Viewing profile — wmf

There is Ashton Kutcher. And it's conceivable to me that some wannabe-Palantir private companies might back these ideas.

Yeah, the US is getting the worst version. We can't have federal digital ID because "mark of the beast" but you have to show ID to watch porn so a bunch of private ID verification …

They need to heat buildings in the summer? Or are they going to store heat for months? There's something I'm missing here.

I think the patent literally covers the idea of a camera that records raw video internally. (Still cameras supported raw for years before Red existed.)

This is a terrible patent (it's basically a list of features) and unfortunately Red was able to milk it for almost 20 years.

Maybe the microcode is copyrighted.

The missing primitive for agents is not a browser. It is a portable identity and a spendable balance... It's funny you should mention that since Cloudflare is also working on bot i…

That's nothing to do with AI though. Attributing cyber attacks is hard. We know this and people have experience doing it.

Because a coverup would be worse. The labs have been talking about AI safety and transparency for years, so if they start to cover up problems it's asking for (unpredictable) regul…

It's an accident and the victim decided not to press charges. That's it. Every day there are crimes that aren't prosecuted because they just aren't worth it.

The specific predictions here are (1) SpaceXAI will borrow ~$250B to build at least 5 GW of GPU capacity in 2027 and then (2) OpenAI/Anthropic/others will pay SpaceXAI ~$300B/year …

Yeah, the dream is to have a "reasoning kernel" that you can plug specific knowledge and skills into ("I know kung fu"). I'm not aware of any progress on the dream though. Plenty o…

Also Nemotron and Arcee.

$1,000 only gets you the Qwen 27B cartridge. For Kimi K3 it would be more like $100,000 (and the "cartridge" is the size of a refrigerator).

FPGAs already configure themselves on boot.

That math doesn't really work. 8B model (FP4) = 4 GB DRAM = 32 Gb DRAM = 80 mm2 8B model (Taalas) = 4 GB ROM = ~800 mm2

Taalas does not have cache so... I agree that Groq with multilayer hybrid bonding could be a good idea.

Each cartridge costs $1,000. Do you still want it?

There's a benchmark for this and a lot of models get negative scores because they're so unreliable: https://artificialanalysis.ai/evaluations/omniscience

This somewhat reminds me of the IBM spiral cache. You should talk to Prof. Volker Strumpen.

Without oversight, a self-defense militia may expand beyond its original scope in unfortunate ways. The history of the Italian Mafia can be instructive here.

The methods for splitting weights across multiple chips are well established. Groq/Cerebras can't hold a model on one chip either.

somewhere = an AWS data center with multiple layers of security and NDAs They won't sell/rent/license the weights to an end user at any price because they don't trust your security…

AIs don't intrinsically know anything about themselves so they often give wrong answers to such questions. This can be fixed by putting info in the system prompt but they may consi…

OpenAI and Anthropic are both designing ASICs.