ericd

7 ms·

91.
by ericd 2mo ago
There is a solution for this, called a CBAM, Carbon Border Adjustment Mechanism. One just went into force across the EU. It should appeal to US politicians, because it advantages the relatively clean US manufacturing base.
92.
by ericd 2mo ago
Good point. It's much more of an issue when running dense models with tensor parallelism. In that case, I'd look for an MoE model instead.
93.
by ericd 2mo ago
Yeah, I might just replicate what you're doing. Main issue right now is just finding spare vram to actually run another model in parallel... And yeah, if I train up a vision adapter somehow, I'll try to put it up/post about i
94.
by ericd 2mo ago
Was recently thinking about doing something similar, do you basically just have the qwens describe what they see for the flashes? Was considering adding a LoRa/vision head to Flash, but seems like it could take a while to get it right.
95.
by ericd 2mo ago
Hermes injects the date/time into the stream, and you tell it to web search for anything newer than the training cutoff date. Works like a charm with deepseek flash.
96.
by ericd 2mo ago
What's your setup? Happy to try to point you in the direction that worked for me.
97.
by ericd 2mo ago
It's much faster than other models on that same hardware in the same size class. I've tested a few, it's by far the fastest I've tested. And it wasn't tens* until recently. Didn't expect this to be one of my be
98.
by ericd 2mo ago
Single request depth on vllm with dspark, I'm getting ~200 tps, I'd say it's pretty snappy.
99.
by ericd 2mo ago
C4a, iirc, followed up by some specific tests for mycotoxins in urine. And yeah, it can be really tricky to find the mold itself.
100.
by ericd 2mo ago
Anecdotally, I've found yearly physicals with full blood workup very helpful in heading off issues before they become much larger issues. For example, a random blood marker pointed at mold exposure. I thought that was strange, I didn&#
101.
by ericd 2mo ago
DIY, Midea 18k BTU, Hyper -22°F, 23.5 SEER2, 15.5 HSPF. Its air exchanger outside is a good bit bigger than other 18ks, I suspect that's part of how they hit the amazing efficiency at low temps.
102.
by ericd 2mo ago
When I first heard about this store, they were using Sonnet, not even Opus, which I found really surprising.
103.
by ericd 2mo ago
Nah, this is a program that someone in CA put together, it’s not a government program iirc. It’s just a dude coordinating a group buy (which is great! Solar United Neighbors does something similar for solar, and it seems very effective).
104.
by ericd 2mo ago
I installed a cold weather variant a couple years ago, it blasted out heat during the -5 degree cold snaps of last winter, maintained its full capacity, and did it surprisingly efficiently. But probably not worth it to go cold climate in Au
105.
by ericd 2mo ago
Lots of cold weather heat pumps can be had for <$3k total, it’s not $2-3k of additional electronics. And beyond cost savings on power (especially not needing to burn $30+ of electricity per day because temps dipped below freezing and res
106.
by ericd 2mo ago
I found Laguna S to be pretty good at coding, pretty fast, but pretty bad as an agent - not proactive, would frequently stubbornly argue things that weren't true, and pretty bad general knowledge. But as a pure coding model, pretty goo
107.
by ericd 2mo ago
Good question, I heard it on a podcast, but going back to the transcript, looks like that's their forecast, not that they've hit it, they estimated a current $70B, but they've been revising their forecasts up, so yeah, it&#x2
108.
by ericd 2mo ago
Maybe, maybe not. Personally, I hope local AI eats their lunch so that the benefits are more decentralized and accrue more to society generally. I don't think you're right about that last prediction, at all. And new use cases are
109.
by ericd 2mo ago
They have something like 80% gross margins, are at a $100B/yr ARR, and are growing at 10x per year... If that keeps up, they're going to be doing more revenue than Google in a year ($400B ARR, 20% per year growth)
110.
by ericd 2mo ago
I don't think that's quite fair. Computers have been my hobby since I was 4, I'm having a blast, this is so much fun . I've always had way more ideas I wanted to try than I've had time/energy to implement, esp
111.
by ericd 2mo ago
Our colo experience was pretty smooth, nothing insane like that, and saved us a bundle. Apparently ymmv. I hate working with AWS APIs by comparison, some of the worst UX I've ever seen.
112.
by ericd 2mo ago
Funnily enough, I seem to remember some people saying that compiler generated assembly wasn’t as good as handmade assembly.
113.
by ericd 2mo ago
For sure, it's not going to be the same, but we need to be able to feed our missile launchers for more than a couple of weeks if we get into a peer conflict. That means building a lot of big physical things, quickly, and transporting a
114.
by ericd 2mo ago
If by heavy industry you mean shitloads of steel production, I mostly agree, but if you count things like domestic rare earth minerals processing, semiconductors, solar, batteries, I'm not so sure. I don't know the official ontolo
115.
by ericd 2mo ago
Well, manufacturing capacity was the basis of US military dominance, which was the underpinning of the post WW2 US-led world order. That's pretty clearly eroded, and we're not sure we could win in a war against China, especially n
116.
by ericd 2mo ago
Heh well and encryption/decryption.
117.
by ericd 2mo ago
Your home router is a pretty good place to run some services… Highly recommend running VPN so you can access your home network from anywhere.
118.
by ericd 2mo ago
401ks don’t guarantee results for a given amount of input, and people still stuff away lots into them?
119.
by ericd 2mo ago
Human spammers frequently don't think they're spamming, they're just marketing. They'd say they wouldn't consider spamming, either.
120.
by ericd 2mo ago
You can kind of get a sense by running these things at home - I'm currently running Laguna. One interesting thing is that per stream doesn't actually slow down that much with multiple concurrents, because the bottleneck remains th

More ›