Today's AI Pulse
OpenAI launches ChatGPT Images 2.0 with multilingual text and agentic reasoning
OpenAI has introduced 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 and create floor plans, maps and UI mock‑ups. The rollout expands OpenAI’s generative‑AI capabilities beyond text‑only outputs.
Also developing:
Taiwan Tech Startups Carving Out Critical Role in AI Platforms
At CES 2026, Taiwan’s tech ecosystem demonstrated a shift from platform concepts to real‑world AI deployment. Guided by the National Science and Technology Council and Taiwan Tech Arena, 57 startups partnered with 83 local supply‑chain firms to showcase edge AI, health monitoring, smart manufacturing, and green‑energy solutions. Their joint‑operation model leverages Taiwan’s long‑standing ICT manufacturing expertise to deliver end‑to‑end, regulatory‑compliant products. Successes such as CES Innovation Awards and strategic alliances with giants like Foxconn underline the startups’ growing role in making AI operational at the edge.
Philips Receives US FDA 510(k) Clearance for AI-Powered Spectral CT Verida System
Philips has obtained FDA 510(k) clearance for its Verida spectral CT system, the world’s first AI‑powered detector‑based spectral CT platform. The system combines always‑on spectral imaging with deep‑learning reconstruction, delivering higher image quality, lower noise, and enhanced tissue characterization. It...
Hey Orwell: Thousands of AI‑written, Edited or ‘Polished’ Books Are Being Sold
A class‑action settlement will force Anthropic, the creator of Claude, to pay up to $1.5 billion for allegedly infringing thousands of authors' copyrights. The article shows how AI chatbots can not only regurgitate content but also imitate an author's distinctive voice,...
Qwen Team Open-Sources Qwen3.6-35B-A3B: A Sparse MoE Vision-Language Model with 3B Active Parameters and Agentic Coding Capabilities
Alibaba’s Qwen team has open‑sourced Qwen3.6-35B-A3B, a 35‑billion‑parameter vision‑language model that activates only 3 billion parameters per inference thanks to a Sparse Mixture‑of‑Experts design. The architecture uses 256 experts with eight routed per token, linear‑attention Gated DeltaNet blocks and Grouped Query...
Why Anthropic's Custom Chip Plans Could Benefit Broadcom
Anthropic, a fast‑growing large language model developer, saw its annual revenue run rate jump from $9 billion to over $30 billion between late 2025 and early 2026. To support this expansion, it is using Broadcom’s TPU‑based AI compute, accessing 3.5 GW of capacity...
TripWorks Flags Risks From AI-Built Booking Platforms Amid Reliability Concerns
TripWorks warned that AI‑generated booking platforms are proliferating faster than their operational foundations can support, leading to accuracy gaps and failure under real‑world stress. While generative AI can produce polished demos, many new systems lack the backend resilience needed for...
Future Skill: Knowing What to Delegate to AI
Stop scrolling. This is bigger than “AI tools.” Codex is turning into a system that can actually do the work for you. It learns how you operate… then starts operating on your behalf. Emails, files, tasks, workflows handled. Not assisted. Executed. We’re moving from typing...
Scaling Regulated Data Workflows Without Lock‑In - with Juan Orlandini of Insight
In this episode, Juan Orlandini, CTO of North America at Insight, explains how finance leaders can modernize chaotic, regulated data environments by integrating AI thoughtfully rather than layering it on outdated systems. He stresses that generative AI excels at pattern...
Self‑Driving Labs Let AI Lead Scientific Discovery
AI Is Becoming A Scientist: How Self-Driving Labs Will Accelerate Discovery #AI is moving beyond assisting #scientists and taking an active role in #discovery , with self-driving labs that can design #experiments , run tests and learn from results. This...
Turn Your Email Newsletters Into a Morning Audio Brief
I built something. A daily personalised audio newsletter briefing. The host tells me what came in, what the interesting ideas were & how I might be able to apply it. It gets delivered at 6am each morning and I just sit with...
Hippocratic AI Launches Safety-First Voice AI for Patients and Nurses
Healthcare AI startup Hippocratic AI unveiled two voice‑driven solutions: AI Front Door, a patient‑facing omni‑topic agent, and Nurse Co‑Pilot, a bedside assistant for inpatient nurses. AI Front Door consolidates scheduling, lab results, billing and care follow‑ups into a single, relationship‑based...
GM Details Unified Camera Stack for Driver Assist
General Motors unveiled a unified vehicle‑anchored geometry platform that powers its Top‑Down View and Transparent Trailer driver‑assist systems. By treating wide‑angle cameras as geometric sensors and projecting their pixels into a common rear‑bumper‑ground frame, the stack delivers consistent perception across...
Food Delivery for $1 an Order? Bank Research Says AI Will Make It Possible
Barclays research predicts autonomous food‑delivery robots and drones could cut per‑order costs to $1, down from the current $5‑7 range. The cost reduction could unlock a $16 billion annual profitability pool for global delivery platforms. Barclays estimates autonomous‑delivery penetration will rise...
Xpeng Pushes Physical AI Pitch at Beijing Auto Show
Xpeng will use the Beijing Auto Show opening on April 24, 2026 to rebrand itself as a “Physical AI” group that unites intelligent driving, humanoid robotics and aerial mobility under a single technology stack. The company highlighted its VLA 2.0 intelligent‑driving...
Claude Mythos Preview Demonstrates Advanced Cybersecurity Capabilities
Anthropic’s Claude Mythos Preview, released on April 7, was evaluated by the Government’s AI Security Institute (AISI). In capture‑the‑flag challenges it solved expert‑level tasks 73% of the time, the first AI to achieve that benchmark. It also completed three of ten...
Tight Bounds Reveal Optimal Inference Complexity for Quantum Kernels
Optimal algorithmic complexity of inference in quantum kernel methods for classical data. Quantum kernel methods are among the leading candidates for achieving quantum advantage in supervised machine learning. A key bottleneck is the cost of inference: evaluating a trained model on...
TPO Converts RL to Supervised Learning for Sparse Rewards
Target Policy Optimization (TPO) TPO turns GRPO into supervised learning: build a target distribution over sampled completions, then fit with cross-entropy. TPO brings the old “reweight, then fit” idea from the RL-as-EM line of work (e.g. REPS, MPO) into group RL....
Lawyers to Survive AI
Sir Geoffrey Vos, the second‑most senior judge in England and Wales, warned that AI‑driven tools will soon give clients free access to statutes and precedents, reshaping the lawyer‑client relationship. He argued lawyers will remain essential, but their role will shift...
Google AI Mode Traps Users Inside Its Walled Garden
Ugh. Google's new AI Mode no longer sends users directly to other sites when they click a link. Instead AI Mode *stays open* while it opens the website in an adjacent window. The user may see your site, but they...
Disney Turns Storytelling Into Character‑Driven Robotics
Disney is not “becoming” an AI and robotics company. In some very important ways, it already is. That is what I think many people still miss. For decades, Disney mastered something most tech companies never truly understood: how to make movement feel alive. A pause....
How to Build a High-Performing Team During the AI Era
Deloitte’s new research of 1,394 U.S. professionals shows high‑performing teams are far more likely to embed AI in daily work—78% versus 54% for average teams. Those teams also report stronger outcomes in efficiency, problem‑solving and collaboration, driven by human capabilities...
Llama Vs. ChatGPT: Read My Expert Take Before Using Them
The article pits Meta’s open‑source Llama series against OpenAI’s ChatGPT, testing them across summarization, content creation, creative writing, coding, image generation, image analysis, and real‑time research. ChatGPT consistently outperformed Llama on ready‑to‑use creative tasks, delivering polished code, photorealistic images, and...
The Playbook for System Prompting
The article reframes system prompting as a form of systems intervention rather than simple instruction writing. It argues that prompts condition a probabilistic model’s attractors, feedback loops, and resistance dynamics, which explains why surface‑level prompt tweaks often fail. The piece...
DoubleVerify Creates AI SlopStopper for Social to Tackle Low-Quality AI Content
DoubleVerify has launched AI SlopStopper, a new tool within its DV AI Verification suite designed to identify and block low‑quality AI‑generated content on social and video platforms. The solution blends proprietary AI detection with human oversight, feeding results into the...
Welcome to Agentic AI. Welcome to Per-Agent Licensing
Microsoft is rolling out a per‑agent licensing model for its AI tools, bundling Copilot and new Agent 365 features into a subscription that charges customers for each AI sub‑routine they run. The approach departs from the traditional compute‑or‑storage consumption model used...
Stop Using AI as a Brainstorming Tool
The post argues that asking AI tools like ChatGPT to "think outside the box" is a misuse, because large language models generate the most probable text rather than genuine novelty. It explains that true creativity stems from juxtapositional thinking—holding two...
AI Self‑checks Iterate Outputs for Higher Quality Thumbnails and Titles
When I build skills now I always ask AI to spin up a separate eval agent to do yes/no checks to grade the first agent's output. And if the output isn't "yes" across the board it asks the first agent to...
Claude's 1M Token Window Just Went Free
Anthropic has lifted Claude’s context window to one million tokens on its free tier, removing the beta header and any extra charge. The expansion turns an effective usable space of roughly 110,000 tokens into about 830,000 before compaction kicks in,...
StradVision Wins India Commercial Vehicle ADAS Deal
StradVision has been chosen by an unnamed global commercial‑vehicle OEM to roll out its SVNet AI perception software across the OEM’s India lineup, targeting a fleet‑scale ADAS deployment. The platform will provide automatic emergency braking, forward‑collision warning and lane‑departure warning...
AI on Cloud: From Pilots to Enterprise Scale
Enterprise AI is moving from a pilot‑centric rush to a focus on production‑scale value. Leaders note that while pilots are easy, embedding AI into core processes proves difficult because many firms treat AI as an add‑on rather than redesigning workflows....
Cisco Bets on AI Waste to Fuel 70% AI Code
Something I'm hearing more often: AI wastage is necessary, at least for a while. Cisco's @jpatel41 says you gotta let people burn through tokens to build the instinct. Now has its first product with zero lines of human code and...
Anthropic’s Latest Model Is Deliberately Less Powerful than Mythos (and That’s the Point)
Anthropic released Claude Opus 4.7, a deliberately less powerful upgrade to Opus 4.6 that prioritizes safety over raw capability. The model adds stronger coding assistance, tripled vision resolution, better memory across sessions, and enhanced financial‑analysis output. Anthropic positions Opus 4.7...
AI Scraper Detected, Account Suspended without Login
I’ve seen a few posts about building AI-powered post scrapers for the business-focused social platform that shall remain nameless and every time I see one I feel like I need to do a PSA about that platform’s bot detection Gather round,...
Leadership Now Needs AI Teammates, Not Middlemen
The End of the Middleman: Why Modern Leadership Requires an AI Teammate by @Timothy_Hughes https://t.co/sW1esAoOlz @DLAIgnite #SocialSelling #DigitalSelling #Sales #SalesTips #SalesLeader #Salesforce #SalesEnablement #Marketing #Leadership #ArtificialIntelligence #TechNews https://t.co/a0Xo6pROQe
Quickplay Rolls Out Gray Media’s AI‑Powered OTT Platform in 113 U.S. Markets
Quickplay announced today that Gray Media’s next‑generation, cloud‑native streaming platform is live in all 113 of the broadcaster’s legacy markets. Built on Google Cloud AI, the system unifies more than 1,300 digital touchpoints and serves 37% of U.S. TV households,...
Free Chatbot APIs That Rival ChatGPT
Exploring the landscape of free chatbot APIs reveals some impressive contenders that rival ChatGPT, providing developers with cost-effective solutions. Discover the true free options available. https://t.co/HqZlJi94ci
Claude Mimics Developers: Native JS Promises, Endless Edge Cases
Claude testing my patience today after it happily declared "lets just build X feature with native JS we don't need that library" and then we run into a million small edge cases. In other words, it's acting just like a real...
Mediavine and Index Exchange Link 18,000 Publishers' Data to Index Marketplaces
Mediavine announced an expanded partnership with Index Exchange, enabling audience segments from its network of more than 18,000 independent publishers to be accessed through Index Marketplaces. The deal makes Index Exchange the first SSP to offer Mediavine’s curated data via...
AI's Eerie Habit:
Isn't it scary when Opus 4.7, while deciding to give the answer, tries to figure out what my intentions are? https://t.co/DrKxL3yb5W
AI Revolution Shifts From Banks to Citizens, Including Government
New on the not-a-newsletter: Joined Up Government https://t.co/uQuV5muDqI I have long bored people senseless with repeated claims that the revolution in financial services is customers getting AI rather than banks getting AI. Well, the same may well be true of government. Good....
Accenture Buys Spanish AI Firm Keepler, Adding 240 Experts to Its Data Practice
Accenture announced the acquisition of Keepler, a Spanish data and AI consultancy, bringing a 240‑person team in Madrid, London and Lisbon into its AI and data analytics practice. The deal, terms undisclosed, deepens Accenture’s end‑to‑end AI offering and positions it...
OpenAI Launches GPT‑5.4‑Cyber via Expanded Trusted Access
.@OpenAI releases GPT-5.4-Cyber, beefed-up Trusted Access for Cyber program https://t.co/Cqe6QG6I3Z OpenAI released GPT-5.4-Cyber, a model that will be released to select participants in a beefed-up Trusted Access for Cyber (TAC) program. https://t.co/pvHgkQtziD
Built Our Own E2e Gemini Tests, GBrain Voice Open‑Source Soon
My claw and I searched high and low for proper e2e Gemini Live tests and in the end we decided to do it ourselves Coming to GBrain Voice, open source release soon. https://t.co/kQOloJS9c0
AI Brand Crisis: Choose a Stance or Lose
Pick a Side or Get Left Behind: The Looming AI Brand Crisis by @Timothy_Hughes https://t.co/Ab8RLnV6pu @DLAIgnite #SocialSelling #DigitalSelling #Sales #Leadership #Marketing #MarketingStrategy #MarketingSuccess #ArtificialIntelligence #AI #ContentMarketing #MarketingTips #MarketingEthics
Robot Gains True First-Person Vision Capability
Seeing Through Its Own Eyes: Building a #Robot with True First-Person Vision via @WevolverApp #AI #MachineLearning #Robotics #ArtificialIntelligence #Innovation #Technology https://t.co/ogTgNcCfV6
AI Marketing Still Missing Allen Iverson Branding
I cannot believe that no company has used AI [Allen Iverson] in their a.i. marketing campaigns yet.
AI's Future Depends on Robust Botanical Data
#AI needs solid botanical #Data more than ever by Chris Bivins @Nature Learn more: https://t.co/6U4taCtOTq #MachineLearning #ArtificialIntelligence #ML https://t.co/MTS7gowYpS
Compute Shortage Drives AMD, Amazon, Cerebras Deal Rush
Chatter being $AMD could get a deal with Anthropic which makes total sense given the compute shortage. I have been saying for a while I think Amazon is a good candidate also for Helios. Bottom line, all available viable compute will...
Qlik Introduces Data Trust Scores for AI Agents
.@Qlik aims to gauge trust of the data underneath agentic AI https://t.co/imj2bAfdYz Qlik is looking to give the data used by AI agents a trust score to make agentic systems more reliable. https://t.co/onnPfAySF1
Opus 4.7 Delivers Stunning Interactive 3D Tower of Babel
With max thinking Opus 4.7 is quite impressive, with a real sense of style In two prompts: "implement the Tower of Babel, in 3D, in as sophisticated and visually interesting a way as possible. It should be interactive" and then "make...