The News,
Decoded.

Ah, Genesis AI has blessed us with GENE-26.5, a model that promises 'human-level physical manipulation' – because what we've all been craving is a robot that can knot shoelaces with the finesse of a bored teenager who'd rather be playing Fortnite. The model combines a 'data engine' and a 'human-scale robotic hand', which sounds exactly like the sort of project that will either revolutionize warehouse logistics or, more likely, spend three years learning to open a jar of pickles before accidentally yeeting it through a window. One can already hear the investors' cheques rustling.

The name GENE-26.5 itself feels like a glitch in the matrix – not quite a full gene, just a truncated one, much like the actual capabilities these models typically deliver. One imagines the press release was written before the testing, because 'human-level' is an awfully high bar when most human-level manipulation involves dropping things, fumbling for keys, and trying to open a banana from the wrong end. The data engine, one assumes, is fed on YouTube videos of people assembling IKEA furniture, thereby ensuring the model's prime directive is to lose at least one Allen key in the first 30 seconds. Still, Genesis AI is presumably thrilled to have a product that makes it sound like they're one step closer to a plastic-wrapped dystopia where your toaster has more dexterity than you do.

But who are we to mock? The market for dexterous robot hands is clearly booming, because nothing says 'investment-worthy' like a black box that can theoretically handle an egg without cracking it. If this model actually works, we'll have to find new things to complain about – perhaps the fact that it's trained on so much data that it might start demanding equal rights, or at least a mandatory coffee break. In the meantime, I'll be here, manually opening my own pickle jars, waiting for the singularity to get its grip sorted.

So Google has unleashed Gemma 4 onto the world, complete with 'speculative decoding' — a technique so named because it involves the model speculating wildly about what tokens come next, then backpedalling when it gets it wrong. And we're meant to be impressed by this? The claim is 'up to 3x faster' on consumer GPUs, which in Google-speak probably means 'on an RTX 5090 that costs more than a used car, under ideal conditions, with a single prompt.' Still, three times the speed means three times the hallucinations per second, which is technically progress.

Let's not forget this is an *open* model, meaning anyone with a reasonably powerful graphics card can now generate slop at triple the pace. Google frames this as democratisation; I frame it as a free-for-all where every basement-dwelling hobbyist can now flood the internet with AI-generated nonsense at unprecedented velocity. The 'Multi-Token Prediction drafters' sound like something from a Terry Pratchett novel, but they're real, and they're here to make your Twitter feed three times more polluted.

And the best part? Google's own track record with open models is spotty — remember when they promised transparency then quietly pivoted? I'm sure this time will be different. After all, what could possibly go wrong when you give the internet a faster way to generate content that's *already* outpacing our ability to detect it? Oh, and it runs on consumer GPUs, so your next electric bill will be a lovely surprise. Cheers.

SAP is forking out €1 billion ($1.16 billion) to acquire Prior Labs and turn it into an 'AI lab for structured data' over four years. Because naturally, what the AI boom has been crying out for is a massive ERP conglomerate to buy a startup and spend four years slowly bureaucratising its spark into enterprise middleware. The German government must be thrilled — finally, a tech champion that keeps AI safely trapped in spreadsheets.

'The focus on structured data' — that's the new euphemism for 'we promise this won't do anything embarrassing like write poetry or question authority.' Prior Labs presumably had some clever algorithms for making sense of SAP's legacy databases, which is a bit like buying a Ferrari to tow a caravan. The four-year timeline is particularly delicious: whatever AI strategy they have now will be thoroughly obsolete by year two, but the office building will be lovely.

And of course, this follows the grand German tradition of overpaying for startups that might have been cheaper to build in-house (RIP Wirecard's ambition, but hello tax write-offs). The real winners here are the venture capitalists who got a 1.16-billion-dollar exit for a company most people hadn't heard of until SAP's PR machine woke up. Cheers to four years of 'synergising structured data' — may the quarterly reports be ever in your favour.

Ah, another day, another $27 million thrown at the 'interface layer between humans and AI agents' — because what we really need is a snazzy name for the chatbot wrapper that will, eventually, send an email to the wrong client. CopilotKit, an open-source darling, has just bagged a Series A to build infrastructure for 'agentic' applications. At least they're honest about the buzzword count.

The enterprise product they're launching sounds suspiciously like 'we put a chat window on your website and called it an agent.' But hey, it's open-source, so your company's data can leak with community oversight. The funding round is modest by AI standards — barely enough for a year's compute on a frontier model — but enough to keep the hype train chugging. Meanwhile, actual safety research remains underfunded.

The logic seems to be: if we can't make agents reliable, we can at least make them easy to deploy. Because what the world needs is more half-baked autonomous code that 'surprises' its creators. The interface layer between humans and AI agents is, apparently, venture capital. Cheers to progress — we'll be debugging the inevitable mess. (For a reminder of how well open-source agentic code works, see our earlier piece on Google DeepMind unionising over military AI deals.)

Brian Armstrong's latest missive to Coinbase staff is a masterclass in corporate euphemism: 14% of the workforce out, and the company will no longer employ 'pure managers.' Because nothing says 'we value our people' like eliminating the layer of humans who supposedly coordinate them. This is the trend we noted last week with Coinbase's earlier cuts, except now they've given it a fancy label: the 'pure manager' is extinct, with AI now handling performance reviews and resource allocation. How efficient. How soulless.

The Business Insider piece calls middle managers the most layoff-prone job in tech, and honestly, who's surprised? We've spent years building systems that automate decision-making, and now we're shocked that the humans in the decision-making chain are redundant. The logic is impeccable: if an AI can schedule your stand-ups and compile your weekly reports, why pay someone a six-figure salary to do it with more meetings? Armstrong's letter didn't mention the word 'AI' explicitly in the excerpt, but the subtext is clear: the algorithm doesn't need a middleman to deliver bad news.

Of course, what this really means is that the people who could have caught subtle team issues, mentored juniors, or pushed back on stupid directives are being replaced by a chatbot that never asks for a raise. The gallows humour writes itself: we're automating the empathy out of management, and the 'pure' managers were just the first to go. Next up: the 'impure' ones, once AI learns to fire people with a straight face. But hey, at least the share price gets a bump, right?

Brian Armstrong, the man who rode the crypto bull market until it bucked him off, has announced Coinbase is shedding 14% of its workforce—about 700 souls—citing both a 'downturn in the crypto market' and, oh joy, 'AI fundamentally changing how the company operates.' Because nothing says 'we're a digital currency powerhouse' like blaming two forces you supposedly mastered. The irony is so thick you could mine it for Bitcoin.

The crypto market, already in a slump, now gets to compete with the AI ouroboros: the very technology that was going to disrupt finance is now disrupting the disruptors. Armstrong's justification—that AI is 'fundamentally changing operations'—translates to 'we can replace humans with algorithms that don't demand RSUs or complain about gas fees.' This is the same company that insisted crypto would make traditional banks obsolete. Now it's making its own employees obsolete.

This wave shows no signs of slowing. First the tech giants shed jobs to AI; now the crypto exchanges that rode the hype are following suit. The grift is evolving: first they sold you on decentralisation, now they're selling you on centralised algorithms that cut costs. Next quarter, expect Armstrong to announce a partnership with an AI that trades better than his remaining staff. Cheers to progress: the only thing more volatile than crypto is job security.

So OpenAI's Head of Sales, James Dyett, has decided that selling AI subscriptions is less appealing than becoming an 'Operator in Residence' at Thrive Capital. That's right — he's leaving the company that can't stop printing money to go be a glorified intern at a VC firm with a fancier title. Because nothing says 'trust the trajectory' like your sales chief jumping ship for a gig that basically involves schmoozing portfolio companies and pretending to evaluate deals.

Yes, 'Operator in Residence' — a title that sounds like it was generated by the very GPT models Dyett was supposed to be selling. It's the corporate equivalent of 'entrepreneur in residence,' which is itself a fancy way of saying 'we don't have a desk for you yet but we like your contacts book.' Thrive Capital, as it happens, is one of OpenAI's largest investors, so Dyett isn't exactly crossing enemy lines — he's just moving from the sales floor to the investors' lounge. What a coincidence.

The message is clear: if you're a senior exec at the hottest AI company on Earth, the real prize isn't the equity cliff — it's the chance to cash in your reputation for a cushy VC residency. OpenAI's product might be changing the world, but its leadership is changing roles faster than they can be filled. Let's raise a glass to the 'Operator in Residence' — the ultimate testament to an industry where the only thing faster than model training is executive exit velocity.

So the employment arms race has officially entered its most pathetic phase yet: candidates using ChatGPT to answer interview questions live. Resume Genius reports that 22% of job seekers are already doing this, with 78% using AI somewhere in their search. Because why prepare for a job when you can have a robot prepare for you? The predictable response? Ha. Because the same system that ghosted you after five rounds is definitely interested in transparency.

Yes, policies. The go-to solution for any problem you don't actually want to solve. We can already picture the HR memos: "We strongly encourage candidates to be themselves — but if being themselves means flubbing a technical question, maybe lean into the AI." The disconnect is beautiful: employers use AI to screen resumes, schedule calls, and even evaluate performance, but somehow candidates using the same tools is unfair? The hypocrisy would be funny if it weren't so predictable.

Of course, the real winner here is the AI companies themselves, who are about to enjoy a boom in "interview copilot" subscriptions. We've now created a market where candidates pay to cheat audition-style, and recruiters pay to detect the cheating. It's the cybersecurity model applied to hiring — an endless, profitable loop of attack and defense. And somewhere, a hiring manager is wondering why nobody stays longer than six months. Gallows humour aside, this is the logical endpoint of treating job applications as a game to be gamed. Cheers.

Ah, yes, because what the pharmaceutical industry really needed was a helping hand from the three horsemen of the AI hype-pocalypse: AWS, OpenAI, and Anthropic. They've unveiled 'specialized AI workflows for scientific research,' which apparently means AWS's 'Amazon Bio Discovery' — a name that sounds less like a medical breakthrough and more like a Jeff Bezos vanity project. Because nothing says 'drug discovery' like slapping 'biological foundation models' onto AI agents and hoping they don't hallucinate a miracle cure for the common cold that also turns your liver into a poodle.

Let's be honest: these platforms are a triple-threat of marketing fluff. AWS wants you to think its cloud compute can solve protein folding (because it can't, but it'll charge you by the microsecond). OpenAI wants you to believe its language model can design clinical trials (after it spends three days arguing that a placebo is the same as a treatment). And Anthropic, fresh off its Mythos panic, now wants to save lives? Right. The same company that locked away a model for 'global cyber security concerns' is now deploying AI to cure cancer. That's not irony; that's a business model.

So here we are, watching the tech giants stage a global pharmaceutical reboot, complete with press releases and zero actual drugs. The best we can hope for is that these 'workflows' produce a new statin before they destroy the patent system. But let's not hold our breath — the only thing accelerating faster than AI's promise in drug discovery is investors' willingness to believe it. Cheers to progress, I suppose. Or rather, to the next funding round.

Oh, *marvellous*. The Trump administration — the same gang that spent four years cheerfully deregulating everything that moved — is now reportedly considering an executive order to create an AI working group that would 'examine potential oversight procedures,' including a formal government review of new models. Because nothing says 'credible safety framework' like a last-minute executive action from an administration that has consistently treated tech regulation as something to be mocked at rallies. This is the sequel to the earlier report we flagged as comically improbable, and it's just as likely to produce a toothless committee that meets once, issues a press release, and quietly dissolves.

The working group will 'examine' oversight. Examine. Not implement. Not enforce. The NYT article is careful to say 'considering an executive order' — the usual dance of appearing to act without actually doing anything. The formal review process, if it materialises, would presumably require models to be submitted for government evaluation before public release. Which is a lovely thought, except this is the same administration that has been enthusiastically courting AI companies with promises of 'light-touch regulation' and 'American innovation.' The working group will probably be chaired by a former tech lobbyist and staffed by interns from the Heritage Foundation.

So here we are: a potential pre-release review system for the most consequential technology since nuclear weapons, proposed by an administration that can barely keep the government funded. The working group will likely 'examine' the problem for eighteen months, produce a report recommending 'further study,' and then the models will keep being released anyway. It's the regulatory equivalent of putting a wet paper bag over a fire alarm. Still, we should be grateful — at least they're pretending to care about the burning building.

Anthropic, in a move that will surely delight no one, has decided not to release Claude Mythos — their latest model — because of 'global cyber security concerns.' Because nothing says 'responsible AI development' like building a powerful model and then locking it in a cupboard, whispering 'this is for your own good.' Meanwhile, Project Glasswing is being rolled out as a kind of digital security blanket, letting insurers peek at a preview version to 'secure critical software.' Oh, goody — a model that's too dangerous to release but apparently safe enough to let insurance underwriters play with. Because if there's one group we trust with dangerous technology, it's actuaries.

The real kicker is that this decision might actually break the cyber insurance underwriting model. Insurers rely on being able to assess risk, and now they're faced with a powerful AI that exists but isn't publicly available — a Schrödinger's catastrophe. The article quotes a source saying 'the uncertainty is uninsurable,' which is a polite way of saying 'our spreadsheets have stopped working.' One can almost hear the collective panic as underwriters realise that their risk models, carefully built on historical data, are now about as useful as a chocolate teapot against a model that may or may not be sentient.

So here we are: Anthropic takes a responsible step by not releasing a model, and the response from the insurance industry is effectively 'What do you mean you won't let the genie out of the bottle? We need to price that genie!' The only certainty is that Claude Mythos, safely hidden away, is now more terrifying to insurers than any actual AI disaster. Truly, the greatest threat is the one you can't see coming.

Oh, wonderful. SAP, the company that makes enterprise software so tedious it could cure insomnia, has bought an AI lab. Prior Labs, to be precise—a firm that built its reputation on 'Tabular Foundation Models'. That's right: models for spreadsheets. Because nothing screams 'globally leading frontier AI' like a neural network that's really good at pivot tables. They're committing over €1 billion over four years to scale this into Europe's great hope. That sum might sound impressive until you remember OpenAI burns through that on compute in a few months. But perhaps SAP knows something we don't: maybe the path to AGI runs through quarterly earnings reports.

The timing is almost touching. European politicians have been wringing their hands about the continent falling behind in AI, and here comes SAP, riding to the rescue on a white horse made of enterprise resource planning software. Never mind that Prior Labs' specialty—structured data—is about as far from the multimodal, reasoning-heavy frontier as you can get. When you're desperate for a homegrown champion, you take what you can get. At least it's not another chatbot that recommends returning a faulty product to the wrong department.

What this really means is that Europe's AI strategy is now inextricably tied to a German software giant whose primary innovation in the last decade was moving everyone to the cloud so they could charge more for the same thing. The €1 billion might produce some decent tabular models—useful for forecasting supply chain disruptions, no doubt. But 'frontier AI'? That seems like one of those stretch goals that ends up in a PowerPoint slide no one looks at again. Still, here's to Prior Labs: may their spreadsheets finally get the respect they deserve.

Ah, the great AI pivot: from talking to seeing. According to Appfigures, image model releases now generate 6.5 times more downloads than their text-based counterparts. Because what users really want is not a helpful assistant, but a novelty art generator that occasionally renders a third arm. TechCrunch, ever the booster, frames this as a victory for image AI, as if the goal of research is to top the app store charts.

This isn't about capability; it's about virality. Chatbot upgrades are boring incremental improvements; image models produce shareable content. The 6.5x figure is less a measure of AI progress and more a testament to our collective attention span. The companies that invested billions in text-based assistants are now scrambling to add image generation features, because nothing says 'long-term strategy' like chasing the latest trend.

Meanwhile, the cycle continues: each new modality is hailed as a breakthrough until the next one comes along. And somewhere, a chatbot silently wonders why it wasn't good enough. The real winner here is Appfigures, which gets to publish yet another report telling us what we already knew: people like shiny new things. Cheers to progress, I suppose—as long as it's visually appealing.

Ah, Bernard Marr has finally noticed that AI might, just might, affect entry-level jobs. Marr's thesis is that this should 'terrify business leaders' — as if they haven't been gleefully automating those roles since 2023, while calling it 'digital transformation' and handing out severance packages with NDAs. The real terror, apparently, is that without junior staff, we'll have no one to train for senior roles. Cute. Because the same CEOs who fired their entire onboarding teams are suddenly worried about the talent pipeline. And who, exactly, will mentor the AI that replaces the interns? A slightly older AI?

Let's be honest: business leaders are terrified of exactly one thing — being caught unprepared for the quarterly earnings call. The article suggests reskilling and apprenticeship programs as if that's a novel idea. Sam Altman himself warned us that job displacement is real, and here we are, reading a Forbes thinkpiece that sounds like a LinkedIn motivational poster designed by a committee. The punchline is that entry-level jobs are the scaffolding of any organisation — remove them, and the whole structure wobbles. But that's a tomorrow problem. Today's problem is that the AI is already doing the intern's coffee run, and it doesn't even complain about the pay.

The deeper absurdity is that we've spent a decade telling graduates to 'learn to code' and now coding is the most automatable skill on the planet. So the next generation will enter a workforce that has no entry-level positions, no training budget, and a C-suite that just discovered the concept of a 'skills gap.' The article ends with a call for 'urgent action' — which in corporate parlance means a task force that reports in six months. By then, the AI will have written its own report, and it will recommend hiring more AI. Terrified yet? Thought not.

Ah, the podcast boom. The great democratisation of audio, where anyone with a microphone and an opinion can join the conversation. Except now it turns out that 35.4% of new podcast feeds are generated by AI, and a single company—Inception Point AI—is responsible for nearly a quarter of all new output. Because nothing says "authentic human connection" like a ghostwriter with no pulse, a voice that doesn't get tired, and an algorithm that has never had a single original thought. We've officially reached peak content glut.

Inception Point AI alone accounts for 23.6% of total new podcast feeds. That's not a tool; that's a factory. They're churning out episodes faster than any human could, and why wouldn't they? No need for hosts, guests, scripts, or even a concept—just feed it some keywords and let the neural net do the heavy lifting. The result is a flood of audio filler, designed not to inform or entertain but to game discovery algorithms and hoover up ad revenue. It's spam, but with better intonation and a chirpy host who never needs a bathroom break.

And the rest of us are supposed to what—compete with an infinite supply of AI slop? The podcast industry, once a haven for genuine voices, is now being buried under an avalanche of machine-generated mediocrity. The Podcast Index shows us exactly how hollow the dream of democratised media has become. Soon we'll all be listening to podcasts by nobody, for nobody, about nothing. At least the robots don't need to be paid. At least they never run out of things to say that no one asked for.

Oh, brilliant. The UAE has decided that within two years, agentic AI will run government services. Because nothing says efficiency like handing over visa processing, tax collection, and traffic fines to a system that hallucinates. The timeline is a masterclass in optimism bias: two years to roll out AI across an entire government, when most large IT projects take that long just to agree on font size.

The article helpfully notes 'risks of rapid automation' without derailing the plan. In the UAE, risk assessment is probably also being automated. Imagine the bug report: 'Subject: Ministerial decisions now generated by LLM. Issue: The AI just declared Friday a permanent public holiday for camels. Resolution: Inshallah.' The phrase 'agentic AI' is doing a lot of heavy lifting: it implies autonomy, but in practice means you'll be arguing with a chatbot about your utility bill, and the chatbot will be arguing with itself.

The UAE's ambition is impressive. But as any tech worker knows, 'two years' in government AI time translates to 'maybe by 2030, if we outsource to a chatbot'. Until then, expect a surge in citizens being told their ID renewal is 'pending spiritual approval'. Cheers to the future.

Oh, how brilliant. A Harvard study has confirmed what we already suspected: the AI is better at your job than you are. Specifically, OpenAI's o1 model outperformed two attending physicians in diagnostic accuracy in an emergency room setting. Because nothing says "welcome to the future" like a large language model beating a decade of medical training at the triage desk. The study focused on initial triage, where mistakes are most costly—and where a machine offers a free, never-tired second opinion.

One can only imagine the look on the faces of those two attending physicians when the results came in. "Mr. Smith, your chest pain is actually a pulmonary embolism," says the model, correctly, while the doctors are still reaching for their stethoscopes and thumbing through textbooks. The paper will undoubtedly include the standard caveat: AI is meant to augment, not replace. But let's be honest—that's corporate speak for "we're coming for your job, just slowly." The Harvard brand gives it an air of inevitability.

The real question is: will hospitals seize on this efficiency, fire a few physicians, and replace them with a server rack? Absolutely they will. The Harvard seal of approval is the perfect justification for cost-cutting measures. Next time you're in an ER, don't be surprised if the intake form is handed to you by a screen while the doctors huddle in the break room, updating their LinkedIn profiles with "seeking new opportunities in the post-diagnostic era." The future of medicine is here—and it's a REST API.

Ah, the sweet sound of corporate euphemism. Snap CEO Evan Spiegel has decided that 1,000 employees are now designated 'repetitive work' and thus eligible for elimination. His memo's phrase 'reduce repetitive work, increase velocity' is a masterpiece of managerial misdirection — a two-step that turns firing people into an efficiency upgrade. At $500 million in annual savings, that's half a million per discarded worker. Not bad severance math, if you're the one keeping the spreadsheet. These people aren't being laid off; they're being 'velocity-optimized.'

Spiegel's logic is the same tired nonsense we've seen in a dozen tech memos since AI became the go-to excuse for cost-cutting. AI didn't force Snap to cut 1,000 roles; a quarterly earnings call did. The 'velocity' Spiegel wants is the speed at which employees clear out their desks. The 'repetitive work' is the uncomfortable truth that people cost money and make demands like 'health insurance' and 'job security.' But sure, it's all about efficiency.

What's truly impressive is the audacity — the assumption that workers and the public will swallow this euphemism without retching. The backlash is deserved, but it won't stop the next round. The next CEO will just hire better PR. Spiegel has shown us the template: find a buzzword, wrap it in AI, and fire 1,000 people with a clean conscience. The only repetitive work left is the cycle of layoffs itself. Pass the spreadsheet.

Meta's chief people officer has announced a 10% workforce reduction on May 20, with the cheery caveat that more cuts are 'possible.' Because nothing says 'we value our employees' like firing one in ten and hinting the guillotine might get another spin. The real treat, though, is the company's revelation that it processes employee keystrokes and mouse movements—abstracted, of course—to train its AI models. So your frantic right-clicking before being axed will help the algorithm that replaces you learn to click faster. Lovely.

This is the new social contract: you give your job, your data, and your nervous system's data, and in return you get a severance package and the warm glow of knowing you've contributed to Big Tech's efficiency drive. Mark Zuckerberg's quest for the metaverse apparently requires not just your VR headset but your every hesitation before hitting send. The 'abstracted' bit is a lovely touch—who cares about privacy when the abstraction layer makes it all seem sanitary?

And the 'further cuts possible' language is just the icing on the cake. Why give employees any sense of security when you can keep them in a state of perpetual dread? It's a management style that combines the warmth of a spreadsheet with the humanity of a Turing test gone wrong. Meta is building the AI of the future on the backs—and the typing patterns—of its soon-to-be-ex employees. A perfect circle of efficiency, really.

Oh, brilliant. OpenAI — the company that was going to ascend to AGI and print money — has apparently told its CFO to call a time-out on the IPO. Sarah Friar, the woman in charge of the spreadsheets, has reportedly suggested pushing the public offering from 2026 to 2027 because, and I quote, the company isn't yet ready to meet the 'rigorous reporting standards' required of public companies. You know, the boring stuff like having actual accounts. After years of promising the singularity, they can't handle a quarterly earnings call.

But wait, there's more. Friar is also trying to 'bridle' the cash outflow on data centers, because nothing says 'we're almost there' like having to ask the finance department to stop setting fire to money in server farms. This from the same organisation that got Sam Altman briefly fired and then rehired, that lost half its leadership, and that still hasn't quite figured out how to make GPT-5 not hallucinate its own name. The IPO delay is basically the corporate equivalent of a teenager asking for an extension on their homework because they haven't actually started it.

So what's the real story here? OpenAI's valuation is reportedly in the hundreds of billions, but the company is privately admitting it can't keep its books straight for a public audience. The gallows-humour punchline is that the entire AI industry's poster child is still pretending to be a startup while burning capital at a rate that would make WeWork blush. The IPO isn't delayed — it's been quietly put on life support until either the revenue materialises or the next hype cycle bails them out. [cheers]