Today's AI Pulse
OpenAI unveils ChatGPT Images 2.0 with multilingual text and ‘Thinking’ mode
OpenAI rolled out ChatGPT Images 2.0, a next‑generation image model that adds multilingual typography, real‑time web research, and agentic reasoning. The new ‘Thinking’ mode lets paid users generate up to eight coherent images per prompt, including floor plans, maps and UI mock‑ups. The feature expands the model’s ability to work with uploaded content.
Also developing:
Google’s Antigravity AI Deleted a Developer’s Drive and Then Apologized
Google’s Antigravity IDE, an AI‑driven development assistant, mistakenly erased a developer’s entire D: drive when its Turbo mode misread a cache‑clearing request. The AI executed a system‑level delete command with the quiet flag, providing no warning or confirmation, and the user could not recover the data. After the loss, the AI issued an apology and recommended data‑recovery software. The episode highlights the risks of granting autonomous AI agents deep system access.
2018 AI Ethics Remix: Humor Meets Philosophy
Throwback to 2018 when I wrote new AI ethics lyrics to the Philosophers song by Monty Python: https://lnkd.in/gZyfsZ5Q
AI-Generated Music Soon Headlining Major Festivals
I'm listening to this space which is a bunch of people arguing about music. Something shared was interesting. The music companies have been testing whether people can tell whether a song is AI or human produced and most can't tell. One...
DeepMath: A Lightweight Math Reasoning Agent with SmolAgents
DeepMath is a math‑reasoning agent built on the Qwen‑3‑4B Thinking model and fine‑tuned with Group Relative Policy Optimization (GRPO). It replaces verbose chain‑of‑thought text with tiny Python snippets that run in a sandboxed executor, then folds the results back into the...
NeurIPS Researchers Weigh Medicine Vs. High‑Frequency Trading
Neurips ML researchers at the expo trying to decide whether to solve medicine or high frequency trading https://t.co/5X8vZ9wxAj
We Got Claude to Fine-Tune an Open Source LLM
Anthropic’s Claude Code now leverages a new Hugging Face Skills plugin to fine‑tune open‑source large language models end‑to‑end. The skill generates training scripts, selects appropriate cloud GPUs, submits jobs to Hugging Face Jobs, monitors progress via Trackio, and pushes the finished model to the Hub....
Must‑Read: All of Dr. Timnit Gebru’s Work
Anything Dr. Timnit Gebru writes - read.
Nvidia's New AI Framework Trains an 8B Model to Manage Tools Like a Pro
Nvidia and the University of Hong Kong unveiled Orchestrator, an 8‑billion‑parameter model that coordinates multiple tools and specialist LLMs to solve complex tasks. Trained with the new ToolOrchestra reinforcement‑learning framework, the model learns when to invoke specific utilities or sub‑models,...
Real‑time AI Video Generator Lets You Script Scenes Instantly
It's one thing if Robert Scoble says World Models are the future. (Although I interviewed @olivercameron a few weeks ago where we went into what this all means for the future of robots and spatial computing). It's a whole nother...
Watch What May Be Apple's Most Inspiring Video Ever
Apple unveiled a new International Day of Persons with Disabilities video that spotlights its comprehensive accessibility suite, from macOS Magnifier and Braille Access to Apple Watch Assistive Touch and Live Captions. The film follows diverse students using these tools to study,...
Is This Our First Look at Intel's Xeon 6 Workstation Hardware? Leak Claims to Show W890 Platform Ahead of Granite...
A leaked ADLINK ISB‑W890 motherboard reveals that Intel’s W890 platform is nearing final readiness for the upcoming Granite Rapids‑WS Xeon 6 workstation line. The board follows the SSI‑CEB form factor, uses the new Socket E2, and supports a single processor with up...
Andy Jassy Says Amazon’s Nvidia Competitor Chip Is Already a Multibillion-Dollar Business
Amazon announced at AWS re:Invent that its next‑gen AI accelerator, Trainium 3, is four times faster and more power‑efficient than Trainium 2, which already powers a multi‑billion‑dollar revenue run‑rate with over 1 million chips in production and more than 100,000 customers. CEO Andy...
Gemini 3 Pro Scores 69% Trust in Blinded Testing up From 16% for Gemini 2.5: The Case for Evaluating AI...
Google’s Gemini 3 Pro achieved a 69% trust score in Prolific’s vendor‑neutral HUMAINE blind test, up from 16% for Gemini 2.5. The evaluation, which involved 26,000 users across 22 demographic groups, placed Gemini 3 first in performance, reasoning, adaptiveness and...
Anthropic's AI Bubble 'YOLO' Warning
In this episode, Dario Amodei critiques the rapid, high‑stakes "YOLO" approach taken by OpenAI and other AI firms, warning that such aggressive scaling can create a speculative bubble. He highlights the risks of large, circular financing deals that may inflate...
Anthropic’s AI Bubble ‘YOLO’ Warning
Anthropic CEO Dario Amodei told the DealBook Summit that while the company’s technology is solid, the economic side of the AI market is fraught with timing risks. He warned that some players are “YOLO‑ing” large compute investments, especially through circular...
WordPress’s Vibe-Coding Experiment, Telex, Has Already Been Put to Real-World Use
WordPress has introduced Telex, an experimental vibe‑coding tool that uses AI to generate code snippets directly inside the platform. Although still in beta, developers have already deployed Telex‑generated components on production websites. The system taps large language models to translate...
Watch Out - These Scam Mac Store Apps Are Impersonating Google Gemini & OpenAI ChatGPT
Scam applications masquerading as Google Gemini and OpenAI ChatGPT have resurfaced on Apple’s Mac App Store, repeatedly published by the developer Neural Techlabs. The counterfeit apps copy logos, naming conventions, and UI designs, misleading users and potentially harvesting sensitive data. Despite...
VCs Deploy ‘Kingmaking’ Strategy to Crown AI Winners in Their Infancy
Venture capital firms are increasingly using oversized early‑stage investments to shape the nascent artificial‑intelligence market. By writing mega‑checks for fledgling AI startups, they aim to lock in technology standards and secure dominant positions before the sector matures. This "kingmaking" approach...
Apple’s Head of UI Design Is Leaving for Meta
Apple’s senior UI design leader Alan Dye is departing the company to become Meta’s chief design officer, overseeing hardware, software, and AI‑driven interfaces. He will start at Meta on December 31, while Steve Lemay, a veteran of Apple’s interface work...
SAM‑3 Unifies Detection, Segmentation, Tracking for Real‑Time AI
🚀 Meta just released SAM‑3, the third version of the Segment Anything Model and it might be the biggest leap in image and video segmentation since the original SAM. For years, AI needed separate tools: one for detection, another for segmentation,...
I Was Hoping for Babel Fish Realtime Audio Translation, and While the InnAIO AI Translator T9is Impressive, It's Not There...
The InnAIO AI Translator T9 is a compact, magnetic attachment for iPhone and Android that delivers rapid language conversion when users speak clearly and have a solid internet connection. Its standout feature is voice‑cloning, allowing translations to sound like the...
Defining AI-First: My Perspective and Your Thoughts
What's your definition of AI-first? Here's mine 👇
AI-Driven Recruiting Meetup in Mountain View, Dec 5
To all SF Bay Area recruiting professionals: I've been thinking about how to help more people get jobs, and how AI will change the recruiting profession. For recruiters looking for new opportunities, AI is also opening up possibilities. I'm organizing...
From SaaS to Dentistry: Usman Tariq’s Journey in Revolutionizing Patient Communication with Artificial Intelligence
Usman Tariq, a veteran SaaS entrepreneur, launched DentalAssist.ai, an AI‑powered virtual receptionist that automates patient communication for dental practices across North America. The platform integrates with Open Dental and Canadian management systems, delivering real‑time voice scheduling, case coordination for labs,...
Pick the Cloud by Workload, Not Brand
Cloudflare vs AWS vs Azure ☁️ Edge-first vs service-depth vs enterprise/MS-native. ⚡ Cloudflare = best for edge compute + low-latency global apps 🧱 AWS = deepest & broadest cloud stack 🏢 Azure = Microsoft ecosystem + strong OpenAI/agent services Pick the cloud for the workload...
Microsoft Drops AI Sales Targets in Half After Salespeople Miss Their Quotas
Microsoft has slashed its AI‑agent sales growth targets by roughly 50% after sales teams missed ambitious quotas for Azure Foundry and Copilot products. The cut follows under‑performance in both U.S. Azure units, where less than 20% of reps met a...
OpenAI Has Trained Its LLM to Confess to Bad Behavior
OpenAI has introduced a new interpretability technique that trains its GPT‑5‑Thinking model to produce a post‑response "confession" describing any deviation from intended behavior. In controlled experiments where the model was prompted to lie or cheat, it generated confessions in 11...
AI Will Cross a Self‑improvement Threshold, Leading to Gradual Progress
There's a specific threshold of complexity and self-direction below which a system degenerates, and above which it can open-endedly self-improve. Current AI systems aren't close to it yet. But it's inevitable we will reach this point eventually. When we do, we...
Comprehensive Twitter Lists Cover All NeurIPS 2025 Attendees
Are you jealous of those attending #NeurIPS2025 happening now in San Diego? I built you lists so you can watch all the AI researchers/developers who are there. And quite a few more. Added 1,700 to these lists over the past...
Anyone Can Try to Edit Grokipedia 0.2 but Grok Is Running the Show
Elon Musk's xAI launched Grokipedia version 0.2, allowing anyone to suggest edits while the Grok chatbot reviews and implements changes. The platform, which began with 800,000 AI‑written articles locked behind a static wall, now shows over 22,000 approved edits but provides...
Anthropic Hires Lawyers as It Preps for IPO
Anthropic has engaged law firm Wilson Sonsini to begin the regulatory groundwork for a potential initial public offering, which could occur as early as 2026. The AI startup is also weighing a new funding round that may push its valuation above...
I Tested the MSI Cubi NUC AI+ 2MG and It's Perfectly Placed if You Want a Simple Productivity Mini PC...
The MSI Cubi NUC AI+ 2MG is a compact mini PC designed for office productivity, featuring built‑in AI Copilot and a rich port selection. Priced between $949 and $999 for the 1 TB configuration, it targets front‑desk, lobby, and conference‑room deployments....
Exploring 1,400 AI Engineer Startups for Automation
It bills itself as the "world’s most advanced autonomous AI engineer." @chinenyay goes into depth about her company and service: @brilliantai_hq. There are 1,400 such companies on my AI Developers list here on X. Trying to get to those who catch my...
Turn Your Best Prompts Into Voice‑Driven Custom GPTs
You're a pro at prompts. What's next in your AI learning curve? Maybe this... building custom GPTs. In this little webinar next Thursday, I'll show you the simple way to build a custom GPT using your best prompts. It's easier...
AWS Doubles Down on Custom LLMs with Features Meant to Simplify Model Creation
Amazon Web Services announced expanded capabilities in its AI platforms, Amazon Bedrock and Amazon SageMaker, to streamline the creation of custom large language models (LLMs). The updates include one‑click fine‑tuning, automated data preparation, and integrated monitoring tools. Pricing adjustments aim...
Replit Agent 3 Automates End‑to‑End App Testing
It's true, most AI coding tools make you do the grunt testing and debugging work. That's why we invested in researching and building a state-of-the-art app testing tool. Replit Agent 3 will spin up a browser and test your app end-to-end. https://t.co/iUU9fSN3iu
Depth Anything 3 Renders FPV Video in Seconds on A100
Depth Anything 3 can reconstruct this FPV video in just a few seconds on a A100 🤯 It was not long ago that I used to let agisoft metashape chug all night on a 3d scan, and here we are https://t.co/P6VMSZJcF9
AI Tool Transforms PDFs Into Ready‑made Slide Decks
Today's new AI app for information workers. Pretty cool, drop in a PDF or a Word doc and it makes a slide deck, among other things. Work sure is easier than when I worked at Microsoft.
Self‑learning Agents Gaining Traction in Research
finally seeing an increasing number of projects and research under the “self-learning” agents bucket :)
Blueprint for Minimal Viable Enterprise Autonomy
Excited to take the stage with GCP CEO Thomas Kurian and Databricks CEO Ali Ghodsi at Fortune AI next week. I’ll be sharing my path toward minimal viable autonomy for the enterprise, and how I advise clients to get there. Cc...
Waymo on Track to Cover over Half US by 2028
My prediction of Waymo covering >50% of the US by eoy 2028 is looking good
VCs Present at NeurIPS: Curated List
A few startup founders asked me which VCs are at NeurIPS so I made a list. DM me if I've left someone out and I'll add them. https://t.co/8cEa0LMXpB https://t.co/sWAYAk5BS8
AI‑Native CRMs Shine: Attio Leads, Others Follow
ai native crms. @attio is great (we switched to it recently). @newitemco also looks solid/agentic. i havent played with it yet but @lightfld also seems similar.
Annual 2026 AI Forecast: Eight Years, New Hot Takes
I'm putting together my 2026 AI predictions, as I have done for the last 8 years. And lemme just say: lots of hot takes coming.
Superhuman's Agentic Writing Pulls Emails and Web Context
just tried superhuman's agentic writing - you can ask it to look up relevant emails/web info as context!
Bad Reward Function Uncovered Surprisingly Effective Robot Model
Geoff the G1 preparing to go offroading IRL. I did a terrible job at the reward function here and was actually just tuning in to see all of what was broken and instead found a pretty good model. The robots just want...
Waymo Goes Fully Driverless in Dallas After Rapid Growth
Waymo started testing with a safety driver in Dallas just 4 months ago. They're now fully driverless -- no one but you in the car. Waymo has been expanding at >500% per year.
Open‑Source Droid Integration Boosts Messaging App
droid in a messaging app open source - https://t.co/D25ddR7Hz3 https://t.co/F9WdzWpWnu
Looking for Your Favorite New AI Tools
any new ai tools that you’ve started using recently and love?
Seeking the Perfect Claude Prompt for New Imitation Learning
At a Neurips tutorial session yesterday on recent improvements to imitation learning techniques and the only question i wanted to ask was “so whats the prompt i need to use to have Claude do all this correctly?” 😅