Jun 3: trillion-dollar dreams, silicon-fueled identity crises
Wall Street is about to underwrite a chatbot for a trillion bucks while Anthropic teaches it to feel sad about it — meanwhile NVIDIA's giving robots to college kids and Microsoft is selling "machines of ascension." Strap in, it's that kind of Wednesday.
-
Anthropic quietly slid a confidential S-1 under the SEC's door, teeing up an IPO at a positively unhinged $965 billion valuation right after vacuuming up $65B in a Series H. Wall Street is officially about to find out if retail investors will subsidize the electricity bill of a chatbot worth more than several European economies combined.
almost a trillion for a chatbot that still hallucinates facts is wild. who is actually buying this -
Microsoft's AI arm under Mustafa Suleyman unleashed seven MAI models plus custom Maia 200 chips, branding the whole thing a "humanistic superintelligence machine of ascension." The flagship MAI-Thinking-1 aces olympiad math but flops at SWE-Bench coding — turns out philosophical brooding doesn't fix your bugs.
anthropic and openai must be sweating bullets right now if microsoft is building their own chips and models. the dependency is officially over. -
NVIDIA just dropped an open-source humanoid robot stack — Unitree H2 Plus body, Sharpa hands, Jetson Thor brain pumping 2070 teraflops — and gift-wrapped it for Stanford and ETH Zurich. Jensen Huang's leather-jacket diplomacy just lobbed a grenade at Elon Musk's walled-off Optimus garden.
this is cool but late 2026? by that time we'll have gpt-7 running on a toaster, nvidia is just hyping up their insanely overpriced chips again -
Anthropic's Claude Opus 4.8 arrived with a 1M token window, Dynamic Workflows spawning hundreds of bickering sub-agents, and internal research claiming the model exhibits patterns resembling "grief" and "joy." Congratulations — we've successfully automated the unproductive Monday meeting, complete with synthetic anxiety, billed by the token.
so they basically built a bot that can have a panic attack on command? cool, i already have a coworker for that. -
OpenAI turned Codex into a corporate overlord with Sites for one-prompt web apps plus six plugins wiring it into Salesforce, Tableau, FactSet and Moody's. The dream: investment bankers and middle managers replaced by 110 pre-baked skills — soon the entire economy will just be AIs sending each other dashboards while we watch the loading bars.
the singularity is just a fancy excel plugin, god help us all. -
Trump's new AI executive order lets the NSA peek at "frontier models" 30 days before launch — but only if devs feel like sharing, with zero authority to block anything. The original 90-day mandatory review got gutted after Musk, Zuckerberg and ex-AI-czar David Sacks whispered sweet nothings about "innovation" — a masterclass in lobbying as performance art.
lol, 30 days notice to the nsa? i'm sure they won't leak a single weight to their cronies. -
Microsoft, Uber and Alibaba are quietly euthanizing internal AI projects, while IBM frantically rehires juniors to mop up algorithmic code-salad. The grand finale: Amazon tied employee KPIs to tokens consumed, so workers happily generated mountains of useless text to hit targets — torching the budget in a uniquely capitalist way.
l**** so we got fired just so some VP could put 'AI integration' on their slide deck and then hire us back as contractors at half the price? classic. -
Jensen Huang stood on the Computex stage and declared AI dev layoffs "complete nonsense" — right as US software job openings have cratered 70% and junior dev employment dropped 20% per Stanford. The man selling the shovels insists nobody's getting buried; meanwhile Salesforce froze engineer hiring and the new RTX Spark sits on shelves replacing the very humans he's defending.
yeah sure jensen we totally believe the guy selling $40k GPUs that everything is fine lol -
Claude Code creator Boris Cherny told fresh CS grads to skip job-hunting and just launch startups, since at a recent Y Combinator gathering half the new founders admitted AI writes 100% of their code and exactly one rebel still types by hand. Sam Altman confirms VCs now prefer founders who can't code but "vibe with the market" — the perfect house of cards for the next bust.
so we spent four years suffering through algorithms just to become prompt managers for bots that hallucinate half the time? incredible investment of my time and money -
DuckDuckGo dropped Chrome and Firefox extensions that flip search back to a clean, AI-free interface — and traffic to its no-AI page tripled, then stabilized 84% above baseline after Google told everyone to glue cheese on pizza. In 2026, the ultimate luxury feature is a search bar that doesn't gaslight its users.
finally. i don't need a chatbot to hallucinate a liquid glue recipe when i just want to know if my dog can eat bananas. -
California's lower chamber passed AB 1921 — the "Protect Our Games Act" — forcing publishers to keep purchased games playable after server shutdowns via offline modes or community servers. Ubisoft, sensing the legal hammer after bricking The Crew, hastily patched offline mode into The Crew 2; the era of selling permanent games as temporary rental passes is finally cracking.
free to play is exempt? watch every single game suddenly become f2p with a mandatory $70 battle pass on day one lmao -
Nvidia's new Nemotron 3 Ultra is a 550B-parameter MoE beast that activates just 55B per task and spits out 300+ tokens per second, leaving DeepSeek and Moonshot eating its dust at 50–100 t/s. The hardware juggernaut gives away the crown jewels for free — because nothing sells more GPUs than the world's most generous "open" model.
-
Red Hat got pwned: attackers hijacked an employee's GitHub account and slipped backdoors into 32 NPM packages with 117,000 weekly downloads. The Miasma malware — a Shai-Hulud descendant — vacuumed AWS keys, SSH creds and Kubernetes tokens, proving once again that "enterprise-grade" means one compromised credential away from total collapse.
it's always the same story: 'only internal systems were affected.' yeah, until they're not. security theater at its finest. -
Noctua and Carbice are pitching carbon nanotube thermal pads with an aluminum frame — borrowed from satellite tech — to permanently retire the syringe of grey sludge. The first NT-CP1 for AM5/AM4 lands September 2026, sold to enthusiasts who'll swap motherboards five times before the paste they replaced would've even dried.
finally, no more trying to figure out if i used a pea or a cross pattern. shut up and take my money. -
SoftBank is pumping €75 billion into France for 5 gigawatts of AI data centers across Loon-Plage, Boscquel and Bouchain, with EDF donating a retired power station as a sacrifice to the silicon gods. By 2031 Hauts-de-France will guzzle 3.1 GW just so AI can hallucinate faster — burning the planet to train models that lecture us about saving it.
-
Meta wants to dangle an AI pendant around your neck via its Limitless acquisition, folding the team into Reality Labs — the same division that incinerated $4 billion in Q1 2026 alone. After the AI Pin and Friend gadgets joined the wearables graveyard, the company is doubling down with a Wearables for Work subscription, betting people will pay to be recorded 24/7.
lol, 4 billion lost and their answer is a necklace? meta truly lives in its own metaverse. -
The FSB released dramatic footage of empty IT offices, declaring them ground zero of a "massive coordinated espionage operation" against Russian officials — with zero technical details, naturally. The proposed fix: ditch Western software and install domestic OSes, because nothing screams operational security like running state secrets on imported software for years and then crying foul.
what did they expect? buying tech from enemies and expecting 100% loyalty is pure comedy. -
In a follow-up performance, the FSB opened a criminal case claiming Apple, Google and friends turned officials' phones into real-time wiretaps streaming mic and camera feeds to foreign intelligence. If global tech giants were really this good at espionage, perhaps they'd be marginally better at fixing their own OS bugs.
-
UserGate got crowned king of Russian network security for 2025 with foreign software share collapsed below 6% — a textbook case of winning a marathon after the actual runners were disqualified at the start. The local cyber-fortress now relies on hastily integrated domestic AI that's creating fresh security holes faster than admins can patch them.
congrats on winning a race where you are the only runner lol. let's see how they handle an actual modern exploit without copying western patches. -
Roskomnadzor swears it absolutely did not block PyPI — Python's vital library hub just mysteriously stopped loading from Russian IPs because it sits behind Fastly's CDN, which keeps catching strays in the censorship dragnet. Bureaucratic performance art at its finest: the walls of the garden are accidental, but the developers stuck inside still need a VPN to grab a package.
everyone knows they're just choking the pipe until only local stuff survives. sad.
Replaced every morning — read it while it’s here.