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 Details AI Push Against Harmful Ads
Google announced that its Gemini‑powered AI models are now filtering harmful ads more effectively, removing 438.6 million ads and suspending 594,000 advertiser accounts in Australia during 2025. The same technology blocked over 8.3 billion ads worldwide and shut down 24.9 million accounts, including 602 million scam‑related ads. Gemini analyzes hundreds of billions of signals to detect intent‑based violations, enabling near‑real‑time review of Responsive Search Ads. The AI upgrade also cut incorrect suspensions by 80 percent and quadrupled the handling of user‑reported issues.
K1x Secures $175 Million for Private Markets AI Tax Tech
K1x, an AI‑native private‑markets tax data platform, secured a $175 million growth investment led by Sumeru Equity Partners, with Edison Partners also participating. The transaction makes Sumeru the majority shareholder and provides capital to deepen K1x’s AI‑driven platform, accelerate product development,...
Nvidia's AI Demand Shifts to Massive Deployment Wave
Nvidia's AI story is getting a major upgrade. Analysts are now convinced as demand shifts from just training models to the massive, long-term phase of deploying them everywhere. This "second wave" could be even bigger. AI
LLMs Finally Deliver Slack’s Long‑awaited AI Vision
Every platform builds on its predecessors, and I'm now using Claude for so many things I'd hoped we'd enable on the Slack platform a decade ago. The vision really did slap, and modern LLMs have now made the vision a...
CoSN 2026: AI Strategy Should Start With Goals for Student Skills
At the CoSN 2026 conference, experts urged K‑12 leaders to reverse the common tools‑first mindset and start AI planning with clear student skill goals. Chris Agnew of Stanford’s AI Hub for Education emphasized that districts should define desired graduate competencies,...
Singapore: Advancing Design AI to Drive Human-Centric Innovation
Singapore is transitioning AI from experimental pilots to purposeful, design‑led integration across both large enterprises and SMEs. Minister of State Rahayu Mahzam emphasized that organisations now focus on how AI can drive productivity, transformation and new value streams. Government programmes...
Australia: UNSW Develops AI Companions to Support Student Well-Being
UNSW researchers have created prototype AI companions, Tom and Mia, to support student well‑being through on‑demand, screen‑based conversations. The digital characters are bilingual in English and Mandarin, targeting isolation among domestic and international students. Designed as a “skilled friend” rather...
India Advances AI Governance Architecture and Financial Integrity
India has launched two high‑profile institutional initiatives to tighten governance in artificial intelligence and financial integrity. The Ministry of Electronics and Information Technology created the AI Governance and Economic Group (AIGEG), a senior inter‑ministerial body chaired by Minister Ashwini Vaishnaw, to...
Could a Digital Twin Make You Into a 'Superworker'?
Richard Skellett’s three‑year project, “Digital Richard,” is a personal AI twin that ingests his meetings, documents and calls to replicate his decision‑making style. Bloor Research has rolled the technology out to its 50‑person global workforce, allowing employees to offload tasks...
Worldly Tackles Audit Turmoil with AI
Worldly unveiled an AI‑driven Supplier Compliance Management platform to centralize social audit data for apparel and consumer‑goods brands. The solution aggregates third‑party audits, brand‑specific assessments, and Higg Facility Social Labour Module inputs, then automatically aligns findings with each brand’s Code...
Even AI Can't Automate My Expense Reports—Need an Agent
We have AI writing code, generating content, analyzing data, and yet I am still doing expense reports manually. Feels like one of the most obvious use cases for an AI agent and yet... If someone has and AI agent that could...
LLNL Combines Machine Learning and 3D Printing for Shockwave Control Experiments
Lawrence Livermore National Laboratory, Imperial College London and collaborators have used a machine‑learning optimization algorithm together with polymer 3D printing to create a void‑filled target that suppresses the Richtmyer‑Meshkov instability during shock‑wave experiments. The AI‑designed void geometry reshapes the incoming...
SAP Embeds AI Across CX in Q1 2026 Release
SAP’s Q1 2026 Customer Experience release embeds artificial intelligence directly into core sales, service, commerce and marketing workflows, turning AI into an execution fabric linked to ERP, supply chain and finance. The update adds features such as an Outlook email‑to‑quote add‑in,...
The AI-Driven Shift in Vulnerability Discovery: What Maintainers and Bug Finders Need to Know
AI‑powered code models are now able to locate real software vulnerabilities with minimal prompts, dramatically increasing the volume of reports to open‑source projects. The surge includes a flood of low‑impact, often invalid findings that consume hours of analyst time, while...
Pair Codex Backend with Claude Code Frontend for Polished UI
My general rule for AI coding agents is simple: use Codex for the backend and Claude Code for the frontend. Why? Codex is written in Rust, and Claude Code is written in TypeScript. In practice, that means you let Codex build the...
AI as the Defender: Reinventing Proactive Cybersecurity Through Intelligent Automation
Artificial intelligence is reshaping cybersecurity by acting as a force multiplier rather than replacing human analysts. Tenable and peers define "AI for security" as the use of machine learning to automate analysis, amplify detection and improve decision‑making, while "security for...
Longing for Interpretable Models Over Opaque Modern AI
I really miss the good old days of linear and logistic regression, decision trees, random forests (kind of), SVMs, XGBoost (kind of), ARIMA, survival models, and similar methods; back when so-called AI was still something you could actually explain.
GPT 5.4/Codex Delivers Robust Solutions, UI Lags Behind
GPT 5.4/codex is honestly solid. Been using it for a bit now and while it's not as fast as what Opus 4.6 was a month ago, couple key things: 1. It's methodical, it uses evidence to come up with a comprehensive...
OpenAI's Big Codex Update Is a Direct Shot At Claude Code
OpenAI announced a major upgrade to its Codex platform, adding agent‑like capabilities that let the model operate macOS desktop applications, browse the web, and generate images. The update also introduces new plug‑ins for GitLab, Atlassian Rovo and Microsoft Suite, plus...
Aged Care Tool Inquiry
The Commonwealth Ombudsman has opened an investigation into the Integrated Assessment Tool (IAT), an algorithm used for aged‑care assessments, after receiving 834 formal complaints. Senators highlighted that the tool, mandatory since November 2025, prevents human assessors from overriding its outcomes,...
First Serious Product Work with OpenCode Beats
This might be the inflection point where I attempt to do serious product work™ using OpenCode and a local model for the first time. Didn't have the bandwidth to experiment as Claude and Codex already exist. But this is alluring. It did...
The Honest AI Onboarding Curve Nobody Tells You About
Small businesses adopting AI agents face a steep onboarding curve that initially hurts productivity. In weeks one and two, output quality drops and speed slows as the agent learns company‑specific workflows, often achieving only about 60% accuracy. By week four,...
Opus 4.7 Uncovers Hidden Health and Music Patterns
Opus 4.7 with Apple Health is wildly fun. So far it’s guessed my social activity from noise levels. It noticed an asymmetry in my walk during a certain week (had knee pain). It’s asking why I stopped listening to music a...
US Nuke Scientists Are Mysteriously Disappearing! - New World Next Week
In this episode of New World Next Week, hosts James Corbett and James Palato discuss Anthropic’s AI model Mythos, a zero‑day vulnerability generator that’s being quietly tested by major tech firms under “Project Glass Wing,” and the unsettling pattern of...
Deconstructing the Mythos Myth
In 2019 OpenAI chose not to release its 1.5‑billion‑parameter GPT‑2 model, citing fears it could generate deceptive, biased or abusive content at scale. Instead the lab offered a less powerful version to researchers while publishing extensive technical details. The decision...
AI Does McKinsey‑grade Research, Drives Millions per Lead
You can run McKinsey-level research for almost nothing to generate millions in revenue. Here's what our Single Brain actually generates for every single lead that books a discovery call - before anyone touches it: 1. Full company dossier: revenue, headcount, tech stack,...
Mozilla Throws Thunderbolt at Enterprise AI Providers
Mozilla’s MZLA subsidiary unveiled Thunderbolt, an open‑source AI client aimed at enterprises seeking data‑sovereignty. The client integrates with deepset’s Haystack platform and supports Model Context Protocol and Agent Client Protocol standards, allowing firms to run any LLM on‑premise or in...
Unintentional AI Adoption Is Already Inside Your Company. The Only Question Is Whether You Know It.
In‑house counsel are confronting AI that has already seeped into daily workflows through browsers, email extensions, and personal devices, often without any formal approval. This unintentional adoption creates a hidden layer of corporate memory built from prompts, scraped emails, and...
Discover Matt's Cutting-Edge Research at Stanford Seminar
It will be great learn about Matt 's latest work at our Stanford Digital Economy Lab lunch seminar. I first got to know Matt back at the Massachusetts Institute of Technology when he was doing his PhD and he's continued...
Gain Full Visibility Into AI Spend with Ramp
Anthropic made its first dollar three years ago. last month it crossed $30B in revenue. that money is coming from somewhere, and your CFO probably can't tell you where. the problem isn't the spending. the companies on Ramp investing the most...
OpenAI Starts Offering a Biology-Tuned LLM
OpenAI unveiled GPT‑Rosalind, a large language model fine‑tuned for biology workflows. Trained on 50 common biological tasks and public databases, it can suggest pathways, prioritize drug targets, and connect genotype to phenotype. The model is deliberately more skeptical to curb...
Happy About Codex Switch, but Fear Rate Limits
I am happy everyone is switching to Codex, but Tibo if you start rate limiting me or making me use worse models...
Claude Falls Short: Poor Voice, Weak Answers
Been trying Claude and what are all you talking about? Voice entry terrible. (Only conversation mode on desktop.) Answers not great. Okay, so a little shorter responses, less bullet-point salad, and antithesis, but it’s not even 10% better than ChatGPT (if at all)
Army’s Work With OPV Black Hawk To Inform Bringing Autonomy To MV-75 FLRAA
The U.S. Army has accepted the first H‑60Mx Optimally Piloted Vehicle (OPV) Black Hawk, equipped with Sikorsky’s MATRIX autonomy suite, to serve as a testbed for autonomous capabilities on the upcoming MV‑75 Cheyenne II FLRAA tilt‑rotor. Recent DARPA‑originated demonstrations, including the...
Professional Identity Vs. Voice: Key AI Assistant Layers
What is the difference between a “professional identity” layer and a “voice” layer when setting up an AI assistant? https://t.co/RQDmR8BMQn via @DLAIgnite #SocialSelling #DigitalSelling #Leadership #ArtificialIntelligence #Tech #TechNews #AgenticAI #AIAgents #AI #FutureOfWork #Strategy #Sales #Marketing
Agentic AI Automates Zendesk Tickets, Tackles HIBP API
I've spoken a bunch recently about experiments with agentic AI to do genuinely useful stuff, such as "Bruce", who now replies to most of our Zendesk tickets. So, what happens when you let it loose on the HIBP API? Some...
Fewer Federal Workers, Same Mission: Why AI Is the Productivity-First Technology Critical to Agency Operations and Efficiency
The federal government is confronting a shrinking workforce—317,000 employees departed in 2025—while still tasked with delivering services. Budget pressures have limited overall spending cuts, but IT outlays rose to $126 billion in 2024, prompting agencies to turn to artificial intelligence for...
Companies Must Onboard AI Like Human Coworkers
AI is supposed to represent our new “co-workers,” right? Well, it would be nice if organizations took the time to help us get to know our new co-workers, which doesn’t seem to be the case. (My latest in Forbes) @forbes...
Seeking a Mid‑range LLM: Sonnet as the Sweet Spot
Claude Opus 4.5 is expensive GPT 5.4 is expensive MiniMax2.7 is cheap, but it can't code well. What's the happy medium? Sonnet?
New Guidelines Recommend AI-Based Breast Cancer Risk Assessments
The National Comprehensive Cancer Network’s 2026 Clinical Practice Guidelines now recommend image‑based artificial‑intelligence risk assessments as a primary tool for breast cancer screening. The guidance advises using AI‑derived five‑year risk scores from routine mammograms, with a 1.7 % risk threshold prompting...
Claude Windows App Fails, Users Threaten Switch to ChatGPT
The latest iteration of @anthrropic's Claude desktop app for Window is so bad. Truly shIte. Better to roll it back. Yet, the company has no obvious window for product feedback. Get it together @anthropic or you'll losemore folks to @OpenAI's @ChatGPTapp
Networked AI Shows Small Cables Power Big Systems
The Network always wins. (in big systems) Hook up small things with cheap cables. Distributed software is hard, needs to really work. Very cool how this is working out @tenstorrent Networked AI
Lack of Governance Coordination on AI Costing Companies
Grant Thornton’s 2026 AI Impact Survey reveals that weak governance and compliance barriers are the top reason—cited by 46% of respondents—AI projects fail or underperform. Although 75% of organizations continue to approve major AI spend, nearly half have not set...
Lead Marketer Demonstrates Claude Code Setup and Workflows
Proud moment 🥲 Our lead marketer, completely unprompted, shared a video about how she set up Claude Code from scratch and showed a few workflows for client work. https://t.co/C0KlrVuHpA
Old Model Cut, Seek New Security Research Options
Until they cut off the old model like they just did for an earlier version. Time to start looking at alternatives for security research if that doesn’t change.
Blue Owl Builds a Capital Platform for the Hyperscale AI Era
Blue Owl Capital has transformed from a real‑estate investor into a dedicated capital platform for hyperscale AI data‑center projects. The firm integrated IPI Partners in January, adding over $11 billion in assets and appointing a digital‑infrastructure head. In May, its Digital...
AI Shifts Value to Human Persuasion, Coordination, Trust
The work people often dislike may be becoming more valuable. As AI takes over tasks, human roles like persuading, coordinating, and building trust are gaining importance. The edge is shifting. What feels inefficient today may be what keeps...
Waymo Matches Human Drivers' Safety, Despite Underreported Crashes
Waymo in SF has comparable safety rate to UberLyft human-driven rides in SF, says Cardozo prof Matthew Wansley, citing academic study, but notes the human-driven crashes in this set are likely underreported. He thinks AVs are making a “positive contribution”...
AI Labs Purchase Defunct Startup Communications, Raising Privacy Concerns
AI labs are buying internal communications of defunct startups to train their agents. Emails, Slack archives, etc. Personally identifiable info is removed by data resellers. But how would you feel knowing your former board/CEO is selling your comms to recover...
Opus 4.7 Improves, yet Anthropic Releases While Mythos Stays Hidden
Opus 4.7 is a big improvement...but also Anthropic is ok releasing it but not Mythos? I'm so confused... https://t.co/OVBq4ZWKyH