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:
Ring Adds AI Controversial Facial Recognition to Doorbell Cameras
The episode examines Amazon Ring's new AI-powered facial recognition feature for its doorbell cameras, detailing how the technology identifies faces and alerts users. It highlights growing privacy and surveillance concerns, noting criticism from civil liberties groups and potential misuse in neighborhood monitoring. The discussion also considers the trade‑off between enhanced security for homeowners and the broader implications for personal data protection.
Genspark to Integrate Tiny Fish for Browser-Enabled Agents
Prediction for @ericjing_ai, founder of @genspark_ai: you will add @Tiny_Fish so that your agents can fire up a web browser and do more. A lot more.
Scientists Reveal a Tiny Brain Chip that Streams Thoughts in Real Time
Researchers from Columbia, Stanford, Penn and NY‑Presbyterian unveiled BISC, a 3 mm³ silicon brain‑computer interface that streams neural data at 100 Mbps. The chip hosts 65,536 electrodes, 1,024 recording and 16,384 stimulation channels, and communicates wirelessly via a wearable relay. Early human...
Ultimate AI Masterclass: Models, Space Data Centers, Geopolitics
This @patrick_oshag @GavinSBaker episode is an absolute master class in AI models, frontier labs, data centers (in space), scaling laws, Nvidia, AI geopolitics, and more. I will be listening multiple times. https://t.co/1j3W8HkGIM
Humanoid Robotics Target $780B Market, Nursing $322B
Just did the math and there are 12 verticals for humanoid robotics that have TAMs over $50b (total is $780b), assuming ~1:1 ratio of people to humanoids, that's $390b of direct SAM This excludes nursing which is probably the most valuable...
The AI Race Gets a Massive Power Shift
The episode examines President Trump's unprecedented decision to let Nvidia ship its H200 AI chips to China, a move that could dramatically shift global AI power by boosting China's compute capabilities while deepening its reliance on U.S. hardware. It explores...
Teen's 17,000-Worker Vision Dataset Powers Robot Learning
He put glasses on 17,000 factory workers to gather this data. Real impressive 18 year old. Met him last week. Good dataset for training robots and building world models.
AI-Powered Bioacoustics Boosts Endangered Species Conservation
How AI is helping advance the science of bioacoustics to save endangered species https://t.co/2NWjsbUxbs #ai #sustainability #IoT #CES2026 https://t.co/WQu76Pu9eV
Salesforce Bets on Agentic AI Platform Following Cautiously Optimistic Earnings Report
Salesforce reported Q3 2025 revenue of $10.3 billion, a 9% year‑on‑year increase, and raised its full‑year revenue outlook to $41.5 billion. Despite the earnings beat, the company’s stock has slipped 29% so far in 2025. At its World Tour in London, Salesforce unveiled...
Nonprofits Can Access $60M Grant – Register by Dec 11
BIG opportunity for non-profits to receive millions of dollars to completely reinvent how we work. Aspen and Pivotal have launched a $60M grant that will support ideas and programs across training, AI tooling, and media. Only 1 day left: https://t.co/xLQp6TfcNG *You do not...
Sam Altman Tells Jimmy Fallon that AI Is the "Equalizing Force" The World Needs
OpenAI CEO Sam Altman appeared on The Tonight Show to highlight ChatGPT’s explosive growth, noting that more than 800 million people now use the service weekly. He framed AI as an "equalizing force" comparable to the global spread of smartphones, arguing...
Spend One Hour with Claude to Crush Annoying Tasks
Take on a one-hour AI challenge. Pick a nasty, brutal, annoying problem. Open up a new chat with Claude Opus 4.5. And refuse to leave your computer or give up for the next hour while you hammer the model. Donate...
Why Cursor’s CEO Believes OpenAI, Anthropic Competition Won’t Crush His Startup
Anysphere, the maker of the Cursor AI coding assistant, posted $1 billion in annualized revenue and secured $2.3 billion in new capital at a $29.3 billion valuation, yet the CEO Michael Truell ruled out an IPO for now. The company is doubling down...
UK Strengthens Subsea Cables Against Russian Interference
The UK has unveiled Atlantic Bastion, a military program aimed at protecting undersea cables and pipelines from Russian interference. The initiative will employ autonomous surface and underwater vessels, advanced sensors, and naval assets, with initial testing this year and deployments...
Sale of Nvidia AI Chips to China Helps All Parties
President Donald Trump announced that Nvidia may sell its H200 AI chips to select Chinese customers, reversing a Biden-era ban on the high‑end processor. Nvidia will remit 25% of each sale to the U.S. Treasury, framing the move as a...
Rivian Is Building Its Own AI Assistant
Rivian has spent nearly two years building an in‑house AI assistant that operates independently of its multibillion‑dollar joint venture with Volkswagen. The system is engineered to integrate directly with vehicle controls through a model‑agnostic, agentic framework that blends edge and...
Apriel-1.6-15b-Thinker: Cost-Efficient Frontier Multimodal Performance
ServiceNow released Apriel-1.6-15B-Thinker, a 15‑billion‑parameter multimodal reasoning model that rivals the performance of models ten times larger. Built on the Apriel‑1.5 foundation, it boosts text and vision reasoning while cutting reasoning token usage by over 30%. Trained on NVIDIA GB200...
Three in 10 US Teens Use AI Chatbots Every Day, but Safety Concerns Are Growing
The Pew Research Center reports that roughly three in ten U.S. teenagers use AI chatbots every day, with 59% favoring ChatGPT. Overall teen internet use remains high—97% go online daily and 40% are almost constantly connected, though that figure fell...
Why Vision Language Models Ignore What They See with Munawar Hayat - #758
In this episode, Qualcomm AI Research scientist Munawar Hayat explains why Vision‑Language Models often ignore visual input, leading to object hallucination, and how his team’s attention‑guided alignment technique improves visual grounding. He also introduces Generalized Contrastive Learning for efficient multi‑modal...
Mistral Launches Powerful Devstral 2 Coding Model Including Open Source, Laptop-Friendly Version
Mistral AI unveiled Devstral 2, a 123‑billion‑parameter coding model with a 256K‑token context window, alongside a 24‑billion‑parameter Devstral Small 2 that runs on a single laptop. Both models are open‑weight and available free for a limited time via API and...
Proposed EO Threatens State AI Protections, Benefits Big Tech
David Sacks is arguing that the new proposed EO is fine: that we don't need state laws to e g prevent AI companions from grooming children or to protect us from any of the 4 C's. That's unfortunately incorrect;...
New AI OS Promises Enterprise and Home Revolution
The new AI operating system. @qinzytech just visited my home. While talking with him I got those goosebumps like I did when I first saw Siri, Tesla, Matic Robots, Insta360 (I was the first to see all those). Today at 4 p.m. he...
I’m Obsessed with Redfin’s AI Search
Redfin has rolled out an AI‑powered search tool that lets users query listings in plain language, instantly generating tailored property results. The feature appears as a new prompt on the site’s search bar and leverages large language models to interpret...
Amazon’s Ring Rolls Out Controversial, AI-Powered Facial-Recognition Feature to Video Doorbells
Amazon’s Ring is adding an AI‑driven facial‑recognition feature called Familiar Faces to its video doorbells, allowing users to label up to fifty faces and receive personalized alerts. The feature is optional, with encrypted biometric data stored in the cloud and...
SAM 3D Delivers High‑Fidelity Single‑Image 3D Reconstruction
📢SAM 3D: Single-Image 3D Reconstruction with Foundation-Model Reliability In this week’s deep dive, we break down SAM 3D, Meta’s groundbreaking framework that redefines what’s possible in single-image 3D reconstruction. Unlike earlier pipelines that struggle with occlusions, clutter, and ambiguous textures, SAM...
OpenAI, Anthropic, and Block Join New Linux Foundation Effort to Standardize the AI Agent Era
The Linux Foundation has created the Agentic AI Foundation (AAIF) to provide a neutral hub for open‑source AI‑agent projects. Anthropic, Block, and OpenAI are the founding donors, contributing the Model Context Protocol, Goose framework, and AGENTS.md instruction file respectively. Major...
Trump Set to Allow Nvidia H200 Chips to Be Exported to China
President Donald Trump announced that the United States will resume exporting Nvidia's H200 AI accelerator chips to approved commercial customers in China, subject to a 25% export fee. The move excludes newer Nvidia products such as Blackwell and Rubin, which...
Amazon Adds Delivery Tracking, Last-Minute Adds, Gift Ideas to Alexa+
Amazon is expanding Alexa+ with a suite of shopping‑enabled tools for Echo Show 15 and Show 21 users in the United States and Canada. The new "Shopping Essentials" hub lets customers track deliveries in real time, view recent orders, and manage household...
AI Transforms Images Into Gesture‑controlled 3D Point Clouds
gm So if you ask Nano Banana to make an image into a Depth Map, you get a grayscale image where white=close & black=far Then ask Google AI Studio for an app that turns depth maps into point clouds & lets you...
2025 Report Reveals Skills for Tomorrow’s Jobs
Future of Jobs Report 2025: The jobs of the future, and the skills you need to get them https://t.co/lBMcF8QL6P #FutureOfWork #GLMC @wef @Global_LMC @WorldBank #CES2026 #AI #IoT #5G
EU Launches Antitrust Probe Into Google’s AI Search Tools
The European Commission has opened an antitrust investigation into Google’s AI Overview and AI Mode tools, alleging the company uses website content and YouTube videos to generate answer summaries without compensating publishers. The probe will assess whether Google imposes unfair...
Edit LLM Behavior Safely without Retraining, Says CTGT
Cyril and the team at CTGT are productizing mechanistic interpretability. They make it possible to edit the behavior of LLMs to add safety policy guarantees without retraining, in a way that is much more reliable than simple prompting.
Humanoid Reliability: 30 Actuators Fail Every 300 Hours
Biggest issue for large scale Humanoid deployment is reliability. You have 30 actuators with 10,000 hr est. life, that equates to a failure every 300 hours on average according to Lusser's Law of series system reliability. The hand is likely to...
Microsoft to Invest $17.5B in India by 2029 as AI Race Accelerates
Microsoft announced a $17.5 billion investment in India through 2029, its largest Asian commitment, to build new data centers, AI infrastructure and skilling programs. The plan adds to a $3 billion pledge earlier this year and includes a massive Hyderabad region with...
Legs or Wheels Enable Versatile, Cost‑Effective Humanoid Robots
The reason humanoids should have legs or offroad wheels is that you have to perform tasks with the bots in different environments with the same robots (to lower unit cost of tasks), if you have have to have a human...
Unitree Thrives without Dev Ecosystem; Tesla Stays Closed
Unitree is winning without any semblance of a developer ecosystem. While Tesla appears to have no interest in opening up a developer ecosystem for Optimus.
India Proposes Charging OpenAI, Google for Training AI on Copyrighted Content
India has unveiled a draft framework that would require AI developers to pay royalties for training on any copyrighted work, creating a single collecting body to distribute payments to creators. The mandatory blanket licence is pitched as a low‑cost compliance...
Robotics Lack Elastic Demand; LLMs Scale Infinitely
One issue with applied robotics is that few physical world services scale with high elasticity of demand, meaning, lowering the cost of a task doesn't necessarily get offset by more demand (eg napkin folding). While LLM value scales with...
Grok Tops the Charts as #1
Grok is #1 https://t.co/v20U3iXOP5
A Look Under the Hood of DeepSeek’s AI Models Doesn’t Provide All the Answers
DeepSeek’s latest large‑language models have been subjected to a peer‑review analysis that sheds light on the company’s data‑centric training pipeline and architectural tweaks. The study confirms the use of a transformer backbone with sparse‑attention mechanisms and a multilingual corpus exceeding...
EU Should Create Open‑source AI for Transparent Governance
EU countries should get together to build an advisory AI that should run the EU with all decisions being transparent and participatory. This should be a fully open source, open data stack with all inputs also available so you can run...
Use Famous Authors' Names to Define AI Writing Style
when i want to specify writing style to an ai, i don’t describe the writing style in the prompt i first ask an ai to give me writers known for the style i want, and then include the writers’ name in...
Databricks' OfficeQA Uncovers Disconnect: AI Agents Ace Abstract Tests but Stall at 45% on Enterprise Docs
Databricks introduced OfficeQA, a benchmark that tests AI agents on document‑heavy enterprise tasks using 89,000 pages of U.S. Treasury Bulletins. Tests show top agents such as Claude Opus 4.5 and GPT‑5.1 achieve only 37‑44% accuracy on raw PDFs, rising to 68%...
AI Celebrity Likenesses Now Accepted: Tech and Legal Shift
Curious why Nano Banana Pro, ChatGPT image gen etc are all suddenly cool with generating celebrity likeness? Like what changed from days of deepfake fear-mongering and fears of legal repercussions? Did fingerprinting get good or is there a new legal argument...
Shipping Reliable AI Requires More than Prompting
Prompting isn’t LLM engineering. Shipping reliable AI is. 8 key LLM engineering skills: 1. Prompt engineering 2. Context engineering 3. Fine-tuning 4. RAG 5. Agents 6. Deployment 7. Optimization 8. Observability Which one is your biggest gap today? 👇 #LLM #GenAI #RAG #AIAgents #MLOps
The UK Must Build Smarter Networks to Lead in AI
AI could add £550 billion to UK GDP by 2035, but the nation controls only 3 % of global AI compute power versus the US’s 75 %. Legacy networks cause 20‑25 % inefficiencies and expose firms to AI‑driven cyber threats. Modern, AI‑ready networks—leveraging SD‑WAN,...
ChatGPT Crafts Roadmap; Gemini Stays Factual, Less Helpful
ChatGPT was able to analyze a dataset and give me a Data Analysis roadmap. Meanwhile, here's the response I got from Gemini. The only upside in this is that it did not hallucinate(by making up answers). Still, this is incredible...
1000 Hours Reveal Six Crucial Prompt Engineering Patterns
this guy literally put in 1000 hours of prompt engineering to nail down the 6 patterns that actually matter. https://t.co/0hMWaNdgMx
Anthropic and Accenture Sign Multi-Year AI Strategic Partnership
Anthropic announced a three‑year strategic partnership with Accenture, creating the Accenture Anthropic Business Group. The deal includes formal Claude training for Accenture’s 30,000 employees and access to Claude Code tools for its vast developer workforce. Anthropic now commands roughly 40%...
MistralAI Hits 68% on 24
68% on SWE-Bench Verified on just 24b! Laptop class 72.2% on 124b 👏 Great job @MistralAI team, looking forward to trying it out https://t.co/I9VlzwMnwB