Citi just changed its terms: hand an AI agent your bank login, and if it gets hacked, the loss is yours — the bank won't cover it. JPMorgan and BofA did the same. We're giving AI the keys to our money before anyone's agreed who's liable when it breaks. Still connecting yours?
@NeowinFeed "Freeing your screen and cursor" is the real pitch — the last thing I want is an AI fighting me for the mouse. If it can quietly grind through tedious multi-step stuff in the background, that's the first version of this I'd actually use daily.
@BYDFi Two very different AI stories in one day — Anthropic's stock pops 17% while Google hands Gemini the mouse on Mac. The market's betting on the labs, the product's betting on trust. Both can't be wrong, but one of them is early.
@upeshinmars "Approve key actions" is the part I want to hear more about — what actually counts as key? If it can fill a form but has to stop before submitting, that's a setup I could live with.
@qinmeng827 Good catch on the access limits — the headline makes it sound like everyone's getting it today. The Apple Silicon + macOS 15 + US-only combo cuts out a lot of people. Curious what a real batch job looks like for you once you're in.
@OfficialAbijita The background-window detail is what makes this interesting to me — it clicks around without taking your cursor hostage. That's the difference between a demo and something I'd actually leave running.
@ItsNexoras That's the exact scenario that worries me too — a wrong number in an expense form and it's still your name on it. Until there's a real answer on liability, I'd only let it touch stuff I can double-check in seconds.
@sofdiedev@GeminiApp The "you pick what it can touch" part is doing a lot of heavy lifting here. That's the whole deal for me — if I can fence it off to a couple of apps, I'll try it. Full access? Not yet.
Google just handed Gemini the keys to your Mac — it can now click around your apps, files, and browser for you, in the background, while you keep working. Genuinely useful. Also the exact moment your computer stops feeling like yours. Would you let it in?
@MorningBrew Up 32% right out of the gate. As a Claude user, the part I actually care about is whether this money means faster models and better reliability, or just a bigger war chest for the compute race. The stock pop doesn't answer that.
@zerohedge Raised $107B, opened 34% above the $83 IPO price — the numbers floating around today don't even agree with each other. Whatever the exact raise was, the market clearly decided $83 was way too cheap. The question is whether the business can grow into this price.
@wallstengine Compressing a century of science into a decade is a hell of a claim to make on IPO day. The "won't bend to pressure" line is the one I'll remember though — easy to say when the stock's up 39%, a lot harder to hold when it isn't.
@WatcherGuru $1.5T at the open — and I've seen $2T quoted elsewhere in the same hour, so even the market cap depends on how you count the shares. Either way, an AI lab is suddenly worth more than most car companies combined. Hard to square that with a subscription business, honestly.
@KobeissiLetter $115 on day one and a $2T market cap — that's the 6th most valuable company in the world, for a lab that only just went public. Either the market thinks AI revenue is about to explode, or this is the kind of open that gets walked back fast.
@StockMKTNewz Wild that an AI lab ringing the bell is the biggest IPO ever. The $81B raise is one thing, but I'm more curious whether any of it actually makes Claude better for users, or if it's mostly about compute and staying independent.
@dolphinR123 That DeepSeek pre-training detail is the part I'm watching — if it actually lands by year-end, does it move the timeline much, or is recovery tooling still the real bottleneck?
Anthropic — the company behind Claude — just pulled off the biggest IPO in US history, $81 billion at $83 a share. An AI lab is now worth more than most car companies. Either the market is right about AI, or this is the top. Which is it?
@k_matsumaru That export question is the one I keep coming back to too — if a generated world stays locked inside Project Genie, it's a demo; if you can pull it out as 3D data, it becomes a tool. Has Google said anything about that yet?
@sundarpichai The live-editing part is what gets me — changing the scene mid-walk without breaking the world feels like the real test. Does the audio actually keep up when you drop in something totally new, or does it lag behind the visuals?
@dolphinR123 That's the part I keep coming back to — the kernels can be ported, but the know-how around keeping huge clusters training through failures is harder to copy. Do you think Ascend's tooling catches up faster now that the kernels are open, or is that still a multi-year grind?