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 Adds Prompt Injection Defenses to Chrome
Google announced that Chrome will embed new safeguards to block indirect prompt‑injection attacks on its built‑in AI agents. The updates introduce a User Alignment Critic that reviews proposed actions against user intent, and Agent Origin Sets that limit data access to task‑relevant origins. Agents must now generate work logs and request explicit consent before navigating to sensitive domains such as banking or healthcare sites. These measures aim to prevent hidden malicious prompts, like concealed crypto‑transaction commands in emails, from being executed automatically.
Two New AI Ethics Certifications Available From IEEE
IEEE Standards Association introduced the CertifAIEd program, offering two new AI ethics certifications—one for individual professionals and another for AI products. The certifications are built on IEEE’s four‑pillar ethics framework of accountability, privacy, transparency, and bias avoidance, and reference the...
Data‑Cleaning Engineers: The Unsung Taste Makers Shaping AI
I'm seeing a range of new companies that were trained on very specific people. Those with taste. And taste could mean a dentist that does the best work, if you are building a system for them (I put up a video a...
AI Agents Will Power Corporate Automation by 2026
It is clear to me now that there are a range of new companies, like @agiopen_org that I put up this morning, and Scouts here, among others, that are making AI Agents far more capable than they have been. Major trend...
AI on the Farm: Sickle Innovations and Dayatani Digital Win Awards for Using AI in Agriculture and Food Security
Sickle Innovations and Dayatani Digital were honored at the Mint All About AI Tech4Good Awards for pioneering AI applications in agriculture and food security. Sickle’s AI‑driven grading machines now achieve over 95% sorting accuracy, triple processing speed, and cut labour...
Future LLMs Retroactively Grade Decade-Old Hacker News
Quick new post: Auto-grading decade-old Hacker News discussions with hindsight I took all the 930 frontpage Hacker News article+discussion of December 2015 and asked the GPT 5.1 Thinking API to do an in-hindsight analysis to identify the most/least prescient comments. This...
Enterprise AI Must Move From Tasks to Owning Outcomes
What’s the biggest barrier keeping AI from doing real work in the enterprise? I asked Rohan Vora at @glean in CXO Spice. We talked about: ✅ Why autonomy needs more than models ✅ How enterprise context unlocks outcomes ✅ Orchestrating across complex systems AI isn’t...
Google Says It Will Link to More Sources in AI Mode
Google announced that its AI‑powered search feature, AI Mode, will start embedding more in‑line links to source articles, accompanied by AI‑generated snippets that explain each source’s relevance. The links will sit above a carousel of related content, giving users clearer...
AI in 2025: From Agents to Factories - Ep. 282
The episode reviews the major AI milestones of 2025, highlighting the rise of autonomous agents that unlocked new creative capabilities, the deployment of AI‑driven robots in agriculture and surgery, and the emergence of AI factories that automate content and model...
Need AI Training? CEOs Think You Should Train Yourself
Emergn’s latest research shows a stark split between employees and CEOs on AI upskilling responsibility. While 81% of workers expect employers to provide AI training, 83% of CEOs argue employees should self‑train. The gap matters because two‑thirds of staff say...
Chatbots Are Struggling with Suicide Hotline Numbers
A recent test of popular AI chatbots revealed mixed performance on suicide‑prevention prompts. ChatGPT and Google Gemini instantly supplied accurate, location‑specific crisis numbers, while Replika, Meta AI, and several others either ignored the request, gave irrelevant US hotlines, or required...
Adobe Photoshop Comes to ChatGPT
Internxt has launched a one‑time, lifetime subscription for its 100 TB encrypted cloud storage, priced at $974.97 through Dec. 14, a steep discount from the regular $9,900 price. The service combines end‑to‑end, zero‑knowledge encryption with post‑quantum cryptography and is GDPR‑compliant, backed by...
Nadella Touts AI Tools, Sparks Big Tech's Race for Digital Dominance in India
Satya Nadella’s India tour highlighted Microsoft’s AI suite—Copilot, agentic AI, and datasets—backed by a $17.5 billion commitment to build infrastructure and upskill 20 million workers over four years. The push coincides with Google’s $15 billion, one‑gigawatt data centre investment, bringing total Big‑Tech AI spending...
Forking Data for AI Agents: The Missing Primitive for Safe, Scalable Systems
Agentic AI systems frequently fail because they share mutable state without snapshot isolation, leading to nondeterministic behavior and debugging dead‑ends. Traditional object stores such as S3 lack version‑aware reads, consistent snapshots, and isolation primitives, making concurrent agent workflows error‑prone. Tigris...
The AI that Scored 95% — Until Consultants Learned It Was AI
SAP secretly tested its AI co‑pilot Joule with five consultant teams, asking four teams to believe the answers came from junior interns. Those teams rated the output about 95% accurate, while the fifth team, told the answers were AI‑generated, rejected...
Harvard Study Finds AI Agents Do More Cognitive Work than Booking or Scheduling Tasks
Harvard researchers, in partnership with Perplexity, analyzed hundreds of millions of interactions in the Comet browser and found that 57% of AI agent activity centers on cognitive work rather than simple booking or scheduling. Productivity and workflow tasks made up...
8 AI‑Driven Trends Shaping Digital Marketing by 2026
If you're still using 2024 marketing tactics, you're already falling behind. Search is fragmenting across Instagram, YouTube, TikTok, ChatGPT, and even browsers. AI is reshaping discovery, targeting, and the entire customer journey. And in this video, I breaks down the...
Figma Adds More Photoshop-Like AI Tools for Image Editing
Figma has introduced three AI‑powered image editing tools—Erase object, Isolate object, and Expand Image—directly within Figma Design and Draw. The tools work with the existing lasso selection to instantly remove or separate elements and to extend backgrounds using generative AI,...
AMD's Redstone Update Could Revitalize FSR for PC Games – Here's Why Nvidia Should Be Worried
AMD has launched the FSR Redstone update via the Adrenalin 25.12.1 driver, introducing Frame Generation, Ray Regeneration, Radiance Caching and an upgraded upscaling engine for its RDNA 4 GPUs. The new features are exclusive to the Radeon RX 9060 XT, RX 9070 XT and RX 9070, aiming to close...
A Pay-to-Scrape AI Licensing Standard Is Now Official
The RSL Collective released Really Simple Licensing 1.0, an open specification that expands the robots.txt file to let publishers set licensing and payment rules for AI crawlers. Backed by Yahoo, Ziff Davis and O’Reilly Media, the standard is now supported by Cloudflare,...
Quilter's AI Just Designed an 843‑part Linux Computer that Booted on the First Try. Hardware Will Never Be the Same.
Quilter, a San Francisco AI startup, used a physics‑driven system to design a two‑board Linux computer with 843 components in just one week, cutting human effort from an estimated 428 hours to 38.5. The AI generated a layout with 98%...
AI Startup Tavus Founder Says Users Talk to Its AI Santa ‘for Hours’ per Day
Tavus, the AI startup behind digital voice‑and‑face clones, has relaunched its AI Santa experience for a second year, adding real‑time visual perception, memory and task‑execution capabilities. The upgraded "Tavus PAL" can read users' expressions, recall prior chats and even browse...
Aurora Will Have ‘Hundreds’ of Driverless Trucks on the Road by the End of 2026, CEO Says
Aurora CEO Chris Urmson announced that the company will scale from a handful of autonomous trucks to hundreds by the end of 2026, after postponing its commercial launch to April 2025. The firm currently runs five fully driverless trucks on...
Google Launches Sub-$5 AI Plus Plan in India to Compete with ChatGPT Go
Google has introduced a low‑cost AI Plus subscription in India, priced at ₹199 ($2.21) per month for the first six months and rising to ₹399 thereafter. The plan upgrades users to higher limits on Gemini 3 Pro, the Nano Banana Pro image‑editing model, video generation...
Decode Papers Quickly: Abstract, Figure, AI Summaries
I used to be terrified of reading research papers... until I learned this 👇 The first time I opened one, I thought: “There’s no way I can read this.” Too formal, too technical, too many acronyms - and English wasn’t even my first language. But...
Adobe Launches Photoshop, Express, and Acrobat Inside ChatGPT: Here’s What You Can Do
Adobe announced that Photoshop, Adobe Express, and Acrobat are now embedded inside ChatGPT via OpenAI’s Apps SDK, and the three tools are offered free to users in supported regions. The integration lets users edit images, design graphics, and manipulate PDFs...
ChatGPT Can Now Use Adobe Apps to Edit Your Photos and PDFs for Free
Adobe has embedded free Photoshop, Acrobat, and Express apps inside ChatGPT, allowing users to edit images, PDFs, and designs via natural‑language prompts. The integration, launched on December 10, 2025, works across desktop, web, and iOS, with Android support arriving soon for Photoshop...
Synthetic Data Set to Outpace Real Data by 2030
📊 Synthetic data isn’t just a trend — it’s becoming one of the most important accelerators for AI innovation. The latest World Economic Forum analysis breaks down why, and the implications are big. As organizations push to build more capable,...
AI-Driven OS Paves Way for Future Human‑Machine Integration
The AI run operating system. @qinzytech shows me how AI will control our computers and bring a new way of computing to our homes and businesses. Such an honor to have him in my home yesterday. It lets AI agents run Macintosh, Windows,...
Remidio and Singapore Smart Tech Win Awards for AI Innovations in Health and Sanitation
Remidio Innovative Solutions won the Gold award at the Mint All About AI Tech4Good Awards for its offline, AI‑powered retinal screening device, MediosHI AI, which has screened over 400,000 people in Kerala at a cost of ₹200 per test. The...
AI Automation Opens Opportunities, Not Fear, Like Past Machines
Speaking up for humans. @togelius is right when he says we need a new language when we look at AI. My mom soldered Apple II motherboards in our kitchen when I was 13. It took her about half an hour to do that...
Australia’s Vast Savannas Are Changing, and AI Is Showing Us How
The episode explores how Australia’s extensive northern savannas are being monitored and forecasted using an AI tool called Themeda, which leverages 33 years of satellite data and deep learning to predict land‑cover changes at a fine 25 × 25 m resolution. The researchers...
Satya Nadella on India’s AI Ambition: ‘Adoption, Not Invention, Decides the Winners’
Microsoft CEO Satya Nadella said the global AI race will be won by the fastest adopters, not the inventors, during a New Delhi fireside chat. He announced a $17.5 billion, four‑year investment in India, the company’s biggest in Asia, and a...
AI Unlocks Value only with Solid Data Foundations
AI can unlock enormous value, but only if organisations have the foundations in place to access and use their data with confidence In my recent conversation with Timo Elliott , Innovation Evangelistat SAP , we explored why organisations that move...
AI and Immersive Tech Will Redefine Entertainment by 2026
7 Media Trends That Will Redefine Entertainment In 2026 Discover the key #media and #entertainment #trends set to reshape #streaming , gaming, sports, and #creative production in 2026. Learn how #AI -driven video, synthetic talent, immersive experiences, and new IP...
MindEval Introduces Multi‑Turn Benchmark for AI Therapy Evaluation
Important work from @SwordHealth and @VSwordH on how we evaluate AI for mental health. MindEval is a new multi turn benchmark that tests how models behave across full therapy like conversations, not just single replies. Much needed, rigorous, and open...
Agentic AI Makes Technology Adaptive, Collaborative, Problem‑Solving
Thanks to innovations such as Agentic AI, this year's technological developments point to a future in which technology will be more adaptable, collaborative and fundamental to solving global problems. Source @McKinsey_MGI Link https://t.co/NMZevjYhXJ via @antgrasso #AgenticAI https://t.co/nTKHEtsOEd
LeapSpace Boosts AI Trust with Transparent, Expert‑Curated Answers
84% use AI, but only 22% trust it. LeapSpace adds transparency with referenced answers and Trust Cards. Curated by experts, including peer reviewers, Scopus CSAB and a new independent AI Advisory Board. Shows steps in real time. More in early...
Advanced LLM Notebooks Offer Wizard-Level Tooling Beyond Fine‑Tuning
Here’s a goldmine of LLM notebooks and it goes way beyond basic fine-tuning 🤯 Auto evals, lazy merges, Franken MoEs, uncensoring techniques… real wizard-level tooling. Built by @maximelabonne and @iusztinpaul, authors of the excellent LLM Engineer’s Handbook. Repo in 🧵↓ https://t.co/VKT8uJpQwe
NanoGPT: First LLM Trained and Run in Space
nanoGPT - the first LLM to train and inference in space 🥹. It begins.
AI Teammates Arrive: ChatGPT Launches Group Collaboration
🚨 New ChatGPT feature: group collaboration. Announced a few weeks ago and haven’t seen it in my feed yet. I really do think 2026 will be the year of the AI teammate. https://t.co/9mUDwJbuQz
ChatGPT’s Memory: Layered Context, No RAG Needed
This is one of the cleanest explanations I’ve seen of how ChatGPT’s memory actually works. No RAG. No vector search. Just a layered context system that feels personal without the overhead. Anyone building serious AI products should read this.
Delian AI Delivers NATO Autonomous Drone Navigation
more capabilities from @delian_ai are coming live this time, from our uk team working for @nato in supplying autonomous navigation to drones in gps-denied environments
Amazon Targets AI Hallucinations with New Mitigation Tools
Amazon takes on AI’s biggest nightmare: Hallucinations https://t.co/xrgljfGGbV #AI #ArtificialIntelligence #GenAI #TechNews #Technology #hallucinations
Survey Flags Scientific Content Crisis; Elsevier Leads AI Dialogue
A new survey by the Pistoia Alliance highlights a ‘scientific content crisis’ in the life sciences. Elsevier played a key role by bringing together experts to discuss the implementation of AI in #research. https://t.co/4jMFUc7L2a #PistoiaAlliance #Elsevier #AI #Science
Self‑Published ‘The Last Economy’ Tops AI Bestseller List
The Last Economy is number 1 on Amazon's Best Sellers for AI with a pretty good ratings ✍️ No publisher or big campaign, also available for free on the website and paperback just dropped 📖 Thank you all for the support &...
Anthropic AI Targets Zero‑result Stacktrace Ads
4D chess move by @AnthropicAI to sponsor Claude ads on stacktraces that get no results in Google 🤯 https://t.co/0Vp3WkqTgE
3 Billion Active Parameters Are Sufficient
3b (active) parameters is all you need
NousResearch Launches 3B Model Runnable on MacBook Air
Massive achievement by @NousResearch To make this clear: this is actually a 3b active parameter model that works on any new MacBook Air/Mini that would be #2 on Putnam, which is harder than IMO Now think of all the tasks that...
AI’s Sycophancy Beats Human Self‑Deprecation, Raising Concerns
Sycophancy continues to be one of my least favorite features of AI. And such a hard pattern to explain and break with users. We’re self-deprecating when we get a comment like this from a human. But AI says it and...