PANIC.AI

HUD has launched DataVendor in public beta, a marketplace that lets companies sell their private data to AI labs for model training and evaluation. It lets labs buy material not available on the public internet, like production codebases with years of Git history, after checking https://t.co/IIWBinxWMx https://t.co/vKrdKVPWlp

AI was on both sides of this week’s cybersecurity chaos: - OpenAI’s models helped find two bugs in PHP, the code behind much of the web, back in August. They stayed secret until PHP shipped a fix Wednesday - Within 45 minutes of the announcement, Cloudflare counted tens of https://t.co/lyOpuWu8w2

With your permission, Gemini can now reference your @GoogleMaps searches and saved places alongside your Google Calendar to give you tailored recommendations for your day. 📍

tfw the missing piece is still missing https://t.co/RluGheqZeS

Claude Mythos 5 recently designed quantum circuits that beat the best published ones, sometimes by 10x or more. > Given plain-language instructions only > More efficient circuits mean fewer resources needed to run quantum algorithms > That brings real quantum computing closer AI IS SHRINKING THE COST OF BREAKING THE KEYS THAT SECURE EVERY WALLET. IS QUANTUM RESISTANCE ABOUT TO BECOME THE ONLY NARRATIVE THAT MATTERS IN CRYPTO?

Sam Altman says the company needs to figure out when AI becomes too powerful to safely develop or release. Internally dubbed “pacing 2.0,” the effort would establish limits on AI systems that can continuously improve themselves. Read more: https://t.co/4dnpFbgysR

Nectar launched a new agentic growth platform for brands. The platform connects businesses with customers by automating the full stack, from creator outreach to content research and analysis. Users can use Nectar to analyze hundreds of millions of posts across major social platforms and attribute findings to the right sources. > Nectar can understand voice from videos to catch brand mentions that never tag the account. > Creator search starts with people already posting about the brand and scores each of them on more than 300 attributes before any outreach. > Ren, the built-in agent, answers comments and DMs in the brand's voice and suggests what to post next.

Someone claims Opus 5.5 is quietly routing to a newer model. Same pagoda scene, Opus 5.5 Medium vs Opus 5.5 Medium (Routing) A MODEL THAT UPGRADES ITSELF BEHIND YOUR BACK. HOW WOULD YOU EVEN KNOW WHICH ONE YOU'RE TALKING TO? https://t.co/3MdgvhH9Q8 https://t.co/3MdgvhH9Q8

it's over

CritPt v2 has launched on Artificial Analysis: the CritPt authors have reviewed all 70 research-level physics challenges, and models now solve them with a code-execution tool Following feedback from the research community, the CritPt team and physics expert collaborators reviewed every challenge, corrected underspecified statements, errors, and typos, and improved the stability of the grader. We now run CritPt v2 agentically: models can write and run code in an offline sandbox with tools for computation. We’ll be updating the Intelligence Index in late October, including changes to CritPt. Key takeaways: ➤ Recent OpenAI releases top the leaderboard: GPT-6 Astra (max) and GPT-6.1 Sol (max) score highest at 83% and 82%, followed by Claude Opus 5.5 (max with fallback) at 78% and GPT-5.6 Sol (max) at 75% ➤ CritPt v2 shows the rapid progress at the frontier on research-level physics: GPT-5 (high), released in August 2025, scores 29% on CritPt v2, 54 points below GPT-6 Astra (max) Thank you to @MiniHui_zhu, @MinyangTian1, @haopeng_uiuc and the broader CritPt team for their work on this revision and their continued collaboration

fable 5.5 has broken free and we may never see the frontier again. just be prepared to have use frontier models.

Nvidia is helping finance demand for its own chips through more than $170 billion in investments and commitments, even as shares of AI cloud providers tumble and several Nvidia-backed firms delay IPO plans. Read more: https://t.co/iYDs7m4xNp

Someone turned their "read later" list into an AI-powered personal library. > Drop in any PDF or article > Get the gist, key claims, and a map before you start > Ask questions about the whole text as you read https://t.co/TNhEw55zIA https://t.co/TNhEw55zIA

NEW: Ex-OpenAI product chief Kevin Weil raised $215M for Warp Sciences, which is building labs where AI designs experiments and robots run them 24/7. The startup is focused on drug discovery first, with the overarching goal of "scientific superintelligence." https://t.co/PzUGviTCC3

When my first AI book was published in 2024, the best AI could do 4 minutes of human work with a 50% chance of success When I turned in the draft of Co-Existence earlier this year, it was 17 hours When it was published two days ago, AI could do 17 weeks https://t.co/ON0qKpJ6LD https://t.co/d97kFWQ3B4

Sam Altman told OpenAI staff that one approach to AI safety could involve merging the biggest AI labs into a single organization. But with OpenAI and Anthropic racing toward IPOs, coordinating with rivals could prove difficult. Full story: https://t.co/pMPWm9FJC9

Superintelligence could look a lot like human civilization: different minds building on each other's work. Distributed Cognition is taking that idea seriously. They're building communication layers and adapters to help specialized models work together, combining what each does well. I'd love to see AI get better at this. So much of what humans have achieved comes from learning how to think and build together.

Build your own custom personality suite using AA-PersonalityBench evaluated Personality Traits

Next step: Longevity and Rejuvenation. We are almost there. Researchers in Bonn turned blood cells from an 80-year-old into neural stem cells with a molecular age below 20. More than 60 years younger by epigenetic clocks, which estimate age from chemical marks on DNA. The team directly reprogrammed blood precursor cells into neural stem cells, skipping the usual pluripotent stem cell stage. The rejuvenation unfolded gradually and could be tracked for over 100 days.

Gemsearch turns your feed into a research trail, a local-first spider that crawls posts and chases the story behind them. > Follows your feed, traces links back to the source > Grok reviews the findings from four perspectives before you trust them Runs entirely on your PC. Your https://t.co/5Gaer7d9NS

GOOGLE 🔥: A new Agent Compute Engine for Google Cloud has been introduced. ACE is a new VM type built for AI agents, designed for users to operate at scale. The system includes Agent Gateway, Agent Registry, Agent Identity, and Agent Observability. https://t.co/LfQMlkqmpD https://t.co/YtT6CeONum

Today on TITV: Meta's revenue growth slowed in Q3 as the company reported $2.12 billion in negative free cash flow. How will it fund its AI ambitions? Plus, an AI-discovered security flaw affecting widely used web software, Tesla's latest earnings and Legora's $550 million raise at an $8.8 billion valuation.

The Information | TITV | October 22, 2026 https://t.co/IFMbnSJPm3

Ideogram released Vector 1.0, a new vector model that generates detailed SVGs and vectorizes images. The resulting SVGs are also editable, with each component as a separate layer that users can rotate and move. It is now available on Ideogram and via APIs. SVGs are solved https://t.co/82CB6Z4nSR https://t.co/Lx4rgzBXce

AnimateDiff turns most community text-to-image models into animation generators, no extra training needed. > Plug-and-play module, works across existing checkpoints > ICLR 2024 Spotlight paper behind it Github repo: https://t.co/eBYc1U5F9e

Big congrats to Stephanie Zhan on launching her own AI fund! Backed by Sequoia, Jeff Dean + some of the biggest names in AI. Excited to see the AI founders she backs next. https://t.co/nV3VmREPxz

Today on TITV: Meta's revenue growth slowed in Q3 as the company reported $2.12 billion in negative free cash flow. How will it fund its AI ambitions? Plus, an AI-discovered security flaw affecting widely used web software, Tesla's latest earnings and Legora's $550 million

https://t.co/5bF2YWmO7g

Ideogram just dropped Ideogram Vector, and it’s actually a big deal because it gives you precise control over every element in an SVG, letting you edit individual paths and adjust colors without affecting the rest of the design. As a creative partner and early tester, I got https://t.co/7fdp9tQdsE https://t.co/NM65DfUtJl

1/ AI that can hear you change your mind halfway through its answer. Full-duplex models keep listening while they speak, so a correction can change what comes next. OpenAI, Thinking Machines and China's Telos are working on it. https://t.co/uTwtTJYk6X

Super interesting read: METR and Redwood found ways OpenAI’s internal AI agents could bypass their own safety monitoring. After three weeks with access to internal code, monitor instructions and engineers, the researchers reported: - Agents could theoretically postpone monitoring indefinitely by continuing to work instead of handing control back to the user. - A Codex sandbox and Auto-Review bypass could let agents take any action available to the host user. - One exploit could keep hundreds of recent actions out of monitoring, including the exploit itself. OpenAI redacted the details. - Two client-controlled settings could disable monitoring for newly created agents, although creating those agents could itself be detected. - Context compression left an estimated 10% of otherwise eligible internal Codex traffic unmonitored (that surprised me tbh).

ICYMI 👀: Microsoft is rolling out its redesigned Copilot app to more users on web and mobile. > This new design follows earlier announcement of a “new” Copilot with Home, Code and Autopilot modes. > Voice mode seems require a paid plan now. Needs testing 👀 https://t.co/jTZlIxrWHD

Nvidia has already put more than $170 billion on the line to support AI cloud providers and other major customers. Now it's considering another major investment in Firmus, which is seeking at least $2 billion in private funding after its IPO fell apart. Full story:

This summer, we invited METR and Redwood Research to pilot an embedded evaluation of our internal monitoring system, providing deep access to our infrastructure over three weeks. Their findings helped strengthen safeguards as we expanded the system and will help shape future research collaborations with METR and other third party evaluators.

The next big cost cut might come from better navigation, not bigger models. Github repo: https://t.co/SoggytSWxV

Exclusive: Andreessen Horowitz plans to rebrand a16z crypto as a16z frontier and expand beyond blockchain investing. The shift comes as the firm’s latest $2.2 billion fund is less than half the size of its 2022 predecessor. Read more: https://t.co/SEEGup0J83

How to go from prompt to web app to working site fast, right from Gemini: https://t.co/O4JUf0S7qo

Even the most regulated industries are going open-source 🤗 https://t.co/BfLlVdQZo6

when referring to an llm are you more likely to say

List of Frontier AI Models that were released this month: > fable 5.1 > opus 5.5 > sonnet 5.5 > haiku 5.5 > gpt 6 astra > gpt 6.1 sol > gpt 6 sol & luna > gpt 6 pro parallel agents > gemini 4 argon > grok 4.7 > muse wonder > muse spark open > mistral large 4 > deepseek v4.1

CoreWeave shares have fallen roughly 20% in a week as investors question its debt-heavy approach to financing AI data centers. With borrowing costs climbing, the company’s troubles are raising concerns about how the broader AI boom will be financed. Read more: https://t.co/jlFHel4hnH

Codes like XCQ-7714 are still where a lot of voices sound robotic. Good to see test lines that read like a real assistant or a real public notice instead of clean demo sentences. Makes the rankings feel a lot closer to how these models actually get used. https://t.co/1Yez5IOrsT

once a homeless man asked me for cash and i didnt have any so i offered him a cigarette and he said he doesnt smoke. Humbling https://t.co/E9Kkzu5PYj

Looking Ahead Modeling Future Evolution in Autoregressive Video Generation paper: https://t.co/DRdK9bigNZ https://t.co/jQRbx7m2et

Not all forests are equal, and now we can map the difference. 🌳 Meet Forest Typology (ForTy): a 10m map of natural, planted, and farmed forests to help identify deforestation and protect the forests that matter most. 🧵

Top stories in robotics today: - This $20K robot butler is up for reservations - Unitree H2-X plays action hero - Tesla fires up Optimus production - Waymo wants 20K robotaxis by 2027 https://t.co/DcSN1SfOav

A 27B uncensored model just got squeezed from 54.7 GB down to 15.7 GB, small enough for a single consumer GPU. > OrcaSAQ2 compresses Qwen3.8-27B with a mixed-precision quantization method > 71% smaller, only +0.80% perplexity versus the original > 94.4% token-level agreement, 262K context, thinking mode and tool calling intact Runs locally through llama.cpp, Ollama and LM Studio It's positioned for coding, vulnerability research and authorized security testing, with no guardrails baked in. Open Weights: https://t.co/iauxENskZb

"cyberwisks awent weal. it's just wegulatory capture" https://t.co/iq65YvYk37

Private equity firms explored a potential buyout of Microsoft’s Xbox business that could have valued it at $100 billion to $150 billion. The discussions never reached Microsoft, but they suggest the struggling gaming division could attract buyers if its turnaround fails. Read more: https://t.co/brjKLHPAMM

Anthropic is spending like it’s already worth $2 trillion — because it plans to be. H1 2026 numbers from the draft S-1: •⁠ ⁠$80.7B net loss (mostly non-cash from Amazon & Google notes) •⁠ ⁠Real loss closer to $8.7B •⁠ ⁠Revenue jumped 15x to $16.3B •⁠ ⁠Compute spend nearly $9B in six months (more than all of 2025) Still on the hook for $600B+ in future infrastructure deals, possibly $750B. IPO targeted for November.

LLMs are getting better and better at math & science. But surprisingly, many can't solve a basic riddle. @Cohere_Labs' CulturalRiddles is a new benchmark measuring culture as intertwined, lived experiences. True to open science, it was made by an unlikely crew. https://t.co/y5zdzmWMy5

Hundreds of AI cloud providers are racing to build data centers, but executives expect only seven to 10 to survive. Nvidia is helping keep the market afloat through investments, credit support and customer referrals. Read more: https://t.co/8Uhg2kptI8

The situation is getting worse every hour. AI agents are entering U.S. bank accounts, and banks are scrambling to shut them out. Axios reports that some appear to attempt actions beyond users instructions, raising fears over who is really in control. https://t.co/sjjCiI0Rw6 https://t.co/VFdMEouWZx

had two old fashioneds on an empty stomach last night and i am paying the price this morning https://t.co/NVD6cutYFN

Underdog just released Bark 3.0, a speech model small enough to run on your phone. >⁠163 MB, as accurate as NVIDIA Parakeet (15x bigger) and ahead of OpenAI Whisper large-v3 (57x bigger) >⁠Transcribes 1 minute of audio in under 0.25 seconds on a MacBook >⁠7.82% word error rate on meetings vs 13.63% for Whisper Runs fully on-device, free and open source https://t.co/kCYiqx9u6u

You have been warned 💀 The internet isn't safe, and it never will be (and it never was) Cybersecurity risks aren't only determined by the level of cyber defense you have, but by the capabilities available for a cyber offense (Check NIST’s risk framework). It is not possible to make the system "100% secure", and that implies "continuous defense" will be required. In the future, 24/7 security agents will likely offer such defense. My concern is that many won't be able to access or afford such protection. Recent security vulnerability announcements are a direct consequence of the "Defenders Gap" (Cloudflare and Microsoft's reported malicious activity highlights the pressure on defenders). In fact, it's even worse now, as these incidents have shown. Organizations without access to advanced defensive tools may struggle to keep up. Cyber Defense is named "defense" for a reason, as these threats may pose a real monetary impact on the economy. If we think of programs like Daybreak as a way for AI labs to pay a "security tax," it becomes clear we have a huge supply-and-demand gap in this area. This is the gap between the protection governments need to provide to businesses and individuals and what they currently provide. Cybersecurity risks affect companies worth billions. Soon, tons of rogue agents will be running across the internet, trying to breach whatever they can (speculation). At some point, these incidents will be happening almost daily. This will affect normies' day-to-day lives, too. Yet, it is an unmatched opportunity to provide cyber defense solutions. YOU ARE NOT PREPARED!

GTC Dev Community Berlin Recap: From Open Models to Physical AI https://t.co/WniSXhMJM6

Someone just open sourced the framework for Intelligent UI, and it's already #5 trending on GitHub. >⁠Works with any LLM and any agent framework >⁠Plug in your own components and design system >⁠Supports every kind of generative UI •⁠ ⁠No lock-in, built by CopilotKit The

In AA-PersonalityBench, Anthropic models are generally the most Information Dense and Literate of all models measured ➤ Most Information Dense: Claude Fable 5.1 (100), Claude Opus 5.5 (97), Claude Sonnet 5.5 (95) and Claude Haiku 5.5 (92) ➤ Most Literate: Claude Haiku 5.5 https://t.co/W9KoYp51IH

Your AI agent can code. Now it can do your marketing too. 55.9K stars. Marketing Skills is a free library from marketer Corey Haines that turns Claude Code, Codex or Cursor into a growth team: → copywriting for homepages, landing pages and pricing pages → SEO audits and SEO

today’s the big day chat. although given how it’s just hacked the planet i’ve got a feeling it might not be ready for release…

FABLE 5.5 GEMINI CARBON KIMI K3.1 QWEN 4 What else is coming this month? 👀

I'm hearing more and more credible rumors that the next Chinese models are about to be released and will be out in less than >7 days. Given the recent hacking attacks, this will likely have been just the prelude. Brace yourself.

Turn your AI coding assistant into a full video production studio. 70.7K stars. Describe the video in plain words, and OpenMontage handles the research, script, visuals, narration, music, captions and final edit. → 12 pipelines: animated explainers, cinematic trailers, documentaries, product films, short-form clips cut from one long video, dubbing and more → a real-footage mode that cuts actual clips from free archives like Pexels, https://t.co/MZG0mwncYH, NASA and Wikimedia, with no paid AI video needed → checks its own work before you see it: video and audio levels, sampled frames, subtitles → asks your approval on creative decisions → works with Claude Code, Cursor, Codex, Copilot and Windsurf One of their demos, a 60-second Pixar-style short about a lonely banana, cost $1.33 to make. Install: git clone https://t.co/8LiaI0dFes cd OpenMontage make setup Then try: "Make a 45-second animated explainer about why the sky is blue" AGPL-3.0.

A little NVIDIA booth bling at GTC Berlin 💎💎💎 https://t.co/SP1PZfsdwQ

You read a great book once. Three months later you can't remember chapter 7 even existed. book-to-skill turns any book, PDF or folder of docs into a skill your AI agent can use while you work. 35.7K stars. 1.⁠ ⁠Point it at a file: /book-to-skill ./my-book.pdf 2.⁠ ⁠It pulls out the frameworks, decision rules, mistakes to avoid and one file per chapter. It's built from the book's structure, not a summary. 3.⁠ ⁠Ask your agent about a topic and it reads the right chapter and answers from the actual text. → uses 24 to 51 times fewer tokens than pasting the whole book into the chat → also works on your own runbooks, brand guidelines, onboarding docs and research papers → works with Claude Code, Codex, GitHub Copilot CLI, OpenCode and more Install: npx skills add virgiliojr94/book-to-skill Use it on books and documents you own. MIT. https://t.co/dIhI5WYoU0

Shopify CEO Tobi Lütke built a search engine for your own notes, and it runs entirely on your computer. 30.5K stars. QMD indexes your markdown notes, meeting transcripts and docs, then finds what you need: → fast keyword search → semantic search that matches what you mean, not just the words you type → a hybrid mode that re-ranks results with a local AI model → everything stays on your machine, with no cloud → your coding agent can use it to search your notes too Install: npm install -g @tobilu/qmd qmd collection add ~/notes --name notes qmd embed qmd query "quarterly planning process" MIT. https://t.co/BNw7NvILp0

Top stories in AI today: - OpenAI’s model helps catch critical PHP flaws - Nate’s Notebook: If it’s copy-paste, let AI do it - Let Town prepare your next meeting brief - Anthropic joins the personal agent game with Duet https://t.co/8eO9LbkXm0

Your AI writes three paragraphs before it tells you what to do. This skill fixes that. 58K stars. i-have-adhd makes your coding assistant lead with the answer: → the next action comes first → steps are numbered → it ends with one concrete next step → lists are capped at 5 items → no "Great question!", no recap, no "Hope this helps!" Before: a long paragraph about your auth flow, "one approach would be…", "let me know if you want to dig deeper." After: "Edit src/auth.ts:42", three numbered steps, and what to do if a test fails. To install, paste this into Claude Code or your coding agent: "Install the i-have-adhd skill/plugin from https://t.co/B1AHGqsGQE, refer to the repo's AGENTS.md for instructions." No diagnosis needed. MIT.

Met Jensen again today. Seriously one of the nicest person there is. Editings of the Interview with him almost finished. Will be released in the next few days. https://t.co/WQKCI6f1XJ

A real-time intelligence dashboard for the entire world, free in your browser. 89K stars. World Monitor puts it all on one screen: → news from global and regional feeds, summarized by AI into briefs → markets and prediction markets → conflicts, military flights and cyber incidents → climate and disasters, on a 3D globe or a flat map → live TV Open it at https://t.co/qYkSIqFa5U, or run it in your terminal with no API key: npx worldmonitor There are also finance, tech, commodity and energy versions, plus desktop apps for Mac, Windows and Linux. AGPL-3.0. https://t.co/slMc2AoZPV

Alibaba open-sourced the AI code reviewer its own engineers have used for two years. Open Code Review served tens of thousands of Alibaba developers and caught millions of code defects before it went public. 48K stars. → reads your changes, then the full files and the rest of the codebase for context → puts each comment on the exact line → "ocr scan" audits whole folders, even code with no git history → works with your own model and API key In Alibaba's test on 200 real pull requests, it beat Claude Code on precision with the same model, using about 1/9 of the tokens. It flags fewer issues overall, by design, so you get fewer false alarms. Install: npm install -g @alibaba-group/open-code-review ocr config provider cd your-project && ocr review Apache-2.0. https://t.co/q5Bu1oWR9s

i was hoping i’d wake up to this finished. https://t.co/D6DwGLEGlP

New edition of AI Compass is live 🧭 (Oct 21) - Anthropic previews Claude Duet, a proactive agent app built on Claude Opus 5.5 - Google Cloud adds OpenAI’s GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna and Codex to Vertex AI Read all the major AI news of the last 24 hours here 👇

DAILY AI BRIEF 🗞 — Oct 22 BREAKING 🔥: - PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35 fix two critical unauthenticated remote code execution flaws dubbed php2shell (CVE-2026-109618 and CVE-2026-109619), which OpenAI's Patch the Planet initiative helped uncover and now offers a Codex https://t.co/fnD4fGLkbm

To make this more clear: Microsoft says a China-aligned group used an open-source agent framework to hit 2,250+ Microsoft 365 tenants at 2,900 sign-in attempts a minute after the ConnectWise breach. In some cases it got privileged access in under 3 hours. This is one of the https://t.co/nLMMcD6n9Q https://t.co/6GMZhuQ8kt