Today's AI Pulse
Anthropic fixes three missteps that briefly dulled Claude Code
Anthropic disclosed three separate changes that temporarily reduced Claude Code's performance and restored the model by April 20. A March 4 switch lowered the default reasoning effort from high to medium, which was reverted on April 7, and a March 26 cache‑clearing bug erased session context, causing forgetfulness and higher token usage until fixed. The issues were fully resolved by April 20.
Also developing:
By the numbers: Tesla to acquire AI‑hardware firm for up to $2B
Mark Zuckerberg Is Excited to Add More AI Content to All Your Social Feeds
Meta CEO Mark Zuckerberg announced that the company will significantly increase AI‑generated posts in users' feeds, framing it as the next era of social media. Meta is embedding AI tools across its apps and piloting dedicated AI‑focused platforms such as the Vibes app, which has already produced over 20 billion AI‑generated images and offers AI‑generated video feeds similar to OpenAI’s Sora. CFO Susan Li highlighted Vibes as a new content type enabled by AI, signaling further expansion into novel AI‑driven formats. The quarter posted $51.24 billion in revenue, a 26% year‑over‑year rise, offset by a one‑time $15.93 billion tax charge.
Meta Denies Torrenting Porn to Train AI, Says Downloads Were for “Personal Use”
Meta has moved to dismiss a lawsuit filed by Strike 3 Holdings that accuses the company of illegally torrenting about 2,400 adult films to train an undisclosed AI model, seeking damages exceeding $350 million. Meta argues the downloads, spanning seven years and...
Young Engineer's Cognitive Architecture Outpaces Leading AI Models
How did a 20-year-old beat Grok. And ChatGPT. And Perplexity. And Gemini? In part one @blevlabs laid out how: he built a very different architecture to take AI to a new human level. Cognitive architectures, he calls them. I call...
How One Mountain Town Hopes AI Can Help It Fight Wildfires
Vail, Colorado has become the first U.S. municipality to deploy Hewlett Packard Enterprise’s AI‑driven Smart City solution, built with partners Kamiwaza, Nvidia and others, to automatically detect wildfires from existing camera feeds, drone and satellite imagery. The platform uses real‑time...
Optimistic Yet Cautious: Balancing AI Progress and Safety
Zoomer or Doomer? Matt Wolfe (@mreflow) on finding the middle ground between optimism and caution in AI. “I probably lean more optimistic, my whole career is sharing AI news and tools, but I still fall somewhere in between.” “We should push forward as...
Meta Lacks Leadership to Rival OpenAI in Consumer AI
Another new thesis. Meta won’t be able to compete with @OpenAI as it enters consumer electronics. The Meta/Raybans debacle and the way it treats the community and how it laid off 600 AI workers tells me it doesn’t have the leadership...
Inside YouTube's Big AI Reorg
YouTube CEO Neal Mohan announced a major AI-focused reorganization, the first leadership overhaul in ten years, aimed at embedding artificial intelligence across the platform. The memo highlighted AI as the "next frontier" and introduced a voluntary exit program for U.S....
Vibe Coding Platform Cursor Releases First In-House LLM, Composer, Promising 4X Speed Boost
Cursor, the Vibe coding platform from Anysphere, unveiled Composer, its first in‑house coding large language model, as part of the Cursor 2.0 update. Composer is a reinforcement‑learned mixture‑of‑experts model that generates code at 250 tokens per second—about four times faster...
Worried About Your Email Password? Surfshark Just Dropped a Feature that Means You'll Never Stress About Phishing Again
Surfshark has launched an AI‑powered email scam checker that scans Gmail messages for phishing cues, currently available as a Chrome extension for Surfshark One and One+ subscribers. In a TechRadar test, the beta tool flagged 17 of 20 suspicious emails, displaying...
Y Combinator Sharpens Pitches Across Diverse SF Startups
The startups in San Francisco are getting much more interesting. Here's 10 who pitched last night at @jowyang's "Llama Lounge" event. If you are thinking of starting a company notice how tight they are at explaining their companies, which are...
👾 Formula 1's Unexpected Edge: Why Racing Matters More as Cars Drive Themselves
As autonomous driving advances, Formula 1 is experiencing a counterintuitive boom by leaning into human drama and high‑performance spectacle rather than technical obsolescence. F1 has modernized operations—shifting to a hybrid broadcast model that halved on‑site staff and leveraging partners like...
AI Turns Holiday Shopping Into Instant Conversational Recommendations
3... 7... 20...? How many Labubus are you buying this year? Holiday shopping is right around the corner, and it will look completely different this year. Alex Sherman, CEO of @Bluefish_AI, on how AI is changing online shopping forever: “Last year,...
Augment Code Rewrites Entire Repos Faster Than Cursor
Forget Roo, Cline, Windsurf, Cursor. Augment Code keeps raising the bar. Tried it on https://t.co/GxGvh6jnPC, a fun project I hadn’t touched in months. Instead of random refactors, it rebuilt the architecture coherently, understanding the whole codebase → structure, dependencies + Git history. And where...
Anthropic Scientists Hacked Claude’s Brain — and It Noticed. Here’s Why That’s Huge
Anthropic scientists injected specific concepts into Claude’s neural activations and asked the model if it noticed anything unusual, finding that the system sometimes reported the injected thought, demonstrating a rudimentary introspective capability. In controlled tests, Claude Opus 4 and Opus 4.1 succeeded...
2D Gaussian Splatting Delivers Real-Time Geometry‑Accurate Rendering
🚀2D Gaussian Splatting: Real-Time, Geometry-Aware Radiance Field Reconstruction In this week’s deep dive, we unpack how 2D Gaussian Splatting (2DGS) redefines the future of real-time neural rendering and reconstruction. By collapsing volumetric 3D Gaussians into surface-aligned 2D disks, 2DGS achieves unprecedented...
Launch Your AI Career with PyTorch Certificate
An exciting new professional certificate: PyTorch for Deep Learning taught by @lmoroney is now available at https://t.co/zpIxRSuky4. This is the definitive program for learning PyTorch, which is one of the main frameworks researchers use to build breakthrough AI systems. If...
ElevenLabs CEO Says AI Audio Models Will Be ‘Commoditized’ over Time
ElevenLabs co‑founder and CEO Mati Staniszewski told TechCrunch Disrupt 2025 that AI audio models will become commoditized within the next few years, even as the company continues to solve current architecture challenges that give it a short‑term edge. He said ElevenLabs will...
Announcing Ambitious AI Intern Roadmap and Safety Strategy
Yesterday we did a livestream. TL;DR: We have set internal goals of having an automated AI research intern by September of 2026 running on hundreds of thousands of GPUs, and a true automated AI researcher by March of 2028. We may...
Open-Source AI Empowers All, Fuels Faster Societal Impact
The beauty of open-source in AI and robotics is that it gives everyone the power to build tools that solve their real problems, instead of relying on a handful of Silicon Valley founders to decide what matters (too often ending...
Adobe and Google Team up to Offer More AI Models and YouTube Integration
Adobe and Google announced an expanded partnership at Adobe Max 2025, embedding Google’s Gemini, Veo and Imagen AI models into Adobe’s creative suite—including Firefly, Photoshop, Express, Premiere and GenStudio. Enterprise users can fine‑tune these models with their own brand data...
AI Tech Is Real, but Valuations Are a Bubble
Are we in an AI bubble? Is it about to burst? Let’s ask Matt Wolfe (@mreflow). “Some companies deserve their valuations, but most of what we’re seeing right now in AI looks insane.” “I’m seeing startups valued at 50× annual revenue....
Innovating Defense: Ali Manouchehri on AI and National Security
In this episode of AI Chat, host Jaeden Schafer interviews Ali Manouchehri, CEO of MetroStar, to explore how artificial intelligence is reshaping national security and modern warfare. Manouchehri discusses the strategic challenges defense faces today, the potential of AI-driven tools...
Built a Baby Care Tracker in Minutes with AI
Got time for a 10-minute app? We have a new baby in the family. Kiana's four months old (my brother-in-law's new daughter). Since my wife is laid off, she is taking care of the child several times a week and...
The Rise of the AI Accountant: Cheaper. Faster. Riskier?
The UK is entering its first AI‑driven tax season, with a Taxfix survey showing 59% of self‑assessment filers intend to use AI tools to meet the Jan 31 deadline. Respondents cite speed (39%), convenience (36%) and lower cost (33%) as primary...
The Missing Data Link in Enterprise AI: Why Agents Need Streaming Context, Not Just Better Prompts
Confluent unveiled a real‑time context engine that couples Apache Kafka event streaming with Apache Flink stream processing, and released an open‑source Flink Agents framework to give enterprise AI agents continuous, low‑latency data context. The platform creates materialized views from live...
Nvidia Hits Record $5 Trillion Mark as CEO Dismisses AI Bubble Concerns
"I don’t believe we’re in an AI bubble," says Huang after announcing $500B in orders.
Music Playlist Recommendations
The post highlights Rebecca Salganik’s research on fairness in music recommendation systems, outlining group, individual, and counterfactual fairness and the problems of popularity and multi‑interest bias. She presents LARP, a multi‑stage multimodal framework that uses contrastive learning to align text...
Chinese Startups Flood AI Open-Source; US Lags Behind
Why are Chinese companies and startups open-sourcing like crazy whereas American ones don't? Where's the open-source AI from @Tinder @RobinhoodApp @NotionHQ @figma @Airbnb @DoorDash @Snapchat @stripe and the hundreds of successful American startups? If they started doing it, in my...
TikTok Can Use AI to Turn Your Long Video Into Short Ones
TikTok unveiled three AI‑driven creator tools at its US Creator Summit: Smart Split, which automatically chops long videos (over a minute) into multiple short, vertically‑framed clips with captions; AI Outline, an AI‑generated pre‑production assistant that drafts video titles, hashtags, hooks...
Mitigate Gen AI Risks for Responsible Innovation
🚀 𝗡𝗮𝘃𝗶𝗴𝗮𝘁𝗶𝗻𝗴 𝘁𝗵𝗲 𝗥𝗶𝘀𝗸𝘀 𝗼𝗳 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗔𝗜: 𝗔 𝗣𝗮𝘁𝗵 𝘁𝗼 𝗥𝗲𝘀𝗽𝗼𝗻𝘀𝗶𝗯𝗹𝗲 𝗜𝗻𝗻𝗼𝘃𝗮𝘁𝗶𝗼𝗻 Generative AI is transforming industries, redefining creativity, and reshaping how we work. But with great power comes great responsibility. 🌟 In my latest article, I dive deep into the key...
YouTube Will Let You Opt Out of AI Upscaling on Low-Res Videos
YouTube announced that videos uploaded in 240p‑720p will be automatically upscaled to HD using AI, with 4K upscaling planned, but both creators and viewers can opt out and original files will be preserved. The rollout includes a new “super resolution”...
Character AI Is Ending Its Chatbot Experience for Kids
Character.AI announced it will eliminate open‑ended chatbot conversations for users under 18, phasing out access by November 25 with a shrinking daily time limit and robust age‑verification measures. The decision follows lawsuits and two teen suicides linked to prolonged AI...
Grammarly Rebrands to ‘Superhuman,’ Launches a New AI Assistant
Grammarly announced it will rename its corporate entity to Superhuman following its July acquisition of the Superhuman email client, while the Grammarly product name remains unchanged. The company launched Superhuman Go, an AI assistant built into the Grammarly extension that provides...
Phia’s Founders on How AI Is Changing Online Shopping
Phia, an AI startup founded by Stanford roommates Phoebe Gates and Sophia Kianni, launched a browser extension and app that aggregates pricing data from more than 150 second‑hand platforms, covering over 350 million items, to help shoppers compare prices and assess...
Each Computing Leap Redefines How We Express Intent
Every leap in computing begins with a new way to express intent. In the beginning, we spoke in commands. The first computers waited for precision. Each keystroke was an instruction carved into syntax. The interface was obedience itself. We learned to think...
Cameo Sues OpenAI over Sora’s ‘Cameos’
Cameo filed a trademark infringement lawsuit in a California federal court against OpenAI, alleging that the "cameo" feature in OpenAI's Sora 2 video‑generation app dilutes and tarnishes Cameo's brand. The complaint seeks unspecified monetary damages and a court order prohibiting...
Research Skills Grow Without a PhD: Learn, Build, Share
During my conversation with @PetarV_93 - Senior Staff Research Scientist at @GoogleDeepMind and Lecturer at Cambridge. He said something that stuck with me. “You don’t need a PhD to think like a researcher. You need curiosity, discipline, and the courage to keep...
Gemini's $100B Quarter Shows Unprecedented Money‑printing Scale
The greatest money printing engine of all time continues. $100b quarter (!!!) is absolutely nuts. Gemini alone is doing: > 7B tokens per minute > 650M MOAs > 300M paid subscribers https://t.co/v6ECBV9FP2
Remote Labor Index Shows AI Automation Still Untapped
A new eval, Remote Labor Index, measures AI's ability to automate real-world, economically valuable projects from remote work platforms. Currently entirely unsaturated (max score of 2.5%) A great collaboration between @scale_AI and @ai_risks!
Celebrating Hugging Face’s ML & Society Impact
Cool to see all the great work and impact of the @huggingface ML & Society team in the past few years! https://t.co/Q4VKw3Ef9w https://t.co/Ae5Bljp5XC
Companies Often Rebuild AI Evaluation Tools After Buying Faulty Ones
"We're building out our own AI evals tool because the one we bought doesn't work at all." Not the first time I've heard this.
AI Threatens Influencers: Only True Creativity Will Survive
AI is crowding out the humans everywhere except here on X. This has deep implications for all media. “Influencers” ability to influence is about to go away even here in X unless they have deep creativity and bring real value that AI’s...
X’s For You Feed Improves with New AI
X’s For You feed just got noticeably better. They are turning on new AI systems. It will keep getting better. If you aren’t following my lists you are hurting yourself. They provide more signal to make your feed better too.
We Finally Recognized GPUs as AI’s Backbone
In hindsight the importance of GPUs for AI was something we really got right!
AI Designs Stunning Slides From Your Text Instantly
More decks to go, but has designing beautiful slides been a pain point? Upgrades from our Presentation Agent @popaiinone enable prompt-based co-creation! Once you nailed the text content, leave the style design (300+ templates), layout, caption/text edits, chart and graphs...
AI-Powered Vision Reimagines Old Delhi Through Countless Eyes
Watching old delhi with a billion eyes (experimenting w/ generative ai + computer vision) https://t.co/v9eWuxmZS1
Deep Dive Into AI Valuations and Infrastructure Costs
Trying to address all the questions on AI valuations, margins to AI vendors and infrastructure, huge investments in data centers, chips, circular economics, robotics, energy and more with @akshay of @theinformation: https://t.co/0qKruLwEvv
Assessing Whether Large Language Models Truly Grasp Reality
How deeply predictive AI systems understand their subject matter? Can large language models figure out the real world? https://t.co/E2bwsorCLs #AI #MachineLearning #MIT @mitsmr @Hal_Good @AlexMachicado @Analytics_699 https://t.co/5KG5LIrdRd
DeepSeek Dip Buyers Reap Big Gains
to everyone who correctly bought the deepseek dip, enjoy your gains
Google Labs' Pomelli AI Auto‑creates Ad Campaigns From Websites
.@GoogleLabs just launched Pomelli, an experimental AI that turns your brand’s website into full-blown ad campaigns 🤯 Feed it your website: ↳ it reads your tone and style ↳ generates campaigns and captions ↳ keeps content consistent and scalable Live in the US, CA, AU...