wmf

7 ms·

151.
by wmf 2mo ago
This is a terrible patent (it's basically a list of features) and unfortunately Red was able to milk it for almost 20 years.
152.
by wmf 2mo ago
Maybe the microcode is copyrighted.
153.
by wmf 2mo ago
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 identity and payment (x402).
154.
by wmf 2mo ago
That's nothing to do with AI though. Attributing cyber attacks is hard. We know this and people have experience doing it.
155.
by wmf 2mo ago
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) regulation.
156.
by wmf 2mo ago
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.
157.
by wmf 2mo ago
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 to rent those GPUs. This money would have to
158.
by wmf 2mo ago
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 of people are making LoRAs for gener
159.
by wmf 2mo ago
Also Nemotron and Arcee.
160.
by wmf 2mo ago
$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).
161.
by wmf 2mo ago
FPGAs already configure themselves on boot.
162.
by wmf 2mo ago
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
163.
by wmf 2mo ago
Taalas does not have cache so... I agree that Groq with multilayer hybrid bonding could be a good idea.
164.
by wmf 2mo ago
Each cartridge costs $1,000. Do you still want it?
165.
by wmf 2mo ago
There's a benchmark for this and a lot of models get negative scores because they're so unreliable: https://artificialanalysis.ai/evaluations/omniscience
166.
by wmf 2mo ago
This somewhat reminds me of the IBM spiral cache. You should talk to Prof. Volker Strumpen.
167.
by wmf 2mo ago
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.
168.
by wmf 2mo ago
The methods for splitting weights across multiple chips are well established. Groq/Cerebras can't hold a model on one chip either.
169.
by wmf 2mo ago
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.
170.
by wmf 2mo ago
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 consider it a waste of tokens since most usage doesn't
171.
by wmf 2mo ago
OpenAI and Anthropic are both designing ASICs.
172.
by wmf 2mo ago
Nope, a small model would be larger than the whole iPhone SoC.
173.
by wmf 2mo ago
If you limit it to 480i I think 4.5 Mbps would be fine. If you use a cherrypicker sports could burst to a higher bitrate.
174.
by wmf 2mo ago
That's not what I'm talking about though. We should have repacked ABC, NBC, CBS, and Fox onto a single channel.
175.
by wmf 2mo ago
Yeah, that's one channel. IMO it's wasteful to give PBS an entire 6 MHz.
176.
by wmf 2mo ago
I heard in some European countries they packed digital SD channels into less spectrum (one 6 MHz channel can carry 4-5 TV channels). We should have done that here instead of broadcasting HD.
177.
by wmf 2mo ago
Prices are higher, hence inflation. People aren't going to understand devaluation because they think $1 = $1.
178.
by wmf 2mo ago
AMD announced a faster CPU two days later.
179.
by wmf 2mo ago
With apologies to Obama, if you like Cloudflare you can keep your Cloudflare. It looks like celld is not intended to be an exact clone of Cloudflare but is for someone who wants to run a bunch of workers in a single region.
180.
by wmf 2mo ago
Maybe Wrangler is a better way to do that.

More ›