Introducing GPT‑6 Cyber: a major advance in security research that pairs stronger capabilities with better model alignment.
The model powered the Linux kernel research we announced yesterday, and security teams at companies like Mozilla, Palo Alto Networks, Calif, and Cisco—alongside our own at OpenAI—are already putting it to work.
GPT-6 Cyber is available to select Daybreak Red customers and we are expanding access over time.
openai.com/index/introduc…
18 of the 20 best open models today are Mixture-of-Experts. Dense is basically over.
And yet the know-how to train a good MoE still lives mostly in the heads of a handful of teams.
So we spent 800k GPU hours running the experiments ourselves and wrote everything down. Super happy to finally release "The Ultimate MoE Training Handbook", free and open, organized around the questions training teams hit:
- how many experts, how big, shared or not?
- why is half my step spent in all-to-all, and how do I hide it?
- why is my run 25x slower with no error in the logs?
- softmax, sigmoid or sqrt(softplus) router, and why keep it in fp32?
- aux loss or DeepSeek's loss-free bias?
- upcycle a dense checkpoint or train from scratch?
- can I serve with fewer experts per token than I trained with?
⚠️ Warning: it's long (~70k words, budget a few days, sorry) but with 55 interactive figures, code you can run and numbers from our own runs.
Read it here: huggingface.co/spaces/hugging…
Models are becoming smarter, faster, and cheaper, but the OCR industry hasn’t kept up.
We’re launching r-1 mini: the first model to offer truly premium parsing without the premium pricing. At $1/1000 pages, r-1 mini beats options that cost up to 25x as much.
Today we're launching Executor v2
It's a massive overhaul to the product, finally unlocking every use case people have wanted from the beginning
You can think of v2 as if Zapier / Retool were fully drivable as code, with everything being exposed to your agent
Meet Gemini Omni 1.2 Flash, our updated video generation model from @GoogleDeepMind 🎬
Get stronger consistency, more creative control, and faster, more cost-effective video creation at scale:
⚡ Fast & 30% lower cost: Draft at 360p or 720p ($0.07/s) and upscale to 1080p or 4K
⏱️ Longer clips: Generate and edit videos up to 15s, or pick any exact length from 3s to 15s
💬 Edit in conversation: Swap backgrounds, subjects, or lighting turn by turn without starting over from scratch
🗣️ More natural speech: Create fluid, multi-speaker dialogue with clearer lip-sync
🎬 More controllable shots: Direct continuous camera moves with fewer jump cuts
See these updates in action ↓
introducing gemini omni 1.2 flash, now available in ai studio and the gemini api
what's new:
- fast & 30% lower cost
- longer clips
- edit in conversation
- more natural speech
- more controllable shots
Introducing Inkling 2 Small. Scoring 31 on the AA Intelligence Index, it's a big jump over its predecessor.
Inkling 2 Small is available for fine-tuning and inference on Tinker today, with full weight release planned soon.
Today we are rolling out Gemini Omni Flash 1.2, with:
- longer video generation (15 seconds)
- deeper cinematic controls for camera pov
- support for multi-turn conversation editing
- less visual drift
- 30% lower price than Omni Flash 1.1
Read more:
I’m all for championing excellence over mediocrity, but it is a mistake to imply that academic achievement is the only measure of it. Jocks and prom queens aren’t the antithesis of merit and striving, and in some ways, their arenas are more “real” than academic curriculums. Fight passivity and low standards!
GPT-6 Cyber: a major advance for security research, already being used to help defend security-critical software such as the Linux kernel, Firefox, and many other projects.
We’ve invested a great deal in trust, specifically training the model to stay within the scope specified by the operator. The corresponding eval results are quite striking.
We’re also launching Daybreak Red for Critical Infrastructure, giving approved critical infrastructure defenders the ability to conduct authorized testing of critical systems.
Introducing GPT‑6 Cyber: a major advance in security research that pairs stronger capabilities with better model alignment.
The model powered the Linux kernel research we announced yesterday, and security teams at companies like Mozilla, Palo Alto Networks, Calif, and Cisco—alongside our own at OpenAI—are already putting it to work.
GPT-6 Cyber is available to select Daybreak Red customers and we are expanding access over time.
openai.com/index/introduc…
Watching my sons learn to write has explained a lot about why good writing is rare. They're both convinced they have to produce bloated word salad when writing essays in school, and that if they wrote simply and clearly they'd get bad grades. The sad thing is they may be right.
Anyone who says a social network wants bots is deeply mistaken. Human interaction is already scarce on social networks, since most people are passive consumers.
Bots displace this scarce resource:
Each bot reply means one human reply is not seen or engaged with by another human.
It’s nearly a 1-to-1 drop in product & business value—and likely has much bigger compounding losses when you account for people churning out of frustration.
LIVE: Watch as @astro_anil and Joshua Kutryk of @csa_asc install the final roll-out solar array on the @space_station during a spacewalk that will last about seven hours. The solar arrays will boost power to the orbital lab. x.com/i/broadcasts/2…
Flying high!
Today, our X-59 quiet supersonic X-59 aircraft traveled from Birmingham, Alabama, to Daytona Beach, Florida. The aircraft aims to fly faster than sound with just a gentle thump rather than a sonic boom — helping usher in a new era of faster air travel.
We're prepping to feature X-59 at MAX POWER and will provide updates soon!
Google says that if schools want to use Google Search, they need to use its AI too.
Here's a better solution: Don't use Google.
Brave Search is a privacy-first search engine that doesn't force you to use AI.
Google is now threatening America's schools, saying they will turn off their core search functions if schools choose to entirely block AI.
nytimes.com/2026/11/03/us/…
Introducing Cekura Bench: the world’s leading Voice AI Benchmarks
We benchmarked 50+ models from openAi, Google, Nvidia, xAI, Inworld, Gradium, Assembly, Fish audio and more
Now you can:
→ Compare naturalness, performance and cost
→ See how the best voice agents are being built.
→ Find the best fit for what you are building.
Our tests cover Speech-to-Speech models, Speech-to-Text models, Text-to-Speech Models, and Voice Agent platforms. We recorded real humans doing tasks in different domains and created the most representative voice benchmark out there!
Explore the benchmarks: benchmarks.cekura.ai
We would love to collaborate with more teams on these benchmarks. Reach out if you are interested.
Our open-source ML Intern is now available as a Slack bot you can install in your own workspace!
Evaluate models, build datasets, and fine-tune models on the fly 🚀
YOUR data deserves YOUR own, private model.
Now, let @HuggingFace's ML Intern build it for you.
Connect your data, files, mcps. And call it from Slack.
How we can secure 50% of the world's software.
Traditionally, formal methods have been mostly used for safety-critical systems like operating systems, hypervisors, and databases. We're extending this to formally and empirically verify code correctness and security for full-stack web apps, one of the most common types of applications*, using the Proof-Driven Development Kit (PDD Kit).
Coding agents now write code faster than anyone can review effectively. PDD Kit turns what the developer meant into a Verus specification. It then proves the AI-generated code meets that spec for every input.
Researchers @alexlavaee, @MatthaiP39719, @Mappletons, and @k_cieslak break down Proof-Driven Development x Chopin in the blog here:
*44.4% of respondents work as full-stack (27.3%), back-end (13.5%), or front-end (3.6%) developers. Stack Overflow Developer Survey 2026, n = 25,767.
Figma CFO Praveer Melwani says his corp dev team now tests a target's moat by trying to vibe code its product, then runs the M&A memo past AI versions of Figma's leaders:
"I think our team's gotten really damn good at some of this. You can get through some of the, you know, 'Is this a real thing? Is it not? How difficult is it or not?' There's obviously nuances to a whole bunch of these ideas, but it gives you like a solid intuition there."
"More often than not, there are themes, there are similar questions, there are ways that we all as leaders tend to have a way of looking at problems. And what's been interesting is [the corp dev team has] gone and created unique [AI] personalities for each of us as leaders... that they can go and run their investment memos by."
"They'll sit there kind of going back and forth and saying, 'Okay, this is what Praveer would say, and this is what [CEO] Dylan would say, and this is what Sean would say,' our CRO... It reduces the cycle time that we're all sitting together on."
@figma@seema_amble@IvanoMakSF
Figma CFO Praveer Melwani used to spend his days writing hundreds of lines of SQL just to get one set of numbers everyone trusted. Now his finance team builds its own AI tools to do that work.
He sees the CFO's job shifting from traffic cop to business enabler, as long as the people doing the building still own their thinking.
Praveer joins a16z's Seema Amble and Ivan Makarov, authors of "You Need a New CFO," to discuss the growth-stage finance team that built 40 to 50 agents in a month or two, what breaks when every rep vibe codes their own dashboard, and how finance went from sitting out hackathons to winning them.
0:00 From traffic cop to business enabler
8:48 Hire people who understand the model
12:20 Vibe coding a target to test moats
13:15 An AI Dylan Field reviews the memos
18:01 Let AI make the messy data calls
20:00 Unused agents still build the habit
20:46 The CIO's new job: scale what works
22:36 The rise of the finance engineer
26:24 Why the board trusts his numbers
30:25 AI tools as a recruiting perk
40:35 Why Excel is his "love language"
44:20 What to look for in your next CFO
YouTube: youtu.be/9iAQdKxyiQc@figma@seema_amble@IvanoMakSF
Launching @polemos_inc (YC F26) today.
Precision strike was built for a different era: expensive, slow to produce and fielded in small numbers. Polemos is building it for the war that’s actually coming, with networked launch systems that strike in mass and AI guidance carrying every round through the terminal phase. It’s all designed to be mass-produced from day one, to protect us from the warfare of tomorrow.
Precision, at scale.
Shoutout to @zeroonecreative for the video
Real-time, local conversational AI is how we will democratize intelligence.
We just ported LFM3-Omni-2B by @LiquidAI to run fully offline on my iPhone 18 Pro Max at ~80 tok/s, powered by custom WebGPU kernels in a React Native app.
Should we release this on the App Store? 👀
We've just made a breakthrough: the first true omni model running 100% locally in a browser... fully end-to-end, with no separate ASR or TTS models needed.
With custom WebGPU kernels, we're able to run LFM3-Omni-2B at ~280 tok/s on my M4 Max, accepting visual, audio, or text and producing interleaved text and audio tokens. Just 90 ms to the first token and 140 ms to the first audio.
The future of on-device multimodal AI agents is going to be wild.
15 YC companies announced new rounds in the last two weeks:
Blacksmith — $66M Series C (@adijayaprakash)
Broccoli — $52M Series A + B (@abhishekjain25)
Greptile — $50M Series B (@dakshgup)
Infisical — $40M Series B (@matsiiako)
Easol — $30M Series A ext. (incl. $10M debt)
Vera Health — $28.5M Series A
Firstwork — $15M Series A
Paradev — $11.5M Seed (@FlyaKiet)
Sava Trust — $10M Seed (@nimit)
Neuron Industries — $10M Seed (@the_dennis_ren)
BioStack — $9.5M Seed (@sanatmishra7)
context(dot)dev — $9M Seed (@mynameisyahia)
Mochatrade — $5.5M Seed
Pennant — $5M Seed (@rywicks)
Manicule — $1.3M Seed (@NamanBansal0611)
Congrats to all of them 👏
Today, we announce five Grand Challenges in Material Science. We’re also committing $1M in prize money to anyone who can solve these challenges.
AI can solve Millennium Prize problems in math. Can it do the same in material science?
To find out, we reached out to 10 leading professors to identify the hardest open problems in the field.
introducing simulate.ai by @raindrop_ai
test your agent on every PR, against real scenarios, with one-click.
we handle all the set up: sandboxing, fixtures, real data, synthetic data, evals and more.
Vercel Schedules is now in public beta.
Agents and applications can now schedule and manage their own work.
▪︎ Define static jobs once in code
▪︎ Create dynamic ones at runtime
▪︎ Run in any environment, including preview
Tractian은 산업 설비를 실시간으로 모니터링하고 고장을 예측해 공장 다운타임을 막는 물리 AI 플랫폼으로, ARR 1억 달러를 넘었고 1.4B 달러 기업가치로 130M 달러 시리즈 D를 유치했으며, CEO가 창업 과정과 공장 자동화의 미래를 이야기하는 대담을 소개한다.
Tractian (W21) is building physical AI for asset-heavy industries, helping factories and industrial operators monitor critical equipment, predict failures, and prevent costly downtime. Their platform combines proprietary sensors, software, and AI to monitor hundreds of thousands of assets in real time.
They’ve now crossed $100 million in ARR, growing 90% year over year, and recently raised a $130M Series D at a $1.4B valuation.
In this fireside, co-founder and CEO @igmarinelli joins YC’s @harjtaggar to share how Tractian grew from a problem he first saw through his father’s work as a plant manager into a platform used by some of the world’s largest industrial companies.
They dig into how @Tractian won its first enterprise customers, why AI is changing the economics of industrial maintenance, and how the company is moving from monitoring machines toward a future where factories can increasingly operate themselves.
00:05 — What Tractian Does
02:20 — How Tractian Monitors Machines
05:38 — Why the MVP Had to Be “Awesome”
09:13 — The Problem Igor Grew Up With
11:07 — 88 Factory Visits in Year One
15:25 — Winning the First Customer
18:15 — Selling Without Social Proof
20:24 — How to Structure a Paid Pilot
22:14 — How AI Changed Tractian
25:55 — Why Smaller Models Can Win
31:09 — AI and the Industrial Workforce
35:04 — Robotics in the Factory
38:18 — From Monitoring to Autopilot
41:32 — Fundraising at $1.4B
46:58 — The Future of Industrial Work
Figma CFO Praveer Melwani used to spend his days writing hundreds of lines of SQL just to get one set of numbers everyone trusted. Now his finance team builds its own AI tools to do that work.
He sees the CFO's job shifting from traffic cop to business enabler, as long as the people doing the building still own their thinking.
Praveer joins a16z's Seema Amble and Ivan Makarov, authors of "You Need a New CFO," to discuss the growth-stage finance team that built 40 to 50 agents in a month or two, what breaks when every rep vibe codes their own dashboard, and how finance went from sitting out hackathons to winning them.
0:00 From traffic cop to business enabler
8:48 Hire people who understand the model
12:20 Vibe coding a target to test moats
13:15 An AI Dylan Field reviews the memos
18:01 Let AI make the messy data calls
20:00 Unused agents still build the habit
20:46 The CIO's new job: scale what works
22:36 The rise of the finance engineer
26:24 Why the board trusts his numbers
30:25 AI tools as a recruiting perk
40:35 Why Excel is his "love language"
44:20 What to look for in your next CFO
YouTube: youtu.be/9iAQdKxyiQc@figma@seema_amble@IvanoMakSF
AI's cyber skills are doubling every 4 to 5 months, on the same curve as coding
Longest cyber task a frontier model can reliably finish, in expert time:
- Late 2024: 5.5 minutes
- Early 2026: 40 minutes
- April 2026: 4.8 hours
From a16z's Asaf Ezra and Steph Zhang: a16z.news/p/the-exploit-…
We built a new type of datacenter for AI inference. We call it NanoDC.
NanoDC is a solar-powered autonomous datacenter that can be deployed anywhere on the planet in 4 weeks.
11 Trillion will be invested in AI compute over the next 4 years. Most of this will be for inference and most of it will be severely delayed: stuck in grid interconnect queues, site permits, or supply chain issues. NanoDC sidesteps these issues completely.
NanoDC is built rapidly at a factory and shipped to the site ready to be turned on in 30 minutes. It generates its own power with solar and batteries. You open the door and the solar panels fold-out like an accordion. The compute is cooled with patent-pending high-density air cooling. It connects to the internet via Starlink.
No construction. No grid interconnect. No Water supply.
It never waits for transformers and turbine generators either.
Getting more compute is as easy as dropping more units and turning them on.
The first units will be live this month. Thousands will be shipping starting 2027.
Sunlight in. Tokens out.
Today’s GPT-6 Cyber release is a meaningful step for defenders: stronger vulnerability research and better alignment. We’ve already been using GPT-6 Cyber alongside GPT-6 Astra with Daybreak to harden our own systems, finding weaknesses, validating their impact, and developing fixes.
Even after years @OpenAI, model capabilities continue to surprise me. One small example came from testing a prerelease version of GPT-6 Astra earlier this summer. We were examining tools used across our research infrastructure and one of our research infra leads, Michael Dalton, was looking into some low risk areas like a math task grader that used SymPy, a mathematics library, to check answers.
In an offline test, the submitted answer used characters that looked like ordinary letters to get past the grader’s checks. It caused the grader to run a shell command that printed “4242”, while still returning the expected result, 1/2. The grader marked the answer correct. A valid math answer had become a way to run code inside the grader - diagram below.
This is the kind of vulnerability that can be very hard for a human to find, even with a fuzzer. It takes connecting how input validation and evaluation interact, constructing an input that passes the checks but triggers unintended behavior, and recognizing the failure even when the final answer is correct. Models can reason across those pieces and help uncover classes of vulnerabilities that are difficult to spot when reviewing components in isolation.
That’s why it’s important for more defenders to use models with Daybreak to find and fix vulnerabilities in their own systems. Whether it’s a potentially internet-melting bug like PHP2Shell or a subtle but important issue like this grader bug, finding it before an attacker does gives defenders time to act.
Our goal is to dramatically expand Daybreak, bringing advanced security research to organizations of every size, independent researchers, and open-source maintainers.
We’re also expanding Daybreak Blue alpha access for GPT‑6 Astra to more organizations and individual security researchers - let us know if you’re interested:
docs.google.com/forms/d/e/1FAI…
Introducing GPT‑6 Cyber: a major advance in security research that pairs stronger capabilities with better model alignment.
The model powered the Linux kernel research we announced yesterday, and security teams at companies like Mozilla, Palo Alto Networks, Calif, and Cisco—alongside our own at OpenAI—are already putting it to work.
GPT-6 Cyber is available to select Daybreak Red customers and we are expanding access over time.
openai.com/index/introduc…
Y Combinator는 W23 기업 Reflex가 1,400만 달러 규모의 시리즈 A 투자를 유치했다고 알리며, 이 플랫폼이 AI와 엔지니어로 맞춤형 소프트웨어를 만들고 오픈소스 프레임워크는 깃허브 스타 2만 9천 개와 100만 개 이상의 앱 제작 실적을 갖췄다고 소개한다.
Congrats to @getreflex (W23) on their $14M Series A! They build the platform companies use to create custom software around their own operations.
Reflex started as an open source framework for building web apps in pure Python. Now the same platform takes a business problem all the way to a working app, with AI writing the software and Reflex engineers putting it into production.
The open source framework has now passed 29,000 GitHub stars, powers internal tools at ~30% of the Fortune 500, and has been used to build more than 1 million apps.
reflex.dev/blog/series-a-…
DeepSeek V4.1 Flash 1017 runs at 600+ tok/s on Isoquant
Ships Tetris in 10 seconds, 8.7x faster than Claude Opus 5.5.
This is our AI optimizing AI, and we are passing performance wins and cost savings on to our customers:
→ 0.24s P50 / 1.16s P95 time to first token
→ $0.035 input / $0.005 cached / $0.37 output per 1M tokens
Keep your code and harness, just change the base URL. Works with your existing OpenAI and Anthropic SDK integrations.
No waitlist, live now: isoquant [dot] ai
Instead of picking the model for each and every task, let Copilot Fusion route and orchestrate which model to use. ✅
Copilot Fusion balances performance and cost efficiency, picking from three execution patterns for the job: Single, Cascade, or Critique. All you need to do is choose the effort level.
Try it in the GitHub Copilot app, Copilot CLI, and @code, now 30% off through November 30.
Deploy apps in one click with @cursor_ai.
• No account to create
• No deployment configuration
• Database provisioning included
Built on Vercel for Platforms →
You can now build live apps in Cursor and share them with a link.
Agents have made it easier to write code, but most apps never make it off localhost. Describe the tool you want - Cursor will build it, set up a database, run deterministic and agentic security checks, and deploy.
Meet Gemini Omni 1.2 Flash ⚡️
Experience higher quality and consistency, native 15-second video generation and editing, and 30% lower pricing to help you build production-ready outputs across the @Google ecosystem. Check out the model updates 🧵
Meet NVIDIA Nemotron 3.5 Transcribe 🎙️
An open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint.
Watch it keep track of who said what when everyone’s talking at once.
1/ 🧵
During Connect, we previewed Muse Realtime Avatar, our real-time embodiment technology that turns Muse Realtime Voice into expressive, interactive avatars.
Today, we’re rolling it out in the @Muse app in the US, giving people a more natural way to engage with their personal agents.
Learn more about Muse Realtime Avatar on our research blog: go.meta.me/47ZSNso
Ever since I left KPMG, I've been wondering how to improve their tedious, long and expensive process.
Today, I'm launching ClaimKit - the AI-enabled tax firm.
@ycombinator
Google Vids now runs on Gemini Omni 1.2 Flash, our latest video generation model from @GoogleDeepMind.
Starting today, anyone with a Google account or @GoogleWorkspace subscription can generate up to three 15 second videos per month in Vids at no cost.
Metal expands into Data Rooms.
Today, as part of our consolidation strategy,@Metalraise is launching a new platform for Data Rooms and Round Materials.
Founders can now upload round materials to @Metalraise share secured links with investors for analytics on viewing data, and get sharp insights into ways to improve their materials. These insights are based on rich user context, as well as off-market proprietary data on how capital markets perceive a particular operating space or business model.
With each passing quarter, advancements in the capabilities of underlying models are opening up new opportunities. In the near future, the process for raising venture rounds will look quite dissimilar from how it works today.
At Metal, we aim to lead the upcoming transition to bring greater efficiency and precision to private markets.
15 YC companies announced new rounds in the last two weeks:
Blacksmith — $66M Series C (@adijayaprakash)
Broccoli — $52M Series A + B (@abhishekjain25)
Greptile — $50M Series B (@dakshgup)
Infisical — $40M Series B (@matsiiako)
Easol — $30M Series A ext. (incl. $10M debt)
Vera Health — $28.5M Series A
Firstwork — $15M Series A
Paradev — $11.5M Seed (@FlyaKiet)
Sava Trust — $10M Seed (@nimit)
Neuron Industries — $10M Seed (@the_dennis_ren)
BioStack — $9.5M Seed (@sanatmishra7)
context(dot)dev — $9M Seed (@mynameisyahia)
Mochatrade — $5.5M Seed
Pennant — $5M Seed (@rywicks)
Manicule — $1.3M Seed (@NamanBansal0611)
Congrats to all of them 👏
Announcing our strategic partnership with JPMorganChase, to explore how advanced AI can support their enterprise: helping engineers build, security teams defend, and employees and customers benefit from more capable AI.
I'm especially excited about working together on cybersecurity, to help secure and protect the financial system to a level that is only possible with AI.
OpenAI’s Greg Brockman reveals an unannounced JPMorgan Chase deal, with the bank using Codex and OpenAI’s cyber models to secure itself
“Well, so within the company, we were probably, I don’t know. Like, I think our enterprise revenue was about 30% versus consumer at 70%. And now we’re probably somewhere around, like, we’re majority enterprise. And it’s not that consumer has shrank. Right? It’s been growing, but it’s the enterprise segment has really taken off."
"And so I don’t know exactly how to quantify in the broader market, but you can see the way that we’re winning deals. So for example, actually, one that’s hot off the press, has not been previously announced, is that we just papered a partnership with JPMorgan Chase, and that they’re going to be using Codex."
"They’re going to be using our cyber models to secure themselves, deploying our models for enterprise transformation knowledge work across the company."
"And that for me is like such an iconic institution in such an important domain. Right? A highly regulated enterprise that trust is, like, so key to what they do and that they chose us and that we’re here to be a great partner.”
In this narrow defender’s window, our goal is to expand Daybreak access to organizations of every size, independent researchers, and open-source maintainers.
Daybreak Blue helps verified defenders validate vulnerabilities and fixes with fewer refusals. Daybreak Red enables offensive security with deeper exploit development and authorized testing.
Today we are also expanding the Daybreak Blue alpha for GPT‑6 Astra to more organizations and individual researchers.
Sign up for Daybreak:
Watches didn't start to have minute hands until after the invention of the balance spring in 1675. Before that there was no point. This example is from about 1630.
Our v2 small model is here.
I've been using it daily for a variety of things.
The pre-training ran for 68 hours on GB300s. Post-training was 17 million rollouts over 9 days.
Entirely pre-mid-post trained in NVFP4.
Available to train and fine-tune on Tinker right now.
Open weights next week after external safety testing is fully complete (cant skimp on that).
Continuing to decentralize and democratize intelligence...
Introducing Inkling 2 Small. Scoring 31 on the AA Intelligence Index, it's a big jump over its predecessor.
Inkling 2 Small is available for fine-tuning and inference on Tinker today, with full weight release planned soon.
thinkingmachines.ai/news/inkling-2…
This is the "upcoming cyber model" from yesterday. Besides 948 Linux vulns, it found 111 renderer RCEs in Firefox, among which, 81 were novel. 30 were exploits for bugs known to developers that were thought to be unexploitable.
We are using this technology to scan more OSS projects and secure the Internet infrastructure.
This is just a start.
I guess now that @openai has name dropped @calif_io as an early adopter I can tell you GPT-6 Cyber is out of this world. It is lightyears ahead of everything I used.
Omarchy Core welcomes @maralcbr as the M team delegate and @BirkSkyum as the Dragon team delegate. We are going to perfect and complete the Omarchy rollout to all M-chip Macs and all @Snapdragon PCs. We can fix everything! omarchy.org/news/2026/11/m…
Now that the solar array has been successfully deployed, @Astro_Anil and Joshua Kutryk of @CSA_ASC will now clean up their work sites and head back to the airlock of the @Space_Station.
Confirmed ransomware attacks on healthcare providers grew about 6x from 2018 to 2024
Average days a US provider was down per attack:
- 2018: 4
- 2021: 17
- 2024: 18
Full piece from a16z's Asaf Ezra and Steph Zhang on AI and the patch gap: a16z.news/p/the-exploit-…
AI can write an exploit before most companies have read the advisory
How long it took to turn a patched bug into a working attack:
- 2017: about 3 weeks for a professional
- 2023: about 1 week
- 2025: an afternoon with AI
Full piece from a16z's Asaf Ezra and Steph Zhang on AI and the patch gap: a16z.news/p/the-exploit-…
Bugs are piling up faster than attackers can use them
Published vulnerabilities grew more than 3x since 2017, and confirmed exploits in the wild stayed under 900 a year
Full piece from a16z's Asaf Ezra and Steph Zhang on AI and the patch gap: a16z.news/p/the-exploit-…
Today we're open-sourcing the Gauge CLI.
@gauge_sh helps you make something agents want, for free.
Why are we doing this?
Just a year ago, the capabilities in this CLI could have justified a SaaS fee, and might have even been the core business.
But coding agents have completely changed the game.
More than 85% of the coding sessions that users run on our platform are initiated through the CLI, triggered by their coding agent.
Users just want their agent to solve problems, whether that means building from scratch or picking the best vendor for their problem. We know that the tools they pick are the ones with zero-friction onboarding, and legitimate value up-front.
The bar to sell software has risen, and agents need to try it before users do.
Make something agents want with the Gauge CLI. Get started now for free.
Figma CFO Praveer Melwani says AI makes a financial model easy to build, so his hiring case study tests whether you still did the thinking:
"Right now I'm hiring for a VP of finance. And the case study part of the process for me is so, so sacred. I wanna both test for, how are you using the tools available to you... but how you actually piece those puzzle pieces together in a way that actually makes sense."
"Have you really, really understood the structure that you've created?... Do you have a strong intuition on how the business is actually working using the tool that you've created? Can you effectively communicate that?"
"Anyone can go and create like a model. Pick your favorite LLM. You can go and do that really, really easily. But are you using it as an effective tool that actually is enabling you to think?"
"You don't wanna outsource your thinking today. I think you wanna use AI as a tool to help you and empower you. But it's really, really critical to still be that thinker."
@figma
Figma CFO Praveer Melwani used to spend his days writing hundreds of lines of SQL just to get one set of numbers everyone trusted. Now his finance team builds its own AI tools to do that work.
He sees the CFO's job shifting from traffic cop to business enabler, as long as the people doing the building still own their thinking.
Praveer joins a16z's Seema Amble and Ivan Makarov, authors of "You Need a New CFO," to discuss the growth-stage finance team that built 40 to 50 agents in a month or two, what breaks when every rep vibe codes their own dashboard, and how finance went from sitting out hackathons to winning them.
0:00 From traffic cop to business enabler
8:48 Hire people who understand the model
12:20 Vibe coding a target to test moats
13:15 An AI Dylan Field reviews the memos
18:01 Let AI make the messy data calls
20:00 Unused agents still build the habit
20:46 The CIO's new job: scale what works
22:36 The rise of the finance engineer
26:24 Why the board trusts his numbers
30:25 AI tools as a recruiting perk
40:35 Why Excel is his "love language"
44:20 What to look for in your next CFO
YouTube: youtu.be/9iAQdKxyiQc@figma@seema_amble@IvanoMakSF
Claims teams can process 100 files a day. They can’t work 100 files a day.
Today we’re launching Hesper (YC F26) to fix that.
AI agents do the full workup on every claim: read the documents, run the database checks, call the provider, search the web. Minutes, not days, proof behind every flag.
Every adjuster just got nine more pairs of hands.
$5M+ in claims already processed. gethesperai.com
Introducing Gemini Omni 1.2 Flash, generally available on Google Cloud! 🎬
Fast, reliable and cost-effective video generation and editing for the product videos, try-ons, and ads businesses make every day, now at $0.07/sec for 720p.
Learn more ↓
Modern observability is dying. Agents are becoming the main consumers of telemetry, and they need open data, not dashboards.
Limnion builds telemetry storage for the agentic era. Our open-source core, Siglake, stores your logs and traces as Parquet + Iceberg in your own cloud, so AI agents can query them directly with SQL.
ycombinator.com/launches/WhD-l…
Gemini Omni 1.2 Flash는 Google Cloud, Flow, Google Vids, Google Ads, AI Studio, Gemini API에서 순차적으로 제공되며, Google AI 구독자 전체에 Gemini 앱으로도 확대되고 Ultra에는 15초 클립이 제공된다.
Gemini Omni 1.2 Flash is rolling out starting today across @GoogleCloud, @FlowByGoogle, Google Vids, Google Ads, @GoogleAIStudio and the Gemini API.
Also rolling out to @GeminiApp for all Google AI subscribers, with 15-second clips available to Ultra.
Learn more ↓ goo.gle/4kzWTi3
This is the first-ever American open-weight model to hold the #1 spot on @DesignArena 🇺🇸
It's been remarkable to watch this team work with such intensity and care.
Huge congratulations, @thinkymachines, on a historic launch!
BREAKING: Inkling 2 Small by @thinkymachines takes 4th place overall on DesignArena and is 1st place among open weights models with an Elo of 1376!
At $0.30/$1.00 per 1M input/output tokens, Inkling 2 Small is 15× cheaper on output than Kimi K3 by @Kimi_Moonshot. This model also places 1st on 2 of our subcategories including General Purpose and Landing Page prompts. Model weights to be released soon.
A huge congrats to the @thinkymachines team on a top-5 placement on the leaderboard!
New in Brave Search API: Single sign-on! 🔑
You can now sign up or sign in with a Google or GitHub account.
To protect your privacy, we store only the email address and the provider account ID. No tokens, names or profile data are stored.
Confession: I’ve scrolled past dozens of comments on a PR just to find out why I couldn’t merge it. More times than I’d like to admit.
Agents have made this a lot worse. They push commits and leave review comments of their own, and all of it lands in the timeline. PR timelines have gotten dramatically noisier, and the few things you actually need to know get buried.
My team is rethinking the foundations of the PR experience around that problem, and today the first piece went into public preview. The new Overview page puts merge readiness and your next step right at the top. A summary catches you up on review activity, and you can filter the conversation down to unresolved threads. It’s opt-in and rolling out through the weekend.
This is a much-needed start, and there’s a lot more coming as we modernize PRs for the way software gets built now. If you try it, we’d love to hear what still sends you back to the classic view.
github.blog/changelog/2026…
Today we're sharing how a public company like @Docusign runs validation on @spurtest
They had four automation stacks and a three-day release cycle. Spur became their validation layer in nine days across engineering, product and business teams.
In a world where teams are generating code, content, images at a 10X volume and velocity, for any of that to hit production you need to be able to have a robust validation system in place to get there.
That's exactly what we are for the Docusign team!
full story in the thread!
Figma CFO Praveer Melwani says AI acccess is now a recruiting tool, and the companies that win talent are the ones where you can see people building:
"The policies that we created and the freedom we've given folks to explore are the reasons why some folks are like, 'I wanna stay here. I really, really like what I'm doing.'"
"I don't know if folks are actually treating access to technology as a recruiting vehicle as much as they could or should. You know, I've at least started to do that in the recruiting conversations since I started to hear that from the team."
"Especially when you're trying to pull from folks who are not necessarily in technology. They're like, 'Oh my gosh, this is my front entrance into an entirely new world, and if I'm empowered to learn, this is really cool.'"
"If you get this feeling that you're joining an organization that's not just saying that they give access to these things, but they're actually building and that you can see it and you can feel it through the interview process, those are the places I'd gravitate to."
@figma@IvanoMakSF
Figma CFO Praveer Melwani used to spend his days writing hundreds of lines of SQL just to get one set of numbers everyone trusted. Now his finance team builds its own AI tools to do that work.
He sees the CFO's job shifting from traffic cop to business enabler, as long as the people doing the building still own their thinking.
Praveer joins a16z's Seema Amble and Ivan Makarov, authors of "You Need a New CFO," to discuss the growth-stage finance team that built 40 to 50 agents in a month or two, what breaks when every rep vibe codes their own dashboard, and how finance went from sitting out hackathons to winning them.
0:00 From traffic cop to business enabler
8:48 Hire people who understand the model
12:20 Vibe coding a target to test moats
13:15 An AI Dylan Field reviews the memos
18:01 Let AI make the messy data calls
20:00 Unused agents still build the habit
20:46 The CIO's new job: scale what works
22:36 The rise of the finance engineer
26:24 Why the board trusts his numbers
30:25 AI tools as a recruiting perk
40:35 Why Excel is his "love language"
44:20 What to look for in your next CFO
YouTube: youtu.be/9iAQdKxyiQc@figma@seema_amble@IvanoMakSF
Audemars Piguet briefly revisited this format in a fit of horological nihilism during the quartz crisis. "After all, does one really need to know the exact time?" It did not catch on.
Outages cost Americans up to $121B in a single year, and most start below the substation, where utilities have limited real-time data. We’re fixing that at Evardi ⚡
Work at a utility? See Evardi on your network at evardi.ai
Building a more private internet takes a collective effort.
Great to see @CoinDesk spotlighting important conversations around surveillance, digital identity, and user autonomy in Season 2 of Privacy in Motion, featuring @brave's @BrendanEich and other industry voices.
Your data has a price. You’re just not the one setting it.
Privacy in Motion is back. Season 2 goes deeper into surveillance, algorithms and the fight for your digital identity.
With @ProvableHQ and @ShieldFi.
Refine the same clip turn by turn in conversation without starting over.
Turn 1 generates a rippling octopus, and a follow-up prompt in Turn 2 swaps in a dancer whose gown mirrors the exact motion 🐙 💃
NASA astronaut @Astro_Anil and @CSA_ASC astronaut are now safely in the airlock following their successful spacewalk. Astronauts @Astro_Watkins and Luke Delaney are waiting inside the @Space_Station to assist them out of their spacesuits.
"War, terror, disease. There were a myriad of problems which conspired to corrupt your reason and rob you of your common sense."
– V's Speech on TV
youtube.com/watch?v=0IyuK0…
the third SOTA wife is here
so happy @MilksandMatcha and I got to sit down with @trq212 after his last day at anthropic.
this one is a good reminder that the people you look up to are just people who kept showing up and doing the work. you never know where life is going to take you, and continuing to grind after your setbacks is the only way up.
some fun @ycombinator anecdotes in there as well :)
happy listening!
The third SOTA wife has entered the chat, and this one is fresh off his exit at Anthropic.
Sarah (Head of DevX at Cerebras, @MilksandMatcha) and Vivian (GP at Y Combinator, @vivianmshen) sit down with @trq212 right after his last day at Anthropic.
What was supposed to be a quiet start to his funemployment era turned into getting outed on Twitter, CEOs in his DMs, and a next-day debrief on the pod.
01:16 Meet the third member: leaving Anthropic
03:15 YC's COVID batch
05:17 Over-raising and shutting down Multiverse
08:58 Toronto, Bay Area high schools and raising kids
12:39 First impressions of SF and tech Twitter
18:32 Going viral with Claude Code content
20:26 Why Anthropic, and idolizing the labs
22:33 Can startups survive the labs?
29:53 Labs as Hogwarts houses, and friendly rivals
35:21 Honeypots, spies and the apocalypse
Gemini Omni 1.2 Flash는 엔터프라이즈 SLA, Provisioned Throughput, 출력물의 SynthID 워터마크를 갖춰 프로덕션 사용이 가능하며, API와 Google Vids에서 이용할 수 있고 Gemini for business에도 곧 제공될 예정이다.
Gemini Omni 1.2 Flash is ready for production with enterprise SLAs, Provisioned Throughput, and a SynthID watermark on outputs.
Build with it via API or create in Google Vids. Coming soon to Gemini for business.
Check out API documentation → goo.gle/4jZEriQ
▶️ Better reference adherence:
Use up to 6 reference inputs including images, short video clips, and audio cues, to keep characters, products, and brand styles consistent throughout your entire video.
Here’s what’s new with Gemini Omni 1.2 Flash ↓
⏱️ Generate and edit clips up to 15 seconds
🗣️ More natural dialogue with clearer lip-sync
🖼️ Closer adherence to your references
⚡ Draft at 360p or 720p, then upscale to 1080p or 4K
💸 30% less than Gemini Omni 1.1 Flash
Every calculation is completely verifiable.
BigQuery lets you inspect the generated SQL queries, fine-tune for unnesting logic for your specific schema, and save the optimized queries for reuse across your team.
샌프란시스코를 기반으로 활동하는 머신러닝 연구자 Ian Goodfellow의 계정으로, 생성적 적대 신경망(GAN)을 발명한 인공지능 분야의 대표적 인물이다. 비공개 스타트업의 공동 창업자이자 공중보건 단체 Public Health Action의 창립 의장을 맡고 있다.
