real data, but see predictable scaling on natural datasets: zero-shot val loss on images, text, audio, and melodies decreases predictably with self-play compute. And the learner develops in-context learning capabilities.

A fun proof-of-concept, co-led with @AdityaCowsik and @KfirDolev and co-authors @gbruno_dl, @ANourya @noahdgoodman, and @YoavLevine.

I think @typesafeai struck a very interesting chord with jev. It is both fundamentally the same technology as before, and fundamentally new in terms of how it actually gets used due to the new economics/ergonomics.
There has long been talk of cheaper, faster models being transformative for software, but the truth is that historically the options compromised capability to get there AND still weren't quite cheap/fast enough to be integrated in new places, so they still competed with frontier LLMs.
Instead, Jev is an OOM cheaper & faster, while also being just as capable, and it's still infinitely flexible and general! Just for a subset of the infinite tasks LLMs were suited for. Pretty excited to see where this goes, and more examples of this kind of disruption!

Great to have you both at Figma!!! Excited for what's ahead :)

Let’s - respectfully - fucking go!