SwellJoe

35 ms·

541.
by SwellJoe 5mo ago
OK, I just tried it with the new mainline ROCm and MTP support, and it is faster, but still uncomfortably slow for interactive coding agent use. It does about 14-15 t/s, which is faster than the 10-11 t/s I was seeing before, but
542.
by SwellJoe 5mo ago
Which public numbers are you referring to that contradict what I said?
543.
by SwellJoe 5mo ago
The cheap money for subsidizing tokens has begun to run out. Not all gone, yet, but it's getting harder to pretend the chatbots are cost-effective to run. Soon, they're going to need to tap a larger pool for money: Everyone's
544.
by SwellJoe 5mo ago
No source, just back of the envelope math. 100k seems optimistic, but I guess I'll try it and see. That would be usable for at least a few use cases, including the security scanning work I'm focused on at the moment (at least, so
545.
by SwellJoe 5mo ago
No, that's a compatibility thing after they changed the behavior of the aliases. Or maybe it was calling `reasoner` instead. Whatever it was, the billing definitely showed 100% DeepSeek V4 Pro usage for the benchmark. My only usage w
546.
by SwellJoe 5mo ago
As far as I can tell you'll have a context limit of about 64k, which is also prohibitive for serious work. (My benchmark maxes out at 90k in context when running, so I'm giving the self-hosted models 128k to leave plenty of wiggle
547.
by SwellJoe 5mo ago
They're leaving us in the dust on solar, while our current administration is still trying to put people in the ground to dig up more coal and die of black lung. https://en.wikipedia.org/wiki/Solar_power_in_China
548.
by SwellJoe 5mo ago
In recent benchmarking I've been doing, DeepSeek V4 Pro was the fastest of 21 models, by a comfortable margin ( https://swelljoe.com/html/bench-report-final.html ). Faster than Claude Opus 4.8, which was the second
549.
by SwellJoe 5mo ago
DeepSeek is the fastest model in the benchmarks I've been doing ( https://swelljoe.com/post/will-it-mythos/ ). Followed not so closely by Opus 4.8 and even less closely by Gemini 3.5 Flash and GPT 5.5. I'v
550.
by SwellJoe 5mo ago
I was surprised by Sonnet's performance, as well. And, it's difficult to say any model is really worse or better based on one attempt across nine bugs (several of which have proven to be intractable for all models, thus far). But,
551.
by SwellJoe 5mo ago
You can theoretically self-host. DeepSeek is big . DS4 (the 2-bit quantization of DeepSeek Flash) runs on my Strix Halo with 128GB, but it's slow as hell. Completely unusable for interactive work. But, I guess a company that cared abo
552.
by SwellJoe 5mo ago
I hope that someday there are AI companies for whom ethical behavior is a selling point. We're certainly not there for the current leaders, though vibes vary a little bit between them. Some seem scarier than others.
553.
by SwellJoe 5mo ago
These days I'm also worried about US companies having my data. I hate that we're at that point, but with Trump talking about taking an ownership stake in AI companies, and tech companies, including the leading AI companies, lining
554.
by SwellJoe 5mo ago
I'll also note that the DeepSeek API seems to be really good at caching and their cached input price is more heavily discounted than most providers at $0.003625 (vs. $0.435 for input cache misses). So, it's hard to spend a lot of
555.
by SwellJoe 5mo ago
I used the native DeepSeek API at deepseek.com. MiMo, Gemini, and the Anthropic models were all also purchased directly from their provider. The other models in the bench were either on OpenRouter or self-hosted.
556.
by SwellJoe 5mo ago
It's already there. It performed well. And, it'll be in the replication run later, as well.
557.
by SwellJoe 5mo ago
I tried adding GPT 5.5 Pro to a vulnerability scanning benchmark I made ( https://swelljoe.com/post/will-it-mythos/ ), and it blew through the $100 budget limit halfway through. DeepSeek V4 Pro cost about a dollar f
558.
by SwellJoe 5mo ago
They're all betting ignorant retail investors will be the final bagholders. It's a license to steal from retirement accounts.
559.
by SwellJoe 5mo ago
You've used the word "inevitably". Are you sure it's inevitable? SpaceX is launching at a ridiculous valuation, has two bad businesses bolted on to one modestly successful one, and all together the revenue puts the compa
560.
by SwellJoe 5mo ago
Is it profitable, though? I mean, $920M here, $920M there, starts to add up to real money eventually, but xAI has spent quite a lot on GPUs, infrastructure, and has ongoing costs for power and cooling. That said, I also think it's inte
561.
by SwellJoe 5mo ago
I presume they're selling their compute because they don't have the capability to use it profitably themselves, which indicates massive capital misallocation.
562.
by SwellJoe 5mo ago
It's years away from $50B/year profit, if it ever gets there. The IPO valuation is insane.
563.
by SwellJoe 5mo ago
So, the idea is to let everyone experience what it's like to be a billionaire, at least in one way: there's always someone to say, "yes", to anything you say. Sounds healthy. After all, it's not like the world'
564.
by SwellJoe 5mo ago
Yeah, I think publishers/advertisers are actually leaving money on the table by using Google ads instead of bespoke targeted ads from direct sponsors. It's a lot harder to manage ads without Google doing all the work, but they&#x2
565.
by SwellJoe 5mo ago
Ah, yeah, Sheldon Brown was great. I was sad when he passed (nearly two decades ago, now). I'm glad to see someone cares enough to keep his site running all these years later. It hasn't ever stopped being useful information.
566.
by SwellJoe 5mo ago
OK, that feels like a choice.
567.
by SwellJoe 5mo ago
Yes, Ian doesn't do any bullshit with ad block detection or blocking. The site works fine with ads blocked. I'm reminded that there should be a tiered ad block tool. I would like to be able to not block traditional ad banners (a
568.
by SwellJoe 5mo ago
That's the worst thing about Google Ads. Google keeps an absurd amount of the money, and they punish your users. And, of course, ads in general became so abusive and intrusive that everyone uses an ad blocker, making ads less effective
569.
by SwellJoe 5mo ago
A tremendous amount of care went into the site, and it shows.
570.
by SwellJoe 5mo ago
I always use this site as a canonical example of The Good Internet. The kind of site that is rare today but used to be most of the internet, and we're all worse off for the change. Lightweight handmade HTML and CSS. Very little JavaScr

More ›