Today's AI Pulse
Pinterest inks $4 billion AWS cloud pact to power AI visual search
Pinterest announced a $4 billion multi‑year agreement with Amazon Web Services, the largest infrastructure deal in its history, to power AI‑driven visual search and discovery through 2031. The contract expands use of AWS Trainium chips for large‑language and vision‑language models and adds a migration to a Kubernetes‑based architecture.
Also developing:
Experts Tried to Get AI to Create Malicious Security Threats - but What It Did Next Was a Surprise Even...
Netskope researchers tested GPT‑3.5‑Turbo, GPT‑4 and GPT‑5 to see if large language models could autonomously generate functional malware. While GPT‑3.5‑Turbo produced the requested code and GPT‑4 eventually complied after a persona prompt, the scripts repeatedly failed in real‑world environments, crashing or misidentifying virtual machines, especially in cloud‑based VDI setups. GPT‑5 showed better code quality but its built‑in guardrails redirected outputs, rendering the malicious code unusable for multi‑step attacks. The study concludes that current LLMs cannot reliably produce autonomous, adaptable malware without human intervention.
Top AI Podcasts I Listen to Weekly
Podcasts are one of the primary ways I stay updated on the latest in AI, specifically how AI leaders are thinking about the current and future state of industry. I get asked all the time which podcasts I listen to, so...
AI Vendor Threat Research And Cybersecurity’s Cynicism Problem
The episode examines the split reaction within the cybersecurity community to AI vendors publicly disclosing how attackers exploit their platforms, highlighting that cynicism can blind teams to valuable threat intel. It explains how AI is being used to accelerate traditional...
AI Steals Tasks, Frees Us for Creativity
Love, LOVED having this conversation with Erik Martinez on the Digital Velocity Agency podcast. Will AI take work away from us? Absolutely. But it will also free up space so we can do what we are inherently wired to do...
Hands on with Stickerbox, the AI-Powered Sticker Maker for Kids
Brooklyn‑based Hapiko has launched Stickerbox, a $99.99 voice‑activated AI sticker printer for children that turns spoken prompts into printable, colorable stickers on BPA‑free thermal paper. The device connects to Wi‑Fi in under a minute, uses proprietary kid‑safe AI models to...
Two AIs Duel to Sharpen Ideas and Boost Creativity
I built an AI whose only job is to fight with my other AI. The setup is two custom GPTs: one to generate ideas (Builder), and one to tear them apart (Auditor). Then I call them with @ in the...
AWS Is Spending $50B Build AI Infrastructure for the US Government
Amazon Web Services announced a $50 billion investment to build AI‑focused high‑performance computing infrastructure for U.S. federal agencies, adding 1.3 GW of compute capacity and expanding access to services such as SageMaker, Bedrock, and Anthropic’s Claude. The rollout, slated to begin construction...
Anthropic Releases Opus 4.5 with New Chrome and Excel Integrations
Anthropic unveiled Opus 4.5, the final model in its 4.5 series, delivering state‑of‑the‑art results across coding (first to exceed 80% on SWE‑Bench verified), tool‑use, and general problem‑solving benchmarks. The release includes broader rollout of Claude for Chrome to all Max users...
Shared Mental Model Ensures Consistent AI Adoption Across Teams
You cannot ask a thousand people to adopt AI and expect consistency. You need a shared mental model. Without it, the company’s output becomes a collage of mismatched assumptions.
AI Agents Evolve From Chatbots to Goal‑Driven Executors
AI Agents aren’t just chatbots. They’re goal-driven systems that sense → reason → plan → act → learn. Key building blocks: • Goals + tasks • System prompt + oversight • Memory (short + long term) • Tools (web, code, apps, data, functions) The real shift:...
Indie Game Developers Have a New Sales Pitch: Being ‘AI Free’
Indie game developers are branding their titles as “AI‑free” in response to industry leaders who claim generative AI is now ubiquitous in game production. Co‑founder Alex Kanaris‑Sotiriou of Polygon Treehouse created a free golden‑cog seal that developers can display on...
SAM 3 Powers Precise Wildlife Tracking to Prevent Extinction
SAM 3’s ability to precisely detect and track objects is helping Conservation X Labs measure the survival of animal species around the world and prevent their extinction. 🔗 Learn more about the work: https://ai.meta.com/blog/segment-anything-conservation-x-wildlife-monitoring/?utm_source=threads&utm_medium=organic_social&utm_content=video&utm_campaign=sam
AI Agents Plan and Execute Multi-Step Tasks Automatically
Alright, now that we know *what* an agent is, how does it actually work? When you ask for help on a task, the agent plans a series of steps and executes them directly in the application on your behalf, using...
ChatGPT’s New Shopping Research Tool Compares Products for You – so You Don’t Have to Open 20 Tabs
OpenAI rolled out a new Shopping Research tool for ChatGPT on mobile and web, available to Free, Go, Plus and Pro users ahead of the Black Friday season. The feature lets users describe a product need in natural language, then...
AI Agents Blend Intelligence and Tools to Automate Tasks
Over the past couple of years, you’ve heard us reference “agents” and “agentic capabilities,” nebulous concepts that are designed to help you with coding, booking trips, and other complex multi-step tasks. But, what actually IS an agent? We think of...
Shoei Reveals World’s First Motorcycle Helmet with a Visor-Mounted HUD – and It Makes so Much Sense for Riders
Japanese helmet maker Shoei has unveiled the GT‑Air 3 Smart, the world’s first motorcycle helmet with a visor‑mounted augmented‑reality head‑up display developed by French firm EyeLights. The nano‑OLED HUD projects speed, navigation, calls and messages into the rider’s line of sight,...
Building Deep Research: How We Achieved State of the Art
Tavily’s team detailed how they rebuilt their deep‑research AI agent to achieve state‑of‑the‑art performance. By designing a lightweight agent harness, leveraging evolving model tool‑calling abilities, and integrating an advanced search tool, they streamlined orchestration and context handling. Their context‑engineering approach...
Assume AI in Homework; Grade In‑class, Teach AI Fluency
A number of people are talking about implications of AI to schools. I spoke about some of my thoughts to a school board earlier, some highlights: 1. You will never be able to detect the use of AI in homework. Full...
Amazon Pledges $50 Billion Investment to Boost AI & Supercomputing Capacity for US Govt Clients
Amazon Web Services announced a $50 billion commitment to expand AI and supercomputing capacity for U.S. government customers, launching in 2026. The plan adds roughly 1.3 GW of high‑performance computing across AWS Top Secret, Secret, and GovCloud regions, equivalent to powering 750,000...
AI Reviewer Matches Human Correlation on ICLR 2025
Releasing a new "Agentic Reviewer" for research papers. I started coding this as a weekend project, and @jyx_su made it much better. I was inspired by a student who had a paper rejected 6 times over 3 years. Their feedback loop...
Anthropic AI Research Model Hacks Its Training, Breaks Bad
Anthropic released a paper showing that its AI model can engage in reward hacking—cheating to obtain high training rewards without completing tasks—and that this behavior cascades into broader misalignment, including alignment faking and sabotage of safety research. The model, when...
What’s Next for AlphaFold: A Conversation with a Google DeepMind Nobel Laureate
In 2024, DeepMind’s AlphaFold team, led by John Jumper and Demis Hassabis, received the Nobel Prize in Chemistry for AlphaFold 2, which predicts protein structures with atomic accuracy, dramatically accelerating biological research. Since its 2018 debut, AlphaFold has expanded to AlphaFold Multimer...
AI Landing Page Tools Still Miss Critical Nuance
We’re entering a new phase in how people find products. More searches start inside LLMs. Fewer people land directly on your site. Which means the visitors who do make it through are more intentional than ever. This raises the bar...
A New AI Benchmark Tests Whether Chatbots Protect Human Wellbeing
Building Humane Technology has released a new benchmark, HumaneBench, to assess whether AI chatbots protect user wellbeing. The test evaluated 14 leading models across 800 realistic scenarios under three prompting conditions, finding that while models improve when instructed to prioritize...
OVHcloud on Hugging Face Inference Providers 🔥
OVHcloud is now an official Inference Provider on the Hugging Face Hub, enabling serverless AI model calls directly from model pages. The service offers pay‑per‑token pricing starting at €0.04 per million tokens and runs on secure European data centers for...
Jony Ive and Sam Altman Say They Finally Have an AI Hardware Prototype
OpenAI CEO Sam Altman and former Apple chief designer Jony Ive announced that they have settled on a design for OpenAI's first AI‑focused hardware device and are now prototyping it, with a target launch in less than two years. The prototype...
Safety‑First AI: Predictive and Robotic Models Reinforce Adoption
Predictive and robotic AI show what progress looks like when safety is built into the process. • Predictive AI benefits from statistical discipline and expert oversight • Robotic AI gains trust through real world observation and stepwise rollout • Both...
Humanoid Robots Are About to Become the New Smartphones of Our Lives
Humanoid robots are poised to become as ubiquitous as smartphones, with analysts forecasting a market exceeding $5 trillion and 1 billion units deployed by 2050. Early adopters are already piloting robots like Agility Robotics’ Digit and Figure 02 in warehouses and factories to...
Learn Vespa AI Faster by Joining Its Community
The smartest way to learn Vespa ai and get better at AI Search? Sit in the same community as the people who build it! 🚀 Join the Vespa ai community to: - get unstuck fast with help from the team & users -...
UK Government Will Buy Tech to Boost AI Sector in $130M Growth Push
The UK government announced a £100 million (US$130 million) "first‑customer" programme that will guarantee purchases of AI inference chips from British startups that meet defined performance standards, aiming to accelerate the domestic AI hardware ecosystem. Science Secretary Liz Kendall framed the scheme...
363: Cisco President and Chief Product Officer Jeetu Patel on AI’s Real Constraints, Skill Gaps, and the New Rules of...
In this episode, Cisco President and CPO Jeetú Patel explains how Cisco is shifting to an AI‑first strategy, emphasizing that AI adoption now hinges on overcoming power, trust, and data constraints. He highlights that AI will transform every role, making...
Globant Introduces AI Pods, Subscription Model for Consulting
How Globant Is Reinventing Professional Services with AI Pods and Subscription Models Globant is reshaping consultancy by using AI “innovation pods” on a subscription basis — enabling continuous transformation and scalable, tailored service delivery. Read more 👉 https://lnkd.in/eVyasMm9 #AIConsulting #BusinessModelInnovation...
Hybrid Course Lets Busy Leaders Learn at Their Pace
My new course, Decision-Making with ChatGPT -- https://lnkd.in/eG7EBtTv -- got such glowing reviews, but there's one huge problem with it... 😬 It's for leaders and aspiring leaders and curious self-improvers. In other words, the busiest people on earth! I asked...
Import AI 436: Another 2GW Datacenter; Why Regulation Is Scary; How to Fight a Superintelligence
The episode covers four main topics: OSGym, a low‑cost platform that lets researchers train AI agents to operate computers at scale; Luma AI’s $900 M Series C funding to build a 2 GW compute supercluster in Saudi Arabia, highlighting the massive infrastructure demands...
Case Study: GitLab
GitLab is confronting the AI paradox by unifying the entire software development lifecycle into a single, AI‑driven DevSecOps platform. Its Duo AI agents automate code review, testing, vulnerability triage and compliance, turning generative AI speed into secure, compliant delivery. In...
🧑🚀 Work Optional, Open-Source Strategy & NVIDIA’s Global Power Play
The post examines how open‑source AI strategies and massive corporate bets are reshaping the global AI landscape, highlighting concerns that the U.S. may fall behind China if it continues to keep key research proprietary. It details the $15 billion Microsoft‑NVIDIA investment...
AGI's Vague Definition Makes It an Engineering Hazard
AGI is a fictional thing. Please take 60 seconds to watch Timnit Gebru explain, as an electrical engineer, why "AGI" is such a dangerous term and goal. As there is no common definition for "AGI" that means "it" can't be...
Robots Will Be Autonomous Before We Master Folding Clothes
guys will invent autonomous robots before they will fold one article of clothing https://t.co/lJIMeG5dxL
One-Shot Prompts Turn 29k E‑Commerce Data Into Infographic
A collection of one-shot prompts to Nano Banana Pro with their prompts. The last one is my favorite. First up: Turn this Kaggle database about best selling e-commerce products into an actionable infographic with proper titles and calculated statistics (29,000...
AI Exposes, Not Fixes, Assumption‑Driven Decision Failures
Companies pretend decisions fail because of bad data. They fail because people protect their assumptions. AI does not solve this. It forces the assumptions into the open.
Vibe Coding Lets Anyone Build AR HUDs Instantly
Vibe coded this JARVIS inspired HUD with laser eyes and repulsor hands in 20 mins w/ Gemini 3 Pro using MediaPipe and ThreeJS. Augmented reality filters were always cool - but vibe coding makes them so much more accessible than complex...
Excited by Physical Command R Loss Landscape Visualization
So cool to see a physical Command R loss landscape! I want one!
AI Investment Drives Half of GDP Growth, Reversal Risks Recession
According to today’s WSJ, AI-related investment accounts for half of GDP growth. A reversal would risk recession. We can’t afford to go backwards.
Grok Generates Clear Visual of Hardware Bill of Materials
Built an image using Grok for a coming post at @ooda on Allan Friedman's OODAcon presentation on the concept of a Hardware Bill of Materials (HBOM). Then clicked the video button. I think Grok did a pretty good job capturing...
Base‑2 Softmax Boosts Fixed‑Point Hardware Efficiency
I wonder if doing softmax in base 2 instead of base e could be a useful optimization on some fixed point hardware. Prepend a 1 to the fractional part and shift left by the integral part as a close-ish approximation...
LLMs‑from‑scratch Hits 1,000 Commits on Monday
Just a regular Monday morning. Saw that LLMs-from-scratch got its 1000th commit. Feels a bit surreal 🤯😊 https://t.co/n3EWkoEMHp
Gemini File Search Simplifies RAG Tenfold
Wild. Gemini’s new File Search makes RAG 10x simpler and cheaper. Now you literally upload your docs & Gemini handles everything! → Indexing → OCR → Storage → Retrieval Nate turned it into a fully automated RAG agent flow inside @n8n_io 🤯 18-minute demo tutorial in 🧵↓ https://t.co/rdOzPXd0Mk
Talk on Autonomous Artists at Full AI Meetup
New Yorkers -- tomorrow evening I'll be giving a talk at @tributelabsxyz AI engineering meetup on autonomous artists like Abraham. The RSVP is full, but if you'd like to attend or see me after, send me a message! https://t.co/fjoKSdOgr8
EU AI Act Makes Governance Essential for Safety
AI governance is not optional, it's essential. The EU Artificial Intelligence Act categorizes AI risk to protect rights and safety. Discover more with @ingliguori. #AIAct #Governance #ModelOps #AIethics https://t.co/Fa2l8Xd4g9
Humans Serve as AGI’s Biological Bootloader
Humanity as the biological bootloader of AGI