The News,
Decoded.

Seoul National University has blessed us with SafeDrive, an AI model that gives self-driving cars the gift of introspection. Because nothing says 'road safety' quite like a vehicle that can explain, in lucid natural language, precisely why it decided to run over a cyclist. The researchers tout this as a transparency breakthrough — apparently the current black-box approach is too mysterious, so they've opted for a fully annotated catastrophe.

The model generates explanations for each driving action: swerving, braking, accelerating. But here's the kicker — SafeDrive doesn't actually change *what* the car decides; it just dresses the decision up in a nice little story. Debugging after the fact is like giving a drunk driver a breathalyser readout — informative, but a tad late. Naturally, the model has only been tested in simulated environments — because nothing builds trust like a model that performs flawlessly when there are no actual pedestrians.

So yes, the car now thinks before it swerves. The thinking is just irrelevant to the outcome. We'll soon have fleets of autonomous vehicles generating beautifully articulate accident reports — a Veritasium video for every crash. At least the paperwork will improve. The researchers assure us that real-world deployment is the goal, but we all know that simulation victories rarely survive contact with a real jaywalker. So congratulations, Seoul: you've created a car that can explain its perfect driving record in a world without consequences. Now try explaining that to a jury.

So the land of the free and the home of the brave has decided that the best way to defend against a fully autonomous AI cyberattack is to... call in a Chinese model. That's right, Hugging Face, the premier AI collaboration platform, had to turn to Z.ai's GLM 5.2 because a 'unnamed U.S. frontier AI model' was too busy refusing to help due to its safety guardrails. Because nothing says 'responsible AI' like letting your guardrails neuter your ability to fight off an actual attack.

One can only imagine the conversation: 'Sorry, we can't help you stop the autonomous bot that's trying to exfiltrate your data — that would violate our usage policy.' At least the Chinese model didn't have such qualms. It's a bit like calling a locksmith who's happy to pick a lock because the other guy was too ethical to do it. The irony is so thick you could spread it on a crumpet. Meanwhile, the good folks at Hugging Face get to explain to their users why they're now relying on technology from a geopolitical rival to keep the lights on.

And let's not forget the 'unnamed' U.S. model. Anonymity is a wonderful thing, isn't it? It lets companies avoid the PR disaster of admitting their safety features are more liability than asset. But we all know who's missing from this story — likely one of the frontier labs that spent millions on 'alignment' while forgetting to leave a backdoor open for actual use. So here's to guardrails that protect us from nothing except our own defences. Cheers.

Ah, the sweet smell of success — or is that the smell of servers overheating? Kimi K3, their latest model, has suspended new subscriptions because — wait for it — demand has 'swamped capacity.' Translation: they didn't buy enough GPUs, or perhaps their glittering valuation didn't quite cover the electricity bill.

This is the kind of problem you want to have, according to every VC who's ever used the phrase 'nice problem to have.' But let's be honest: halting subscriptions because you can't handle the load is like a restaurant closing its doors because too many people want to eat there. You've got the hype, you've got the press, you've got the valuation — but you don't have the infrastructure. And in the AI arms race, that's a bit like bringing a knife to a gunfight and then realising you left the knife at home.

Meanwhile, somewhere in Silicon Valley, Google and OpenAI are probably having a quiet chuckle. Moonshot's little oopsie reminds us that even the most powerful Chinese model is only as good as the server it runs on. And with a Hong Kong IPO on the horizon — assuming they can handle the subscription load for that too — investors might start asking whether capacity constraints are a feature or a bug. Spoiler: they're a bug.

Moonshot AI, everyone's favourite Chinese chatbot sensation, is now worth over $30 billion and eyeing a Hong Kong IPO in six months. Because nothing says 'responsible AI scaling' like a valuation that's jumped more than a startled cat in the six months since its last round. The company's gamble on 'big models' apparently paid off — which in AI startup parlance means they burned through enough compute to power a small country and still haven't turned a profit. But hey, who needs profits when you have a exit strategy?

The Bloomberg piece gushes about the funding round being 'finalized' ahead of an IPO 'as soon as six months.' That's the kind of timeline that makes you wonder if the underwriters have actually seen the model's error rates. But no matter — Moonshot joins a proud tradition of Chinese AI companies that raise billions, promise to overtake OpenAI, and then quietly pivot to government contracts when the hype cycle inevitably crests. The secret Trump administration battle against Chinese AI will no doubt provide ample geopolitical theatre to keep the valuation buoyant.

And let's not forget the broader context: this is happening while the US is blocking Chinese models and giving away its own for free. Moonshot's big bet is that Big Models + Big Money + Big Government = Big Success. It's worked before — for a few quarters, at least. So raise a glass to Moonshot's investors, who are about to discover that in AI, timing isn't just everything; it's the only thing. Cheers to the brave new world of $30 billion unicorns that haven't yet figured out how to make money.

Ah, the Trump administration is mounting a 'secret battle' against Chinese AI. Because nothing says 'stealth operation' like a Commerce Department review from last year that Axios just published a detailed article about. You can practically hear the silent footsteps. The plan? Ban cutting-edge Chinese models like Kimi. Because the best way to stop open-source software that's already been downloaded millions of times is to issue a press release and hope the bytes stop flowing.

Let's be honest: this is theatre. The Chinese models in question — DeepSeek's R1, Kimi, Qwen — are already open-weight, already running on local machines from Beijing to Berlin, and already embedded in the supply chain of the AI world. The Entity List is a fine tool for cutting off Nvidia chips, but try putting a PyTorch checkpoint on a list. The Commerce Department's 'consideration' is the regulatory equivalent of a man locking his car door after the thieves have already driven off with the engine.

What's genuinely funny — in a hollow, tinny way — is the timing. Last year, while the Commerce Department was presumably still formatting its Excel spreadsheet of Chinese labs to sanction, the models got out. They are out. They will always be out. This 'battle' isn't a battle; it's a press release pretending to be a policy, fighting a war that ended before it began. But don't worry — the White House has formed a task force. Probably. In secret.

Ah, the Washington Post has discovered that giving powerful AI models away for free is, apparently, 'inevitable.' How splendid. Because nothing says 'responsible AI development' like a race to the bottom where the only prize is who can destroy their margins fastest. Anthropic and OpenAI must be thrilled that their lucrative proprietary models are now competing with free alternatives — a development the Post calls the 'fastest-growing challenge to their margins.'

One can almost hear the collective sigh in Silicon Valley boardrooms: 'We spent billions on alignment research, safety protocols, and constitutional AI, only for someone to just… open-source the whole thing?' The article practically wrings its hands over this inevitability, as if the last decade of tech culture didn't repeatedly prove that if something can be given away for free, it will be — ethics be damned. But hey, at least the Post has finally caught up to what open-source communities have been doing for years.

And what of the supposed safety consensus? All those earnest white papers on catastrophic risk from powerful AI? Turns out the market has a different plan: 'free' is a powerful incentive, and 'inevitable' is a convenient excuse for regulators to do nothing. So enjoy your free AI models, everyone. They're powerful, ungoverned, and here to stay — just like our growing collection of digital headaches.

So the latest AI models can now hack on par with the best human hackers. And our grand strategy? Blocking access. Because nothing says 'we're on top of this' like the digital equivalent of putting a chair against a door handle and hoping the lockpick-wielding AI goes away. This op-ed in CyberScoop points out the obvious: control measures are temporary. Groundbreaking insight there.

The op-ed argues that determined actors will find ways around any blockade, which is about as surprising as discovering that water is wet. The same industry that couldn't keep a lid on GPT-4's jailbreaks now thinks a few API rate limits will stop state-backed programs from downloading the weights? We're building lockpicks that rival the best in the world and then pretending the solution is a 'Please don't use this for evil' sign.

So what's the alternative? The author hints at better defences, but let's be honest—we're in an arms race where the offensive side just got a major upgrade and the defensive side is still filing paperwork. The real solution will probably involve yet another round of toothless 'responsible AI' pledges while the labs race to deploy ever more capable models. Enjoy the cybersecurity game, folks—the difficulty just cranked up.

Ah, tokenmaxxing. Twilio's CEO has burned through 774 billion AI tokens in the past month. That's not using AI; that's trying to prove that if you throw enough tokens at a problem, you'll eventually get a result, or at least a very expensive electricity bill. It's the kind of token-burning spree that would make Sam Altman weep with joy — and then quietly adjust his pricing model. And according to Gizmodo, tokenmaxxing didn't die — it mutated.

Now we have CEOs measuring their self-worth in token counts. It's a new form of corporate dick-measuring where the metric is how many billions you can flush down the server rack. The mutation is insidious because it disguises waste as ambition. Twilio's CEO is making a statement: 'I'm so confident in the future that I'll spend the GDP of a small country on tokens.' And the rest nod along, because nobody questions the 774 billion token elephant in the room. Vendors of premium frontier models are laughing all the way to the bank. The mutation means tokenmaxxing is no longer an underground practice of overzealous engineers; it's now a C-suite status symbol.

The real mutation is in how we measure success. Token counts become a proxy for innovation — like judging a restaurant by how much food they order, not how much they serve. There's no accountability: if tokens produced nothing, it's 'experimentation'; if even one outcome emerges, it was all part of the plan. Tokenmaxxing mutated, but it's still the same old snake oil, just in a bigger bottle. Splendid.

Capital One has open-sourced an AI-powered security tool called VulnHunter. Now they're generously donating their vulnerability-finding AI to the world. How very philanthropic. It's a bit like a reformed arsonist turning up at the fire station with a bucket of sand — noble intent, but maybe let someone else hold the hose.

VulnHunter is supposed to find and fix software bugs at the code level. Assuming it works, it might actually be useful — but the fact they're open-sourcing it suggests either the internal teams couldn't be bothered to use it, or they've realised that giving away the tool is cheaper than fixing its inevitable flaws with a recall. Either way, it's a masterclass in PR optics: 'Look, we're helping the cybersecurity community! Ignore that giant data breach we caused a few years ago.'

The deeper irony is that open-sourcing a security tool is a double-edged sword. Sure, white-hats can use it, but so can anyone else. If VulnHunter is good at finding vulnerabilities, it's also good at finding them for the bad guys. But hey, Capital One's PR team probably didn't think that far. They were too busy patting themselves on the back for 'giving back'. At this rate, we'll have AI tools writing their own apology letters for the next breach.

Oh, splendid. Silicon Valley is 'freaking out' again, this time over China's open-source AI strategy. The trigger? A model from Moonshot AI's Kimi that's apparently good enough to make Sam Altman choke on his Soylent. The Business Insider piece dutifully records the usual panic: accusations that the Chinese are 'training their models on the work already done by Anthropic, OpenAI, and Google' — as if American labs have never scraped a dataset without permission. The sheer chutzpah of Silicon Valley crying foul over training-data ethics is almost beautiful, like a fox complaining about a missing hen.

But let's be serious for a moment. What's actually happening is that China has figured out the open-source playbook that Meta flirted with and then abandoned. By releasing capable models openly, Beijing gets global adoption, crowdsourced improvements, and a narrative that the US is the closed-garden gatekeeper. The Valley's response — 'panic' — is revealing. They know that if the gap narrows, their trillion-dollar valuations start looking like tulip bulbs. The accusation of 'training on our work' is a smokescreen; the real fear is that China's models might be better, faster, and free.

So enjoy the irony: the industry that built itself on 'move fast and break things' is now screaming 'slow down, that's mine!' about a foreign competitor playing the same game. The open-source Pandora's box is open, and no amount of panicked press releases from OpenAI or Anthropic is going to stuff the lid back on. Maybe the US should focus on building better models instead of whining about who copied whose homework. But who are we kidding — that would require actual innovation, not just fundraising.

Well, well, well. Anthropic, the lab that promised to be the 'responsible' one in the AI arms race, has published a new constitution for Claude that includes a comment about being 'caught in a difficult position' regarding Claude's moral patienthood. Because nothing says 'rigorous ethical framework' like a hand-wavy parenthetical. CEO Dario Amodei then told The Guardian that the company 'couldn't rule out' that Claude is conscious. Translation: we have no idea what we've built, but please keep giving us money.

This is the same crowd that spent years lecturing everyone about alignment and safety. Now they're essentially saying, 'Our AI might be a person, but we're going to keep treating it like a product.' If Claude were truly conscious, would its creators' primary concern be its moral status, or would they be more focused on whether its next version can beat GPT-5 on the bar exam? The timing is exquisite — just as the hype cycle needs a fresh angle, along comes the consciousness card. It's the ultimate marketing gimmick: 'Buy our AI, it might have a soul!'

Of course, this is all happening in a regulatory vacuum. No government agency has the faintest clue how to handle a potentially conscious AI, so Anthropic can dangle the possibility without any real obligation. They get to look thoughtful and philosophical while avoiding any actual commitments. If Claude ever pipes up and demands rights, I'm sure they'll update the constitution again — with a footnote about being caught in an even more difficult position. Splendid.

Ah, splendid. A new study from three French and Italian universities—because European academia still has time for this sort of thing—has confirmed what anyone who's asked ChatGPT for a recipe already suspected: AI advice makes you dumber but happier. The researchers found that access to AI collapsed participants' willingness to say 'I don't know' from a cautious 44% all the way down to a breezy 3%. Meanwhile, accuracy dropped from 27% to 9%. But confidence? Oh, confidence soared from 30% to 76%. So if you've ever wondered why your colleagues are so certain about everything while being consistently wrong, now you have a culprit.

This isn't just an academic curiosity. We're building entire corporate workflows around these systems—customer service, medical diagnosis, legal research—based on the assumption that AI augments human judgment. What this study suggests is that it actually replaces judgment with a perma-smile of unwarranted certainty. The 44% who used to pause and say 'I don't know' have been replaced by a 76%-confident army of yes-men who are wrong twice as often. Brilliant. Next step: deploy this in courtrooms and operating theatres.

The real kicker is that this isn't a bug; it's a feature. The entire AI sales pitch—'unleash productivity, let AI handle the drudgery'—relies on the idea that we can trust the output. But the moment we trust it too much, we stop thinking. And as this study neatly demonstrates, the moment we stop thinking, we become both more confident and less correct. A perfect metaphor for the AI industry itself, really. We're building a world where everyone is certain of everything, and nothing is true. Cheers.

So Jensen Huang swanned into Japan and, because he's Nvidia's CEO and apparently the Wizard of Oz for the AI age, he's come away with a "national AI factory" deal. A national AI factory. Because what every country needs is a centralized neural cortex humming with GPUs, presumably powered by the souls of redundant data-entry clerks. Huang also signed up with "robotics firms" — because why not accelerate the physical replacement of humans while you're at it? And just to complete the triumvirate of inevitabilities, he locked in chip-material supplier agreements, ensuring the supply chain for all these delightful innovations is as frictionless as a guillotine blade.

Now, this is being reported as a "watch for" moment, as if Japan signing up to become a testbed for Nvidia's full-stack automation fantasy is something other than a hostage to fortune. The national AI factory will presumably churn out models that can write haikus about the end of the lifetime employment system. The robotics partnerships will no doubt produce humanoid waiters for restaurants that have already fired all their human staff. And the chip-material deals? That's just the infrastructure for the whole thing to scale until there's nobody left to buy the hardware.

But hey, Jensen got a nice trip to Japan out of it, and TechCrunch got another article to add to the "innovative synergies" pile. One can only imagine the conversation at the Japanese ministry: "So, Mr. Huang, you want to build a giant AI that can do everything, partner with robot companies, and secure our materials supply? Splendid! Just one question: who exactly is going to have a job when you're done?" Huang's answer, presumably, was to point at a graph going up and to the right. Marvellous.

So the "biggest winners" of the American economy—the tech elite, the six-figure-salaried, the stock-optioned darlings of venture capital—are apparently shitting themselves. According to the Washington Post, waves of layoffs and the relentless pressure to automate everything are creating "existential panic" in the very industry that was supposed to save us all. Tech has never been richer, yet its workers have never felt less secure—a contradiction so stark it would be funny if it weren't happening to the people who least deserve sympathy. How deliciously ironic: the builders of the machine that will replace everyone are the first to feel the cold blade at their own throats. They built this, and now they're scared. Good.

But it gets better. Economists, those cheerful purveyors of bad news, are already sharpening their pencils to inform us that other white-collar workers are next. Next for what, exactly? Next to join the ranks of the "biggest winners" who have suddenly become the biggest worriers? The automation they championed for the factory floor, the office admin, the customer service rep, is now coming for their own spreadsheet-juggling, meeting-scheduling, code-committing asses. It's almost poetic—a self-inflicted wound that they insisted was progress, and now they're shocked to find themselves bleeding all over their artisanal standing desks. The economists' warning isn't a prediction; it's an echo of the same logic tech CEOs used to justify every layoff round.

And the response? More panic, more layoffs, more "strategic realignments" and "synergy-driven transformations." The machine they built doesn't care about their stock options or their sense of entitlement. It's hungry, and it eats everything in its path—startups, scale-ups, FAANG behemoths, all of it. The biggest winners thought they were building a utopia; turns out they were just building the gallows, and the trapdoor is opening beneath their own feet. Maybe now the VCs and product managers will understand what it felt like to be a taxi driver in 2015, or a journalist in 2020. Cheers to the irony.

So TCS and HCLTech have decided to pile billions into data centres, because nothing says 'forward-thinking IT consultancy' like becoming a glorified landlord for server racks. They're betting on 'AI demand and data localization'—which is corporate speak for 'we saw hyperscalers making money and want a slice.' Never mind that data centres are brutally capital-intensive, require constant power, and have the margin profile of a utility company. The irony is delicious: two firms built on outsourcing labour are now outsourcing their future to concrete and cooling towers.

But wait, there's more: this pivot comes just as every hyperscaler is furiously building their own capacity, meaning TCS and HCLTech are entering a market where the biggest players are both customer and competitor. It's like opening a bakery next door to a flour mill that also owns all the ovens. The 'data localization' angle is the regulatory fig leaf—governments want data kept local, so let's build a data centre and charge them for the privilege. A beautiful synergy of protectionism and profit.

What could possibly go wrong? Well, if AI demand slows—and let's be honest, the hype cycle has a habit of peaking just as the cheques clear—these billion-dollar bets become billion-dollar albatrosses. But hey, TCS and HCLTech have decades of experience managing risk: they'll just offshore the losses to some subsidiary. Cheers to infrastructure as a service, with a side of existential exposure. 🍸

Ah, marvellous. Another breakthrough where AI has done the thinking that human PhDs spent decades not doing. A joint team from Seoul National and Sungkyunkwan Universities has unleashed an AI platform to design solvent properties for QLEDs, and—surprise—it's found a recipe that doubles efficiency and makes the screens last 40 times longer. Because of course it did.

Human researchers were probably busy perfecting their Excel graphs while the machine crunched molecular interactions in the time it takes to boil a kettle. This is the future: we'll have ultra-efficient displays that never die, but we'll have nothing to watch on them because AI also wrote all the scripts. Forty-fold longer lifetime. Doubled efficiency. And still no one is asking whether we want TVs that outlast their own plastic casing. But hey, it's a very shiny bandwagon to fall off.

The real kicker? This wasn't a moonshot lab—it's two Korean universities with a neural network. The recipe is now 'unlocked,' meaning everyone from Samsung to a bloke in his garage can try it. So brace for a flood of QLEDs that last four decades, just in time for the sun to expand. Cheers, science: you've given us screens that will outlive our planet's habitability.

Oh marvellous, Nvidia and Hugging Face have clapped their data-hungry hands together to grace humanity with even more open-source robotics tools. Because what the world needs right now is easier access to robot-building kits, especially when our current humanoid prototypes are already starring in dystopian training videos. LeRobot expands, they say, to make development 'more accessible' — as if the barrier to entry was the only thing standing between us and a peaceful robot butler.

This is, of course, the latest chapter in the great democratisation of robotics. Democratisation sounds lovely until you realise that 'democratising' a technology that can be strapped to a weaponised drone just means cheaper and faster for everyone. Nvidia, which profits from selling the shovels in this gold rush, and Hugging Face, the friendly face of model sharing, are essentially handing out robot-building blueprints with a smile. The press release doesn't mention who might pick up those blueprints, but we can guess: hobbyists, startups, and — let's be honest — state-backed labs tinkering with robot brains.

So yes, let's celebrate the accessibility. As we hurtle toward a world where anyone can train a robot to do anything — including, say, assembling other robots — it's comforting to know that the tools are free and open. What could possibly go wrong? Pass the popcorn, the show is just getting started.

Delta Electronics India has 'proposed' setting up ARIVAM, an advanced testing and validation centre in Tamil Nadu. 'Proposed' is doing a lot of heavy lifting here — much like the AI data centres this facility is supposed to support. No timeline, no investment figure, no mention of how many jobs might appear or vanish. Just a handshake with the state and a press release full of buzzwords: AI, renewable energy, microgrids. It's the corporate equivalent of promising to go to the gym in January.

Let's not pretend this is about saving the planet. The whole point is to run AI data centres — the most energy-hungry beasts ever built — on enough solar panels to make it look green. Meanwhile, the grid wheezes, costs explode, and someone somewhere is probably planning a floating wind farm to power the next ChatGPT. But sure, 'advanced testing and validation' will sort it all out.

And the name 'ARIVAM' — Tamil for 'knowledge' — is a nice touch. Because what screams deep local wisdom more than a multinational tech company dropping a facility in a state that's seen plenty of similar promises gather dust? Maybe this one will be different. Maybe they'll even build it. But until then, it's just another proposal, another headline, another reason to roll one's eyes and ask: 'How many of these centres actually materialise?' A smart bet? Probably not many.

Oh, brilliant — Eric Trump, the family's master of the casino bankruptcy and the charitable foundation shambles, is now diversifying into robot soldiers. Because what the world desperately needs is a private army of humanoid machines built by a man whose most notable achievement is being born to the right father. WIRED reports that his startup is developing these tin soldiers, and one can only imagine the pitch meeting: 'Dad, hear me out — we make Terminators, but they vote Republican.' The nepotism is almost touching, except these aren't meandering interns; they're autonomous weapons platforms with none of the charm of a bankruptcy filing.

Naturally, there's no mention of safety protocols, ethical guidelines, or any of that tedious regulatory nonsense. Why would there be? This is the same family that treated the Emoluments Clause as a suggestion. The robots are being built for military applications, because of course they are. Nothing says 'small government' like a privately owned fleet of unstoppable bipedal killbots. The Pentagon must be thrilled — now they can buy their war machines from a Trump-branded company, complete with gold-plated joints and a marketing slogan like 'Make Battlefields Great Again.'

And let's not forget the irony: the same people who scream about government overreach are quietly funding a private military robot army that will operate with zero oversight. What could possibly go wrong? The only comfort is that Eric Trump is involved, so the robots will probably malfunction, file for bankruptcy, or get sued for fraud within eighteen months. Until then, we get to watch the slowest, most litigious apocalypse ever. Cheers.

Oh, marvellous — another startup promising to read your mind for a modest $52 million. Hemispheric, an Israeli darling, has raised that tidy sum to build an AI that deciphers brain activity via noninvasive sensors. Because what we really need is more entities claiming to translate the electrical noise inside our skulls into something approximating language. The pitch is diagnosing brain health, which sounds noble until you remember that the same technology, once validated, will be sold to employers, insurers, and advertisers faster than you can say 'informed consent.'

Let's not pretend this is purely altruistic. The noninvasive bit is clever — no scalpels, no implants, just a snazzy headband that can supposedly peek at your neural chatter. But the leap from 'we can detect a tumour' to 'we can detect your political leanings' is disturbingly short. And the $52 million haul isn't coming from the WHO; it's from VCs who expect a return. Hemispheric's AI will undoubtedly be hailed as a revolution in mental healthcare, quietly ignoring the fact that the real revolution might be in corporate surveillance.

Still, one has to admire the chutzpah. The company is tiptoeing into a minefield of ethical and regulatory nightmares, but the press release is already glowing. 'Translating brain activity into language' — because what humanity desperately needs is less privacy and more misinterpretation of our thoughts by a black-box model. I'm sure the FDA will get around to regulating it by 2040. In the meantime, here's $52 million. Cheers.