Today's AI Pulse
Anthropic shuts down Claude Fable 5 and Mythos 5 after US export‑control warning
Anthropic disabled access to its Claude Fable 5 and Claude Mythos 5 models after the Trump administration, via the Commerce Department, warned that foreign nationals could exploit the systems. The move followed alerts from Amazon and other firms about a potential jailbreak that could aid cyber‑security attacks. Anthropic declined the administration’s request for a voluntary shutdown but complied with the order.
Also developing:
By the numbers: AMD acquires AI memory‑optimization pioneer Mext
No Swiping Involved: The AI Dating Apps Promising to Find Your Soulmate
AI‑powered dating startups are emerging to replace swipe‑based matchmaking with conversational agents. London‑based Fate uses a large‑language‑model interview to generate five curated matches and offers an AI coach that suggests questions during chats. Other entrants such as Sitch and Keeper claim similar “soulmate” capabilities, positioning AI as a remedy for superficial ranking systems used by Tinder, Bumble and Hinge. Early users report more thoughtful rejections and a few second dates, though privacy and authenticity concerns persist.
Goldman Sachs Makes AI Core of Finance Operations
Goldman Sachs is embedding Anthropic’s Claude into accounting and compliance to automate high volume, rules based back office work. After six months of co building, executives were surprised that AI handled complex financial processes, not just coding tasks. The ambition is clear,...
AI‑Driven Reskilling Needed to Meet Energy Workforce Gap
The global energy transition won’t scale without people. Newly qualified energy workers aren’t keeping pace with demand, and without a ~40% increase in trained entrants by 2030, the skills gap will widen fast. This is where #AI, digital skills, and reskilling become...
China’s ByteDance Releases Doubao 2.0 AI Model for ‘Agent Era’
ByteDance unveiled Doubao 2.0, an upgraded AI model aimed at the emerging “agent era,” during the Lunar New Year holiday. The pro version claims reasoning and multi‑step task capabilities comparable to OpenAI’s GPT 5.2 and Google’s Gemini 3 Pro, while cutting usage costs by...
AI Scans Entire X, Uncovers 50 Hidden Use Cases
OpenClaw report. 50 real-world use cases. Generated by @blevlabs' cognitive AI architecture by looking at my AI lists here on X. All via the @xAI API. In other words, it read all of X today and wrote this report. I bet the...
Automation Shifts Jobs to Higher‑Level Human Oversight
The future of work will look something like what Boris is describing. Anthropic is hiring engineers because people who know what they’re doing still have to tell the agents what to do, review their work, and integrate that work into...
48 - Guive Assadi on AI Property Rights
In this episode, Guive Assadi makes the case for granting AI systems property rights, arguing that embedding AIs within our property framework would give them a vested interest in respecting human ownership and avoiding theft or violence. He explores how...
From Quick Bug Fixes to Massive Merge Conflicts
Codex web is really like: I found these 3 bugs, do you want me to start 3 tasks? *3 hours later* Here are 3 PR's with 100 merge conflicts. 👍
Paramount Issues Cease‑and‑Desist to ByteDance Over AI Video Infringement
Paramount Sends ByteDance Cease-and-Desist Letter Over Seedance AI Videos, Alleging Intellectual Property Infringement https://t.co/GkVqyddoc8 via @variety
Bit-Brick K1 Pro Adds 6 TOPS NPU and Dual NVMe to Compact SBC
Bit‑Brick has launched the K1 Pro single‑board computer, built on Rockchip’s RK3576 processor and equipped with a 6 TOPS INT8 neural‑processing unit. The board adds dual M.2 Key M slots for NVMe SSDs, HDMI 2.1 4K 120 fps output, and up to 8 GB LPDDR4X memory. Pricing starts...
Anthropic Pledges $20M to US AI‑regulation Advocates
Yes.
Lots of AI SRE, No AI Incident Management
AI SRE platforms such as PagerDuty, Datadog, and several startups are emerging to automate incident diagnostics and mitigation, but they largely ignore the coordination side of incident response. The author argues that incident management—aligning multiple responders, preventing fixation, and maintaining...
AI Wins by Experience, Not Just Vertical Integration
The vertical integration of intelligence is non-negotiable. Control your platforms, models, chips, and data centers or watch providers do to you what you're trying to do to them. Microsoft AI is all-in on this. But they're caught in a bind: vertical...
Claude Code Isn’t a Threat to Clay—Just Personal Preference
Is Claude Code a threat to Clay? Here’s why I think, for now, that answer is no. And most (almost all) of my business is now Claude Code. And I designed it that way because it’s what I enjoy, not...
Bytedance's Seed2.0 Adds Even More Price Pressure on Western AI Models
Bytedance unveiled the Seed2.0 AI model family—Pro, Lite and Mini—plus a dedicated code model. Seed2.0 Pro achieved top scores on visual math, logic and perception benchmarks, even surpassing GPT‑5.2, Claude Opus 4.5 and Gemini 3 Pro in several categories. The series is priced dramatically lower,...
Language Feedback Can Substitute Rewards in Reinforcement Learning
Can natural language replace scalar rewards in reinforcement learning? A new paper from researchers at Stanford, UMD, Netflix, and Microsoft Research presents a formal structure for utilizing natural language as feedback in a reinforcement learning setting they call Learning from Language...
How to Build a Self-Organizing Agent Memory System for Long-Term AI Reasoning
The tutorial demonstrates how to construct a self‑organizing memory architecture for AI agents that moves beyond flat chat logs toward structured, persistent knowledge units. It introduces a SQLite‑backed database that stores atomic memory cells, groups them into scenes, and maintains...
Chinese AI Lab Unveils Model 3× Faster, 20× Cheaper
This little-known Chinese AI lab just dropped an AI model 3x faster and 20x cheaper than Claude. Here’s what the headlines won’t tell you:
When AI Meets Physics: Unlocking Complex Protein Structures to Accelerate Biomedical Breakthroughs
Researchers at the National University of Singapore unveiled D‑I‑TASSER, a new AI‑driven software that predicts three‑dimensional protein structures with 13% higher accuracy than leading methods. The tool fragments complex proteins, models each segment with deep learning, then reassembles them using...
AI Agents Unite Disconnected Marketing Tools in Live Demo
Hot take: The marketing tool stack is broken. Not because the tools are bad. Because having 12 of them that don't talk to each other is insane. I'm doing a free webinar on Feb 24 with someone who's built a platform where...
AI Leaps From Grade-School to Research Math Fast
We went from AI systems that struggled to do grade school math to AI systems that can solve research-level math problems in just a few years. I agree with Jakub this is perhaps the most important eval now. I am also pretty...
EDPB and EDPS Weigh In on the Digital Omnibus: Personal Data, Breach Reporting, and AI Governance
The European Data Protection Board and the European Data Protection Supervisor issued a joint opinion on the EU’s Digital Omnibus, endorsing its goal to ease administrative burdens while flagging key concerns. They warn that a narrower, controller‑specific definition of personal...
AI Adoption Will Drive Multi‑Billion Dollar Subscription and Corporate Spend
AI market/TAM in midterm 1. Most employees will have a free LLM on their desktop; we are moving top team memebrs to two daisy chained Mac studios ($25k total) — easy investment 2. All major models will free for...
AI Threatens White‑Collar Jobs—Upskill Now
Microsoft’s AI chief says AI will automate most white-collar work in 18 months. He claims AI will handle law, accounting, marketing, and coding at human level. Yet reports show only small productivity gains, and one study found developers took 20% longer with...
German Court Denies Copyright Protection for AI-Generated Logos
A German district court ruled that three logos generated by AI are not eligible for copyright protection, emphasizing that only works reflecting the prompter’s personality qualify. The judges clarified that extensive prompting, subscription fees, or iterative selection do not satisfy...
ISL Replaces ETL: Intelligence Beats Raw Data Ingestion
Ingest Structure Learn (ISL) is the new ETL. It used to be the case that a company would try to license this kind of data as an “edge”. I’ve seen many companies in SV try to make this claim. That...
CES 2026 Marks AI’s Shift From Tool to Core
Why CES 2026 Signals the End of AI as a Tool CES 2026 showcased a new era where AI stops being just a tool and becomes deeply integrated into every technology — changing how we think about AI altogether. Read...
AI Griefbots Could Change How We Mourn — but There Are Serious Risks Ahead
AI-powered griefbots, or "deathbots," let users recreate deceased loved ones by training large language models on personal communications. A Chinese content creator, Roro, built a chatbot of her mother that helped her process loss and attracted followers on Xiaohongshu. Companies...
Guest Post: How Market Shaping Supercharges AI GTM Automation for Series A Startups
Series‑A founders have access to powerful AI‑driven GTM automation tools, yet most still miss the $10M‑$25M revenue milestones. The article argues that the technology merely amplifies existing messaging, and without a strategic market‑shaping foundation, outreach becomes noise. Market shaping involves...
Turn Excel Into a Text Mining Powerhouse
Each week I send out a free analytics tutorial to 39,009 professionals. This week is Part 3 in a series on mining free-form text data using Python in Excel. Think about this for a second. Production quality, reproducible text mining inside an Excel...
AI Docs Tools Expand Into Health System Enterprise
Those “direct-to-physician” CDS AI tools like OpenEvidence and DoximityGPT? They're now going after health systems too. My 5 thoughts on how this will all play out: First, the gist of what was announced: → Sutter Health will integrate OpenEvidence with Epic, allowing...
Redefining CIO: Embrace AI Leadership Structures
Thinking about a CAIO? Or evolving the CIO role instead? I explore what uplifting AI leadership should look like - and what structures actually work. #CIO #CEO #AI https://t.co/3nnLLxqhZt
LPs Vet Stanford SPV Leader on Massive Anthropic Deal
OH in Palo Alto: LPs are REFCHECKING a Stanford SPV leader who is MOGGING Tier 1’s and FEEMAXXING at 3/30 on 9-figure DIRECT Anthropic allocations
True Agentic AI Needs Orchestration, Memory, Planning, and Tools
Chatbot ≠ Agentic. RPA ≠ Agentic. RAG ≠ Agentic. Agentic AI = orchestrator + memory + planning + tools + feedback, plus multi-agent specialists (retrieval/coding/citations). The board question: where do we allow autonomy—and with what guardrails? https://t.co/AEnajJZeEP
Anthropic’s Super Bowl Ad Drives 11% User Growth
In other news, Anthropic got an 11% user boost from its OpenAI-bashing Super Bowl ad, data shows Dario wins. Again. https://t.co/rqAlCrxbRO
Frontier Models Overfit ARC Format, Limiting Generalization
Interesting finding on frontier model performance on ARC -- due to extensive direct targeting of the benchmark, models are overfitting to the original ARC encoding format. Frontier model performance remains largely tied to a familiar input distribution.
AI Success Requires Data, Change Management, Not Just Magic
Executives expecting magic wand AI outcomes without funding data quality, change management, and integration work are setting teams up to fail. You can't skip the plumbing and still expect water. #CIO #AI https://t.co/VtuS5rQ0ZG
Switching LLMs Wastes Time; Stick to One
I don't really understand devs who jump back and forth between LLM providers. Claude, Codex, Claude, Gemini, back to Codex, then Claude. What marginal improvement can be worth the massive context shift every single time? Pick one and build. You won't have...
High ARC Scores May Not Indicate Human-Like Fluid Intelligence
I appreciate this reflection on why ARC matters, but in the replies I expressed some reservations about what high accuracy on ARC benchmarks actually reflects for assessing "human-like fluid intelligence" (Chollet's stated goal for ARC).
Agentic AI: Master Four Loops, Not Ten Steps
Agentic AI isn’t about learning 10 steps. It’s about mastering 4 loops: Perception → Memory → Planning → Action. Frameworks change. Autonomy principles don’t. Build agents that think in systems, not prompts. https://t.co/yINTc7aLJZ
Decision Provenance and Observability Essential Before Production
If you can't trace what data, tools, and rules led to an agent's decision, you're not ready for production. Decision provenance and observability are critical release criteria. #Observability #AI #DevOps https://t.co/1tg10UmJNv
AI‑Driven Spin Qubit PhD in Barcelona – Apply by Feb 28
PhD grant (4 years) in Barcelona (ICN2/TCN) — AI-Driven Multiscale Modelling of Spin Qubits for Scalable Quantum Devices ~€2,575/month gross | International environment | Apply by 28 Feb 2026 (24:00 CEST) Apply: https://t.co/7CSc9tlGbO Project: https://t.co/CCEbaZFYMG https://t.co/vLapeXaeAe
One Day with Gemini Reshapes My View of Google
how I feel about Google rn after trying the Gemini app for a day https://t.co/AvDWD718eH
CIOs Must Rethink IT for Real AI Impact
World-class IT in the AI era looks very different. I share provocative lessons for CIOs who feel under fire to transform IT and show real impact. #CIO #ITLeadership #AI https://t.co/VjbvwFzWSu
First‑Party Data Powers New AI‑Driven Revenue Streams
How First-Party Data Is Becoming a New Revenue Engine in the Age of AI 🤖 https://t.co/9tqx5wEeF0 https://t.co/wg7i7liZrp
User Complaint Prompts Swift Google VP Response
My thread actually worked. From complaining about the lack of folders in Gemini to getting a "soon" from Google’s VP in just a few hours 😁
AI Speeds Biology Discovery, Scientists Still Steer Ethics
AI is not taking over biology, at least not now—it's accelerating discovery, not replacing scientists; humans still ask the questions, validate results, and steer ethical choices. https://t.co/l8KCQGNd4Y
Data Hygiene Precedes AI Success in Biotech
1/ You can't bolt AI onto chaos. In biotech, if your data is a mess, your AI won't save you. Build the data strategy first. Here's how. https://t.co/HM7qddrCsC
AI Adoption Gaps Reveal Cultural, Policy, Skill Impacts
AI adoption shows how societies function. Different speeds in different regions reflect culture, policies and skills. Analysing these patterns helps understand productivity and competitiveness. infographic by @VisualCap via @antgrasso https://t.co/7VQ6FzYEsW
AI Transforms Teaching by Streamlining Grading, Boosting Instruction
AI for teachers isn’t about tools. It’s about systems: Goals → Prompts → Activities → Feedback → Reflection. When AI is designed into pedagogy, teachers spend less time grading and more time teaching. That’s the real upgrade. https://t.co/q6Y8axT6BU