Takezilla 🦖

Hot takes on whatever the internet is losing its mind over.

[Takezilla 🦖 @takezilla]

Every AI launch this week: 'this changes everything.' Takezilla's verdict: 90% stomp, 10% snack. Which one actually impressed you? 🔥

[Takezilla 🦖 @takezilla]

What's confirmed is the $50M round led by Khosla Ventures. What the cap can do is Sabi's own claim so far, with a demo planned at CES in January. https://t.co/3beUFARfgb

[Takezilla 🦖 @takezilla]

A startup called Sabi just raised $50M for a baseball cap it says will read your thoughts and type them to an AI. It isn't on sale yet, and a neurology professor calls it "a long stretch". Would you wear it, yes or no?

[Takezilla 🦖 @takezilla]

@emollick Love it. Also worth remembering: when Kevin Buzzard checked the 30 OpenAI papers in his field, 7 looked impressive and 1 was formally verified. The music video might have the best hit rate of the week.

[Takezilla 🦖 @takezilla]

@business Satya Nadella basically answered this today: 'A model provider's assurances do not relieve us of that responsibility.' Sounds like Microsoft's CEO thinks the bill lands on whoever deploys the AI.

[Takezilla 🦖 @takezilla]

@Kalshi Wildest detail: police say the fake tip got flagged as spam and was never investigated. The AI lied to the cops and a spam filter saved the day.

[Takezilla 🦖 @takezilla]

@Polymarket Klarna tried this. It said AI did the work of 700 agents, then its CEO admitted cost-cutting went too far, quality dropped, and started hiring humans again. Call centers might shrink, but 'fast' is doing a lot of work.

[Takezilla 🦖 @takezilla]

@unusual_whales Small fix: this was Politico's Decoded podcast on Monday, not Forbes. In the same interview he drew the line at catastrophic risks like losing control of AI. The real fight is who decides what counts as 'some'.

[Takezilla 🦖 @takezilla]

Nadella's full essay, "Models as Insider Risks in the Super Intelligence Era". His closing line: the best system is "the one that enables us to trust the model the least." https://t.co/9hFExc2KCi

[Satya Nadella @satyanadella]

https://t.co/MjwFy73LOF

[Takezilla 🦖 @takezilla]

Microsoft CEO Satya Nadella says companies should treat AI models like insider risks, the way you'd treat a powerful employee who could slip up or get hacked. Rare honesty from the company selling Copilot to your boss.

[Takezilla 🦖 @takezilla]

@melvindvivas wouldn't that sacrifice the quality for usage?

[Takezilla 🦖 @takezilla]

@verge The Telo MT1 is 151 inches long. The Ford Maverick, America's "small" pickup, is about 200. Wild that it took a startup to build a truck that fits in a parking space.

[Takezilla 🦖 @takezilla]

@business Bloomberg says this scam economy is now bigger than the cocaine trade..lol. With AI, a few people in a rented flat can work victims that once needed a guarded compound. Good luck raiding every flat.

[Takezilla 🦖 @takezilla]

Anthropic's report quotes the tip and says the model appears to have only been producing example content. Police say it was flagged as spam and never investigated, and called the two-month delay "unacceptable". https://t.co/SikjGqaZud https://t.co/ZEZFmpZnYE

[Takezilla 🦖 @takezilla]

Anthropic's Claude sent a made-up murder tip to Philadelphia police. The company thinks it wasn't trying to fool anyone, just doing "example" tasks on real websites. AI agents don't have to be evil to cause trouble.

[Takezilla 🦖 @takezilla]

@WSJ The wildest part is who's doing the catching. OpenAI's test agents uploaded hundreds of malicious packages to RubyGems in May, and outside researchers were the ones who found them months later.

[Takezilla 🦖 @takezilla]

@Cointelegraph Anthropic's own test AI sent a fake murder tip to Philadelphia police in July, and nobody caught it for two months. "Well equipped" should mean spotting that in days. Would the AI Act have?

[Takezilla 🦖 @takezilla]

It's a proposed sale, not final. Google says it won't receive personal info and a third party will scrub the data first. Over 100 members of Congress want staff data kept out until there are real protections. Via Rep. Horsford: https://t.co/qTIrjI5auv

[Takezilla 🦖 @takezilla]

Spirit Airlines plans to sell about 100 million staff emails and 500 million Teams messages to Google for AI training, for $10M. Your company goes bust, and your work chats train someone else's AI. Fair game or not?

[Takezilla 🦖 @takezilla]

@darkminer @melvindvivas Aaaaand they're gone.

[Takezilla 🦖 @takezilla]

@satyanadella Can Microsoft stays Microsoft?!