Welcome to blocmates. weekly AI newsletter
It’s been a big week for keeping things alive. An Australian entrepreneur saving his cancer-stricken dog with AI; OpenAI’s IPO intent; and a whole universe of AI agents in hashing things out in MiroFish’s emergent world.
Speaking of the future, we’ve been given more prescient insight from SXSW and GTC than Nostradamus could shake a stick at.
Let’s take a look back on what’s pushing ahead:
MiroFish's predictive engine • Anthropic’s judicial backing • OpenAI’s IPO intent • The web measurement apocalypse • Huang’s $1T prediction • The FBI’s shady loophole • ChatGPT’s canine cure • Models & Tools • Financials
P.S. There’s a lot in this newsletter – we suggest getting comfy and tucking in on desktop via the READ ONLINE tab at the top of this mailer.
THE WEEK IN AI
Our weekly news run has had a revamp. Get familiar with our news tags, so you can get straight to the point like the simpletons we are.
..MODEL BEHAVIOR ..
#1 Predict the future with MiroFish
Do you ever feel like you’re living in a simulation? If you do, MiroFish might offer you a glimpse into how it would work. This is just a headline version, but we published a deep dive into the AI prediction engine earlier in the week too good not to reshare.
The story goes Guo Hangjiang, a senior undergrad in China, topped GitHub's Global Trending list in March 2026 with an open-source AI prediction engine that he built in 10 days. It works by feeding it a document – e.g. a news article, policy draft or financial report – and it spawns thousands of autonomous agents, each with a unique personality, memory, and social logic, then drops them into a simulated world and watches what emerges.
He sent a rough demo to Chen Tianqiao, founder of Shanda Group, and one of China's early internet giants. Within 24 hours, Chen had $4.1 million committed. The repo hit 28,600 stars, overtaking OpenAI, Google, and Microsoft on GitHub’s global weekly trending list.
Target applications include financial decision support, policy testing, PR crisis simulation, and marketing strategy evaluation. Public opinion simulations can model how sentiment evolves after a news event, generating 90-day trajectories of things like polarization curves and amplification cycles. Financial forecasting captures psychologies and behaviors traditional models can miss entirely, while you can even inject new variables mid-simulation and watch the digital world reorganize in real time.
Whether MiroFish specifically dominates or competitors emerge, agentic-powered, simulation-driven forecasting may have just flipped the predictive industry entirely on its head. Watch this space.
.FRONTIER NEWSR.
#2 Anthropic gets ‘AI safety case trial’ boost
Nearly 150 retired federal and state judges filed an amicus brief this week, supporting Anthropic in its lawsuit against the Trump administration. The rolecall of judicial support joins an even longer list of heavyweight supporters including major tech industry groups, former senior national security officials, Microsoft, and rival AI companies. Anthropic contends that the Pentagon's supply-chain risk designation punishes the company for protected speech about AI policy, and that current procurement laws do not give the Pentagon or the President the power to blacklist a company on those grounds. A hearing on whether to grant Anthropic a temporary relief is set for March 24. If any favorable ruling actually matters, however, remains just as big a question, given the Trump administration usually does whatever the fuck it likes.
..FRONTIER NEWS ...
#3 OpenAI’s IPO intent confirms enterprise pivot
This week's OpenAI all-hands crystallized what the IPO story is really about. Fidji Simo, OpenAI's CEO of Applications, told staff the company is "orienting aggressively" towards high-productivity use cases and that the goal is to transform ChatGPT into a productivity tool – with the IPO potentially landing as soon as Q4 2026. Forget the advertising, Altman knows enterprise is where it's at and is intending to close the gap on competitors, Anthropic. Approximately $10 billion of OpenAI’s $25 billion annualized revenue comes enterprise, while Anthropic's annualized revenue has surged past $19 billion – with roughly 80% from enterprise clients – Dario Amodei confirming $6 billion was added in February alone, driven almost entirely by Claude Code.
“Yeah, your marketing strategy is screwed.” Matthew Prince at SXSW.
..MODEL BEHAVIOR...
#4 Bots will outnumber humans online by 2027
Cloudflare CEO, Matthew Prince, dropped a humdinger of an insight at SXSW last week, saying AI bot traffic will exceed human traffic on the internet by 2027. The maths is pretty straightforward: agentic AI will blow up a single user search query by the nth degree as they scuttle across the internet simultaneously looking to fulfil that request. Before generative AI, bots accounted for around 20% of web traffic. Naturally, this will render traditional traffic measurements meaningless, up-end ad and marketing models, blow apart infrastructure and hosting costs, and totally redefine the consumer web experience as we know it. No biggy then.
..FRONITER NEWS...
#5 Huang hangs $1T forecast through 2027
At GTC this week, Jensen Huang doubled his own demand forecast – from $500 billion to $1 trillion in orders for Blackwell and Vera Rubin chips through 2027. Nvidia’s Miracle Man went further, telling analysts the total across all product lines will exceed that figure once new CPUs, Groq inference chips, and storage racks are factored in. "I am certain computing demand will be much higher than that," Huang said. The chip developer’s pivot from training to agentic inference is the engine behind the upgrade, while Nvidia looks to position itself as the full infrastructure stack, not just the GPU supplier. Company earnings have given Huang’s claims weight: last month, Nvidia posted a 65% in annual revenue at $215.9 billion for FY 2026 – their highest annual result ever.
..FRONTIER NEWS ..
#6 The FBI is buying your location data
This is mildly terrifying: During a Senate Intelligence Committee hearing on Wednesday, the FBI’s Bro-in-Chief Kash Patel confirmed the bureau is buying our location data – and no warrant is required. Till now, the practice has been a relatively blunt instrument when used by humans, but with AI able to cross-reference things like movement patterns, identify relationships, and build behavioural profiles at scale and speed, the same purchase fundamentally becomes a much more powerful surveillance tool. The admission reverses what his predecessor Christopher Wray told Congress in 2023, in what was called "an outrageous end run around the Fourth Amendment."
..MODEL BEHAVIOR ..
#7 Google Stitch causes Figma stock dive
Figma's stock took a long walk off a short plank this week, diving roughly 12% after Google launched an updated version vibe-design tool, Stitch. Released in beta on Tuesday, Stitch has similar AI-powered features to Figma – prompt generated design, agent critiques and voice controls – but Google is offering it for free. The updated tool converts natural language descriptions into high-fidelity UI designs without requiring traditional wireframes, and includes interactive prototyping capabilities. Figma's stock is now down around 35% year-to-date, and is becoming one of the unwanted poster childs of the software sector's AI-inflicted slide.
..MODEL BEHAVIOR ..
#8 Man cures dog’s cancer with ChatGPT (kinda)
The feel-good story we needed this week. When standard surgery and chemo failed his dog Rosie, Sydney entrepreneur Paul Conyngham used ChatGPT to map a plan and Google DeepMind's AlphaFold to model the protein driving her mast cell cancer and identify a potential drug target. Working with UNSW researchers, he developed a personalized mRNA vaccine that reportedly reduced the tumour by 75%. AI luminaries shared the good news, though researchers quickly pushed back, noting the treatment was AI-assisted rather than a cure, with conventional immunotherapy playing a role too.
TOOLS OF THE WEEK
The latest tool, model and LLM news, updates, tips and tricks – you name it, we’ll play with it til it breaks.
OpenAI GPT-5.4 Mini & Nano: Stars of the week. They bring GPT-5.4's intelligence to a much smaller, faster architecture. Developers are flocking to them because they are 2x faster than the previous "mini" and support a 400k context window.
Anthropic Claude 4.6: Claude 4.6's 1M token context window is now generally available as of this week. It also introduced "Context Compaction," which prevents the model from getting "confused" or "forgetful" (context rot) during extremely long conversations.
Google Gemini 3.1 Pro & Chrome Integration: Google expanded Gemini 3.1 Pro this week with a new "Auto Browse" feature in Chrome. It can now autonomously handle web tasks (like booking travel or researching competitor pricing) directly in your browser side panel.
Cursor Composer 2: Cursor’s highly optimized in-house coding model that significantly reduces latency and API costs. It features "Predictive Refactoring," allowing it to anticipate changes across multiple files and prepare code structures in the background before you even finish typing.
Microsoft MAI-Image-2: The upgraded text-to-image model has set a new standard for typographic accuracy and spatial reasoning. It is now deeply integrated into the Windows 12 "Canvas," enabling users to generate complex layouts with perfect text rendering for professional design assets directly on the desktop.
Google AI Studio (Antigravity Agent): Google transformed AI Studio into a full-stack "vibe coding" platform by introducing the Antigravity agent. Leveraging Gemini 3.1 Ultra’s massive context window, it can now build, connect to a Firebase backend (including Auth and Firestore), and launch a live, production-ready web app from a single prompt in seconds.
FINANCIALS
Covering significant funding stories, deals and partnerships, and investment opportunities globally.
AMI Labs has launched with a $1.03 billion seed round, the largest in European history. Based on Yann LeCun’s JEPA architecture, the Paris-based startup is positioned as the primary rival to LLM-centric paradigms. The round establishes a massive war chest for developing "world models" designed to achieve human-level reasoning in physical and digital environments.
Earendil Labs has raised $787 million in a massive financing round to scale its AI-driven biologics platform. Led by Dimension Capital and DST Global, with participation from Sanofi and Pfizer’s Biotech Development Fund, the capital will accelerate a pipeline of over 40 AI-generated therapeutic programs. The deal highlights the shift toward AI as a "production engine" for real-world medicine.
Nexthop AI has closed a $500 million Series B to build AI-optimized networking infrastructure. Led by Lightspeed Venture Partners with participation from A16Z, the deal signals that the "connective tissue" between GPU clusters has become a standalone investment category. The funding will be used to resolve the networking bottlenecks currently limiting the scale of trillion-parameter model training.
PQuince has secured $500 million in new fundin, marking it at a $10.1 billion post-money valuation. The round establishes Quince as one of the first AI-native consumer "unicorns" to reach mega-scale. The investment will be used to expand its AI-driven supply chain and demand prediction models, which currently bypass traditional retail layers to offer luxury-quality goods at radical price points.
RoboForce has raised $52 million in a round led by YZi Labs, bringing its total funding to $67 million. Featuring high-profile backers like Jerry Yang and Myron Scholes, the round is dedicated to scaling "Physical AI" foundation models. The capital will drive the manufacturing of general-purpose industrial robots integrated with NVIDIA’s Jetson Thor edge processors for deployment in heavy labor sectors.
UNTIL NEXT TIME, FRIENDS.
We end the note with an extremely valid crash-out, and a big mood for the following week. 🤖
Until the next one 🫡
