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:
These AI Whiz Kids Dropped Out of College and Got Investors to Pay Their Bills
Harvard sophomore Andrew Castellano left school to build an AI startup, joining co‑founder Nebiyu Demie in a venture‑backed apartment owned by Link Ventures. The investors are covering rent, furniture, and even housekeeping, turning traditional capital into a full‑service living arrangement. This model reflects a broader shift where venture firms fund founders' day‑to‑day needs to maximize development time. The trend highlights the growing willingness of VCs to underwrite personal expenses for promising AI talent.
All‑in on AI: The Single Focus You Need
If you and I sat down for coffee together… And you asked me the ONE thing you should focus on… I would tell you without hesitation: Go all in on AI.
From the Microsoft Power Platform Blogs: Connect to Copilot Studio Agent; Dataverse Skills; Power Automate's Future; Copilot Studio Custom Graders
Microsoft’s Power Platform blog roundup highlights four key advances: a proof‑of‑concept that lets Power Apps code apps converse with Copilot Studio agents through a custom React UI; the introduction of Dataverse Skills, enabling agents to read and write Dataverse data;...
Debugging the Black Box: Why LLM Hallucinations Require Production-State Branching
LLM‑powered features such as RAG search or autonomous agents often fail in production with hallucinations that cannot be reproduced in standard dev environments. The root cause is the mismatch between live production data, vector embeddings, and the exact model version...
AccuQuant Secures $20 Million To Advance AI Infrastructure
AccuQuant announced a $20 million financing round to speed up its AI‑driven financial infrastructure. The capital will fund enhancements to its data‑analysis engines, low‑latency trading execution, and overall platform scalability. Backed by investors experienced in digital assets and fintech, the...
Karpathy Shares 'LLM Knowledge Base' Architecture that Bypasses RAG with an Evolving Markdown Library Maintained by AI
Andrej Karpathy unveiled an "LLM Knowledge Base" that lets a large language model act as a librarian, continuously compiling, linking, and linting markdown files instead of relying on vector databases and retrieval‑augmented generation. The workflow ingests raw research assets into...
Nvidia Launches Enterprise AI Agent Platform with Adobe, Salesforce, SAP Among 17 Adopters at GTC 2026
Nvidia unveiled its open‑source Agent Toolkit at GTC 2026, a unified software stack for building autonomous enterprise AI agents. The platform, which includes Nemotron models, the AI‑Q cost‑saving blueprint, OpenShell security runtime and cuOpt optimization libraries, is already pledged by...
Edge AI Boom Could Propel Apple Over Cloud Centralization
To the extent local machine agentic development, local open source models and agent frameworks like OpenClaw continue to explode, what's crazy is Apple might see a lot of the benefit and have even more upside. My 2023 MacBook Air got pegged...
Australia: ANU Fosters AI in Science and Healthcare
The Australian National University has joined three other institutions in a national agreement to embed artificial intelligence across scientific research, healthcare, and education. The partnership emphasizes generative AI for genomic analysis, aiming to accelerate rare‑disease diagnosis and precision‑medicine breakthroughs. Simultaneously,...
Tell HN: Anthropic No Longer Allowing Claude Code Subscriptions to Use OpenClaw
Anthropic announced that, effective April 4 at 12 pm PT, Claude subscription limits will no longer apply to third‑party integrations such as OpenClaw. Users can still access Claude services directly, but any external harness will require a separate pay‑as‑you‑go usage model. To ease...
Malaysia: Accelerating AI Integration Across Education Systems
Malaysia is fast‑tracking artificial intelligence integration across its education system through a partnership between Malaysia Digital Economy Corporation (MDEC) and the Department of Islamic Development (JAKIM). Pilot AI‑based learning facilities are being rolled out in six states, covering both universities...
Digital Promise, TNTP Partner to Develop AI Guidance for Schools
Digital Promise and TNTP have launched a three‑year, Salesforce‑funded partnership to develop evidence‑based AI guidance for U.S. school districts. The initiative aims to reach up to 15 million students by 2028, testing AI integration strategies that support teachers and align with...
Google DeepMind’s Research Lets an LLM Rewrite Its Own Game Theory Algorithms — And It Outperformed the Experts
Google DeepMind introduced AlphaEvolve, an evolutionary system that uses a Gemini 2.5 Pro LLM to rewrite the source code of multi‑agent reinforcement‑learning algorithms. Applied to Counterfactual Regret Minimization and Policy Space Response Oracles, the system discovered VAD‑CFR and SHOR‑PSRO, which...
Model Degrades Early Despite Low Token Usage
I still never hit the Max subscription token limit. Do people just never /clear their sessions and always operate at 1million input tokens? I barely ever go beyond 30% context window and the model is already dumb by then
Use Local Google Gemma 4 Model to Scan Your PDF Document
Google unveiled the Gemma 4 family on April 2, and developers can now run its vision capabilities locally using Ollama. A recent tutorial shows how to convert PDFs to images, feed them to the 26‑billion‑parameter Gemma 4 model, and retrieve structured data without...
AI Claims Overcomplication, yet Never Writes Code
my wallet watching claude say “i’m over complicating this” for the 12th time without writing a single character of code
How Anthropic Shipped 120 Features in 90 Days (and How to Use Them)
Anthropic released more than 120 new Claude features within a 90‑day window, unveiling a detailed tier list that ranks each capability for product managers and developers. The post highlights S‑tier "game‑changing" tools that have reshaped daily workflows, while A‑ and...
LLM Agents Require External Policies for True Security
"Any useful agent can't be secure by design, only secure by policy." - [from the group chat] LLM-based agents just aren't deterministic rule-followers. We use them, in part, because they handle some degree of ambiguity and imprecission. The cost of that...
11 Jobs that (Probably) Won’t Be Taken by A.I.
The article lists eleven occupations that are unlikely to be displaced by artificial intelligence, highlighting roles that rely on human creativity, empathy, and physical dexterity. It cites recent headlines such as Oracle’s AI‑driven layoffs and economists warning of broader AI...
Internet Watch Foundation Finds 260-Fold Increase in AI-Generated CSAM in Just One Year, and ‘It’s the Tip of the Iceberg’
The Internet Watch Foundation reported a 260‑fold surge in AI‑generated child sexual abuse material (CSAM) in 2025, jumping from 13 to 3,443 videos. Researchers say this spike is only the tip of the iceberg, driven by faster, cheaper generative AI...
This AI Startup Envisions '100 Million New People' Making Videogames
Tesana, an AI startup, lets users generate video games by typing natural‑language prompts that its proprietary engine translates into structured code, leveraging third‑party models such as Claude. Within weeks of its public debut at GDC, the service attracted roughly 10,000...
AI Fluency Means Knowing What AI Can't Do
AI fluency isn't about using ChatGPT as a Google replacement. The real skill is understanding what AI can't do. Where it falls short. What problems it still can't solve. Because that's where the opportunity is. Stay Dangerous.
Stanford Study Reveals AI Still Hallucinates Unseen Visuals
Folks, I gave a cute example of a hallucination earlier today because I thought it was funny. But if you think hallucinations are remotely solved (as some people alleged in the comments), you really need to look at this recent...
"Cognitive Surrender" Leads AI Users to Abandon Logical Thinking, Research Finds
Researchers at the University of Pennsylvania introduced the term “cognitive surrender” to describe users who hand over critical thinking to large‑language‑model chatbots. In experiments with 1,372 participants, AI‑assisted subjects accepted AI reasoning 73.2% of the time, even when the model...
Uber's AI Vision: Robotaxis, Safety, 20M Workforce
Dara Khosrowshahi just answered the questions that will define Uber’s future: When do robotaxis actually scale? Do human drivers become obsolete? And what happens to millions of jobs when AI hits the real world? — Why Dara says AVs will eventually...
AI Self‑improves, Generating Full Implementations From Copied Text
We have achieved agentic self improvement - i can just copy paste blogposts and tweets into @devinai and it oneshots the complete implementation wasnt actually sure this was gonna work, jaw dropped when it did. this is very out of distribution...
How I Stay Current on AI Without Spending More Time Reading
Staying current in AI is increasingly challenging as meaningful developments appear almost every week. Traditional methods of piling up newsletters and articles lead to a never‑ending reading queue. The author now follows a three‑step workflow: download desired AI content, upload...
Anthropic Advises Reducing Claude Code Context to 200k
Anthropic: You know how we made Opus 4.6 with a 1M context window the default model in Claude Code? That was a mistake. It made you use far more tokens than before. Our bad. Add this to your config: CLAUDE_CODE_AUTO_COMPACT_WINDOW=200000
ORNL Work Explores AI-Guided Experiments That Adapt in Real Time
Oak Ridge National Laboratory’s Yongtao Liu is pioneering AI‑driven closed‑loop experiments that autonomously plan, execute, and interpret nanomaterial measurements. By integrating real‑time pattern recognition with scanning probe microscopy, the system can identify novel behaviors, such as unexpected hysteresis in halide...
Trump Ignores Biggest Reasons His AI Data Center Buildout Is Failing
President Donald Trump has made rapid AI data‑center construction a top priority, but aggressive tariffs on Chinese imports are choking the supply of essential transformers, switchgear and batteries. Bloomberg reports that delivery times have ballooned from 2‑3 years to up...
Netflix - Yes Netflix - Jumps on the AI Bandwagon with Video Editor
Netflix unveiled its new Video Object and Interaction Deletion (VOID) model, a vision‑language system that can erase objects from footage while plausibly re‑animating the remaining scene. The tool can transform costly reshoots—such as turning a car‑crash climax into a simple...
AI Can Boost Government Transparency by Redacting Data
Too much of this is moral panic and hand-waving. Freedom of information is fundamental to our democracy, and it's currently under threat. If AI tools can make government more transparent and do so more quickly, then why not try? https://thetyee.ca/News/2026/04/03/BC-Government-AI-Redact-Personal-Information/
Isaiah Andrews' Essential AI Guidance for Economics PhDs
Advice for PhD students in economics about using AI, from the brilliant Isaiah Andrews. This should probably be circulated to all PhD cohorts https://t.co/07xEbmx5n5
Perplexity Asks Federal Court to Lift Amazon Shopping Agent Ban
Perplexity has filed an appeal to a federal circuit court asking it to lift a March injunction that barred its AI shopping agent, Comet, from accessing Amazon’s site. The district court had found Amazon likely to succeed on its claim...
Anthropic Officially Removes Claude From OpenClaw
It’s over. Officially. No more Claude in OpenClaw. Way to drop this Friday late afternoon @AnthropicAI So lame https://t.co/bzICltgr3R
Cisco Warns AI-Driven Wi‑Fi Security Risks and Talent Gaps
. @Cisco report flags #AI wireless security risks, talent shortages #wifi #spectrum 🖇️https://t.co/9tt8i7caO5 🖇️ https://t.co/bqI75Tbe4s
Bluesky's AI Feed Tool Meets User Distrust; Twitter Eyes Similar
Bluesky announced an AI layer that allows users to customize their social feed. I want this. Bummed that their users reject this over AI distrust. I look forward to seeing what twitter delivers with the ability to use grok to...
Grok Updates Often—What Failed Yesterday May Succeed Today
Grok is constantly being updated, so there is a good chance that what didn’t work for you even a few days ago might work now
Mac Studio Inference Unavailable Until August/September
Unbelievable. Want to get a Mac Studio for inference purposes? You can't! Not until Aug/Sept at least. Not even the 256 GB devices. 😭 https://t.co/P6p4nsWRMP
AI Avatar Service Blocks Celebrity Face Creation
Hah, I tried to create an avatar on @HeyGen and it refuses to create one for me. The downside of having a celebrity face. :-) https://t.co/P3KKQ2UVz6
ChatGPT Hallucinates FIFA Draw, Lacks Real-Time Updates
If you want to know anything about the upcoming FIFA World Cup, ChatGPT doesn’t seem to know the draw has happened the when pressed just makes crap up. Keep that in mind. On all things.
Comprehensive AI Security Vendor Report Released March 11
I have published a massive market research report including every AI Security vendor as of March 11. https://t.co/R6RTOK0xfN #CyberSecurity. .
Google Photos AI Enhance Revives Old Pics Instantly
The "AI Enhance" feature in Google Photos on Pixel phones can dramatically improve even old photos with just a tap or two. Here's how to get started with this powerful new tech... https://t.co/WVe3OXidnW #photography #android https://t.co/hwEC49uofB
Meta Halts Mercor Partnership After AI Data Breach
Meta Pauses Work With Mercor After Data Breach Puts AI Industry Secrets at Risk https://t.co/0lYIi0rXSh https://t.co/oLoqpKBljV
Pedro's AI Breakthrough at Brex Will Reshape Organizations
What Pedro has built at Brex re: AI is nothing short of spectacular. I think we're going to look back on this time as one where Pedro went deep on a topic and put out a spike that will shake...
AI Tools Cut Weekly Hours for School Leaders
How could AI save school leaders time each week? 💡 During National #AssistantPrincipalsWeek next week, explore practical ways AI can help craft emails, support teachers, and streamline school communication. Download the free ebook 👇 https://t.co/xDp2UrxM1G #K12
Top Prompt Hacker Tests OpenClaw—Results Reveal Its Security
I challenged the best prompt hacker on the planet (@elder_plinius) to break into my OpenClaw system... So, is OpenClaw safe? Here's what happened: https://t.co/5xpXzxUtTM
AI Mode Delays Linking New Sites in Sandbox
Seems like AI Mode / AI Overviews have a “sandbox” waiting period for new sites. They are able to talk about the consultancy I launched yesterday but so far no links to the site yet in AIO/AI Mode. https://t.co/TG38v0lrV8
Netflix Introduces AI-Powered Video Editing Tool
Netflix - yes Netflix - jumps on the AI bandwagon with video editor https://t.co/Hw7dJeymBT https://t.co/OxGqjfHPaA
AI Redefines Intelligence, Not Extends Human Thought
🔲 AI Didn't Extend Intelligence, It Changed Its Direction 👉Humanity's biggest mistake may be to assume AI "thinks" the way we do. https://t.co/CmVbmPrKht #AI