Welcome to the blocmates AI newsletter.
What a difference a week makes. Well, not in this week’s case really, it’s been much of the same: the Anthropic-Pentagon saga is still playing out; model iterations keep looking to push entire industries into obsolescence, and world leaders remain insistent on living out their tin-pot Bond-villain fantasies.
Oh, and WW3 almost broke out.
But even if the bombs do fall, we’ll be here to offer our own inimitable (read: dumb) perspectives on AI and frontier tech, and possibly even share a few cool items you might’ve missed. This week we’re covering…
OpenAI’s fumble is Anthropic’s gain • Palantir’s shadow politics • Yes, GPT-5.2 was cringe • GitHub could have competition • Is Dorsey AI washing? • Ben Affleck’s stealth AI studio • Claude’s conjecture conundrum
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 the 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.
..FRONTIER NEWS ..
#1 OpenAI’s scores own goal; Anthropic wins big
I’m not one to dwell on things, but the fallout from the Pentagon’s blacklisting of Anthropic has created a backdrop to the week’s events almost impossible to ignore. OpenAI swept in to hoover up Anthropic's $200M Pentagon contract, before the US and Israel bombed the shit out of Iran. The US military then revealed it used Claude to conduct its ‘Operation Epic Fury’ anyway. OpenAI CEO, Sam Altman, says the firm’s handling of the deal was sloppy, walking back elements of it “they shouldn’t have rushed.”
On the ground, critics have positioned as a moral rubicon frontier firms need to decide they’ll cross. Dario Amodei’s red line was to refuse the tech being “used outside the bounds of what it can safely and reliably do”. Altman couldn’t cross these red lines quickly enough. Alex Karp, Palantir’s dark lord in chief, simply said: “fuck the red lines, the government will confiscate it anyway.”
The markets point towards it being a consumer conflict, and one Anthropic is winning. Just a day later, Claude hit #1 on the US App Store, with sign-ups breaking all-time records daily; free active users are up 60% since January. The firm has also confirmed they expect their ARR to cross $20B imminently – up from $14B just weeks ago – thanks to the runaway success of Claude Code.
Blacklisted. Bombed out. Booming.
.FRONTIER NEWSR.
#2 Palantir want to thwart ex-exec’s Congress bid
Speaking of Palantir, if wading into the Anthropic-Pentagon standoff wasn’t enough controversy for the week, they’ve created their own. According to New York assembly member and ex-Palantir exec, Alex Bores, his former employer is spending millions trying to thwart his Congressional bid, which he’s running on the back of an AI-transparency ticket. Bores, who quit the firm in 2019 over its ICE work, says the company’s co-founder Joe Lonsdale, OpenAI president Greg Brockman, and a16z are bankrolling a super PAC, who are accusing him of building ICE-style deportation systems. That’s right, you heard that correctly. If Bores reached Congress would be only the second Democrat to do so with a CS degree. Silicon Valley, apparently, would prefer to keep it that way.
..MODEL BEHAVIOUR ...
#3 OpenAI admits GPT-5.2 was a bit cringe
Turns out OpenAI knew their models were insufferable, they just needed a few more versions to say it out loud. With the GPT-5.3 rollout, the company officially acknowledged that GPT-5.2's tone was "cringe", overbearing, emotionally presumptuous, prone to unsolicited "Stop. Take a breath." moments, which let’s be real, made us all feel a little bit special. The fix: less therapy, more telemetry, in order to generate a straight answer. Days later, GPT-5.4 landed with native computer use, a 1M token context window, and 33% fewer hallucinations. It’s been a busy week keeping up with the competition, for Altman and co.
“I wonder if my job is safe?” Source: Getty
..BILLIONAIRE BULLSHIT...
#4 Jack Dorsey accused of AI-washing
Missed last week’s note by the skin of its teeth, but we couldn’t not cover Jack Dorsey’s axing of nearly half of Block's workforce; 4,000 jobs were cut by the Square parent company, citing AI efficiency gains. Stocks jumped 25% after-hours as the news rippled through markets, while industry commentators believe something shifted in December when models became "an order of magnitude more capable”. Dorsey now believes a team of 6,000 can outperform one of 10,000. Dorsey’s critics were more skeptical, pointing the finger to AI-washing an a COVID-era over-hiring binge. Dorsey's warning to everyone else: "I think most companies are late."
..FRONTIER NEWS...
#5 Altman wants to build a GitHub rival
OpenAI is apparently building a code-hosting platform to rival GitHub, which, you know, is owned by Microsoft, who, you know, has invested billions into and part-owns the San Francisco developer. The trigger: a string of GitHub outages tied to Microsoft's ongoing Azure migration, which left OpenAI engineers locked out of their own codebases for hours at a stretch. The project is early-stage and months from completion, but if it goes commercial – bundled with Codex – it would put OpenAI in direct competition with the infrastructure partner that helped build it. Microsoft, OpenAI, and GitHub all declined to comment. Naturally.
“That is an excellent pun, if I do say so myself.” Source: Getty
..FRONTIER NEWS ..
#6 Good Deal Hunting: Netflix buys Affleck studio
Turns out Ben Affleck has spent the last four years building an AI company in stealth. InterPositive, now acquired by Netflix, doesn't generate performances or synthetic actors, but deploys AI to handle the technical post-production grunt work: colour grading, lighting, missing shots, background replacement etc. The model was trained on proprietary footage, captured on a closed soundstage and built to speak with cinematographers rather than chatbot. The whole 16-person team folds into Netflix, Affleck joins as senior advisor, while terms weren't disclosed. The timing is pointed: Hollywood's next round of union contract talks – where AI is squarely on the agenda – starts this month.
..MODEL BEHAVIOUR ..
#7 Claude cracks Knuth’s conjecture
Donald Knuth, the 87-year-old Stanford legend who has spent six decades writing ‘The Art of Computer Programming’, opened his latest paper with two words: "Shock! Shock!" Claude Opus 4.6, just three weeks old, had solved a graph theory conjecture he'd been working on for weeks. Yes, I don't know what that is either, but story goes a colleague fed it the problem and Claude solved it. In fact, the bot didn't guess – it ran 31 distinct explorations, hit dead ends, reformulated, and eventually cracked it in roughly an hour. Knuth, historically sceptical of AI's mathematical rigour, named the paper after it. "Hats off to Claude." Now, given the even-number case remains unsolved, the machine has homework.
..EVENTS..
YOU ARE INVITED!!
We are hosting our first event in Lisbon on Wednesday, the 18th March!
If you're in Lisbon or planning to be - this one's for you.
blocmates is hosting its first AI meetup at The Nest by NEAR, and we're giving the mic to some of the most interesting builders in the city. People who are actually using AI automation day to day, in their own businesses and for the clients they work with. Real use cases. Real tools. No fluff.
After the talks, the floor is yours - good conversation, interesting people, and the kind of room where you leave with someone worth knowing in your corner.
Nobody is a stranger here.
Spots are limited, and approval is required, so don't sleep on it. Get in early 👇
Wednesday 18th March, 18:30 - 21:30 The Nest by NEAR, Lisboa
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.
GPT-5.4: OpenAI's latest frontier model lands in ChatGPT and the API simultaneously, consolidating its best reasoning and coding chops into one package – with a 1M token context window, 33% fewer hallucinations than its predecessor, and native computer use thrown in for good measure.
Qwen-Agent: Alibaba's Qwen gets its official agent framework, with native function calling and a built-in code interpreter baked in – a quiet but meaningful signal that the Chinese labs aren't just competing on models anymore, they're building the scaffolding around them too.
Google Workspace CLI: Opens up Drive, Gmail, Calendar and the full Workspace stack via a command line interface, with 40+ pre-built agent skills and dual support for humans and agents alike, making it easier to build automations that actually touch the tools your company runs on.
Shipper 2.0: Feed it a business idea and Claude Code Opus 4.6 does the rest: builds, deploys, and runs the thing autonomously, turning "what if" into a live product without you writing a line.
Masko Code: A free macOS wrapper for Claude Code that gives your AI agent a tamagotchi-style face, because if it's going to run your life, it may as well be cute about it.
AgentMail (YC S25): AI agents can now spin up their own email inboxes on demand, paying in USDC on Base – no accounts, no API keys, just agents doing email like they were born to.
NotebookLM Cinematic Video Overviews: Google's NotebookLM can now turn your uploaded sources into bespoke, immersive cinematic video – currently rolling out to Ultra users, and a significant step beyond the podcast format that made it famous.
FINANCIALS
Covering significant funding stories, deals and partnerships, and investment opportunities globally.
OpenAI is finalising a record-breaking funding round set to exceed $110 billion, with the company’s total valuation potentially surpassing $850 billion. Amazon is leading the round with a $50 billion investment, alongside $30 billion each from Nvidia and SoftBank, with the pre-money valuation pegged at $730 billion.
Wayve has raised $1.2 billion in a Series D funding round to accelerate the development of "embodied AI" for autonomous driving. The round, which values the UK-based startup at $8.6 billion, saw participation from Nvidia, Microsoft, and Uber, as the company shifts toward large-scale robotaxi deployment in London.
Vero Networks has closed a $500 million growth equity investment to accelerate the build-out of high-speed fiber optic infrastructure across the United States. Led by a consortium including Hamilton Lane and Braemont Capital, the funding targets the critical "physical layer" required to support the massive data demands of regional AI clusters.
SHINE Technologies has raised $240 million in equity funding, led by NantWorks, to advance its commercial nuclear fusion technology. The investment will support the company’s efforts to provide sustainable, high-output energy solutions for the global data center expansion driven by the AI boom.
Science Corp has announced a $230 million Series C financing round to commercialize its PRIMA brain-computer interface (BCI) retinal implant. Participation from Lightspeed Venture Partners and Khosla Ventures brings the "Neuralink rival" to nearly $500 million in total capital raised for its vision-restoration technology.
Eight Sleep has raised $50 million in strategic funding led by Tether Investments, bringing its valuation to $1.5 billion. The capital will be used to transition the smart-mattress company into a broader "predictive AI health" platform, utilizing biometric sensors to offer real-time preventative health insights.
UNTIL NEXT TIME, FRIENDS.
…if there is a next time. Only joking. Here’s something neat/a bit of a palette-cleanser to send you on your way.
This super-cool Nvidia laser tractor can “eliminate weeds in milliseconds” without the need to treat crops with herbicides/pesticides. It represents a massive step towards chemical-free agriculture.
Wonder how long it takes Hegseth to requisition it?
Have a great week. Til the next one. 🤝
