SwellJoe

11 ms·

271.
by SwellJoe 3mo ago
This is exactly the kind of model that's been needed in the middle. Realistically self-hosted, Good Enough intelligence, MoE so it's fast on limited bandwidth systems like Strix Halo and DGX Spark. For a while there's been no
272.
by SwellJoe 3mo ago
It's also got vision and audio. So, the better comparison is any of the other large Chinese open models that are better and cheaper than Gemini Flash.
273.
by SwellJoe 3mo ago
3.5 Flash was always too expensive for a "flash" model. They marketed it as "near frontier" level, but there are several order-of-magnitude cheaper open models that compete with it.
274.
by SwellJoe 3mo ago
I'm not willing to believe in phone-based frontier models anytime soon. Though, Gemma 4 12B is a beast that runs comfortably on the current top of the line phones (or would run fine if allowed to run, I think there's some kind of
275.
by SwellJoe 3mo ago
"I think that in 10-15 years, we are going to have consumer PCs (and phones!) running models doing pretty much anything that frontier models can do right now." I don't think it'll take 10-15 years. Gemma 4 31B in the 4-b
276.
by SwellJoe 3mo ago
Even very large models failed tests like this up until a year or two ago...this is a very, very, small model.
277.
by SwellJoe 3mo ago
I've been seeing more "grounding" messages in the thinking log of Claude models. I guess that's "do a web search, read the docs", probably designed to mitigate hallucinations. I am surprised by how much the rea
278.
by SwellJoe 3mo ago
In my benchmarks of security auditing abilities of models, DeepSeek was roughly an order of magnitude cheaper than either GPT 5.5 or Opus 4.8, less than ten cents per task vs. roughly a buck each for the best American models at the time. GP
279.
by SwellJoe 3mo ago
You still don't know what's going on in there.
280.
by SwellJoe 3mo ago
That's interesting. The "best" current models, Fable and especially GPT 5.6, are also lying liars that lie all the time. Seems like we're going the wrong way on hallucination.
281.
by SwellJoe 3mo ago
Nobody does caching as well as DeepSeek, so I guess it's a big enough difference in the implementation to make it difficult. If you use Reasonix with DeepSeek it gets silly, as it is append-only to work with how caching works. It gets
282.
by SwellJoe 3mo ago
I find the 4-bit QAT with MTP to be entirely usable speed on both my boxes (Strix Halo and a desktop with two V620 GPUs, which are slightly faster than the Strix Halo).
283.
by SwellJoe 3mo ago
I have seen occasional weird behavior that I guess could be attributed to hallucinations, but for security auditing, DeepSeek v4 Pro is among the best models I've tested, competitive with Opus 4.8 and GPT 5.5 (MiMo and GLM also did wel
284.
by SwellJoe 3mo ago
Qwen is the most censored of the Chinese models in my testing, which makes me wonder in what other ways it is compromised. Open weights doesn't really reveal what's in there. And, in my tests, existing Qwen models are not at the p
285.
by SwellJoe 3mo ago
Kimi K3 weights aren't available yet. And, the claim is about Inkling. I'm not saying it's wrong, I just don't understand why Inkling would be the trigger, rather than a prior model that's better, at least for code.
286.
by SwellJoe 3mo ago
I'm not sure I buy the direct causation? Why didn't GLM 5.2 have that effect? It's a better model, by some measures, and a little easier to run as it's quite a bit smaller.
287.
by SwellJoe 3mo ago
That sounds complicated. I'll just use my month of Kimi and then cancel. I have too many AI subscriptions to use them all, anyway. I subscribed mostly to test it. I mean, if it turned out to be competitive, I would keep it, but if it
288.
by SwellJoe 3mo ago
It let's me choose different thinking levels in Kimi Code. Not sure if it actually works, yet, but it says "Thinking set to high." when I change it from max.
289.
by SwellJoe 3mo ago
Yep, with Reasonix, DeepSeek is free real estate. Seems to just go and go for pennies. And, DeepSeek is what I use for any task that works best with an API. It's cheap enough to where I don't think about cost, made even cheaper by
290.
by SwellJoe 3mo ago
Yeah, I'm finding I end up switching to Codex and GPT 5.6 a lot lately because I've either run out of Fable usage or Fable refused to do the task. Most recently it refused to work on a WiFi configuration UI for a robot. No idea wh
291.
by SwellJoe 3mo ago
Not sure how the economics work for the Chinese models, but DeepSeek did the same task for a dime.
292.
by SwellJoe 3mo ago
I tried Kimi K3 on a task I've done with every other model I use regularly ( https://swelljoe.com/post/i-let-every-agent-implement-its-ow... ) and found it chewed a lot longer on the problem and ate up almost the
293.
by SwellJoe 3mo ago
They're being generous.
294.
by SwellJoe 3mo ago
Investors are clearly not familiar with anything Musk does, because it's all wildly overvalued. A dip because of a scrubbed launch is a blip on the radar compared to the catastrophically bad ideas Musk is promising to implement. Data c
295.
by SwellJoe 3mo ago
This is why I'm not wasting my time on these things (either in participating or in paying attention to the results). The noise is just too high. I'd be furious if I'd spent a bunch of time on doing actual work for something l
296.
by SwellJoe 3mo ago
I buy DeepSeek and MiMo directly from them, on the assumption they're better at it, have a vested interest in me having a good experience and they'll get caching and other stuff right. They're also cheap enough that it's
297.
by SwellJoe 3mo ago
I think that's probably a good thing. Sycophancy seems to be correlated with AI psychosis. GPT 4o was creepy sycophantic and has a body count. It'll be good for chatbots to be more interested in facts than in agreeing. (Then again
298.
by SwellJoe 3mo ago
Oh, she's sassy.
299.
by SwellJoe 3mo ago
Even their demo clips are soundly in the uncanny valley, and I assume normal folks using it and having to pay for each clip will have generally worse results.
300.
by SwellJoe 3mo ago
That's not a rename, though, Antigravity is a new product. Google is kinda famous for killing a product and replacing it with a worse one, though, so not very surprising.

More ›