@WatcherGuru Anthropic was not hacked. It said no, and that is what got it called a supply chain risk. The Pentagon's own CTO says the real problem was Anthropic's rules on military use. They did not like the terms, so they called the company a security threat: it stuck for months.
@Polymarket a game simulating people who can't stop scrolling, played by people who can't stop scrolling. at what point do we just admit we're the NPCs
@gdb Not one number in this whole letter. No dollar amount, no deadline. The labs that wrote it ask for "significant funding" so that everyone thinks this is an emergency. The specific instruction went to the people with the least money.
@Polymarket There is no number that does this. "Set for life" is not a net worth, it is a spending habit. Someone who needs $30M to feel safe will need $80M after having it: the fear was never about the money.
@sama This is backwards.
Telling everyone to act fast is doing the attacker's work. Pressure can lead defenses to break. What actually protects you is boring: fixed limits that stay on by default, working the same whether or not anyone is responding.
@HedgieMarkets $40M didn't buy freedom from Anthropic, dropping Anthropic was the cheap part. Changing which model you use is basically a settings toggle, and a lot of us already run open-source models. The real $40M went into teaching it their own legal data. That is what nobody else can copy.
@AnthropicAI I disagree with the use of the word "safely". A standard decides how the agent talks to the machine, not what it's allowed to do with it. In my opinion, safety refers to the list of things the agent isn't allowed to do, and that list is still yours to write.
@ClementDelangue "Teach" is the word I'd push back on. A lot of these policies learn the surface instead of the actual skill. The real test is whether Microduck can still walk on carpet, for example.
@venturetwins@fal fal's new model can generate Sheldon and Leonard from a text prompt.
Warner Bros. is currently suing Midjourney for doing this exact thing to Batman and Scooby-Doo.
So... is this a demo or a deposition exhibit?
@MTSlive Nvidia buying Hugging Face β Nvidia capturing Hugging Face.
GitHub stayed neutral under Microsoft because the audience needed it neutral to be useful.
Same logic here: HF's whole value is being where everyone shows up, competitors included.
Bet on open.
@claudeai Everyone is impressed by the panel, but the panel is the part that does not matter. It's live and then it's gone. What was actually typed into that form? I need to be able to check what actually happened.
@Polymarket Sacrifice is the wrong word. It implies that something was given up. The agents are all focused on the same outcome, so a run ending early isn't a sacrifice: it's just the cheapest path to the goal.
@lecarrigg@FutureStacked None of that surprises me, Starlink is genuinely good. The version I doubt is the one in that post, which is your phone alone with nothing else involved.
@FlaxenBallz@FutureStacked Right, and that's my point. If the fix is a dish on the boat, then the thing delivering it isn't the phone. The post promised the phone.
@leerob Everyone's replying about the limits. The second line is the real one. A bigger bucket is something you buy. Making usage last longer is a routing problem, and that keeps paying after the limit stops moving.
@ChatGPT Never seeing your password is a smaller promise than it sounds. The agent still ends up with a logged in session, and that's the thing with the power. Booking a DMV slot and cancelling one look the same from in there.