HNHacker News
TopNewBestAskShowJobs

Balinares

885 karma · joined August 5, 2024

I was born on Earth, and have been living there ever since.
submissionscomments
Balinares··on Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint
I wonder how well it performs in practice, because I can't help seriously doubting those benchmarks. That would put this 6GB model in Opus 4.6+ ballpark. Granted, that's mostly to Qwen 3.8's credit, but it's hard to believe that Qwen's already unbelievable capability density can still be compressed this much more.
Balinares··on 25 years of mass surveillance is enough
The number of people abject enough to end up with transphobic views is thankfully nowhere near that. Most people, as it turns out, are not creepily obsessed with other people's genitals. Most people also aren't driven by a horrible urge to find minorities to torment. Though those who are would sure claim otherwise, of course; trying to normalize their awfulness is part of the playbook.
Balinares··on 25 years of mass surveillance is enough
Because then they'll move on to the next victims and in having surrendered without even a fight you've granted them momentum for no good reason whatsoever, and you'll have fewer allies left too.
Balinares··on I can't stop thinking about Papua New Guinea
You are however under no obligation to grant your instincts unrestricted access to the pilot seat.
Balinares··on Irish, Scottish and Welsh leaders meet to plan break-up of United Kingdom
Scotland at least mulled independence quite seriously pre-Brexit, and one of the strongest arguments against which the UK advanced was that in leaving the UK Scotland would also leave the EU. The EU presidency, at the time held by Spain which was also concerned about its own independantist movement, supported that argument.

Then the UK voted for Brexit. Scotland and Northern Ireland voted against but England voted for it, and it outweighs both, so the UK left, dragging Scotland and Northern Ireland along and out of the EU. Neither were hugely amused.

So to answer your question, the Venn would likely be somewhat disjoint. That's a simple summary of a messy bag of thorns, though, so the details are likely way more nuanced.

Balinares··on An open letter to Dario: if you mean it, open the weights
He sells himself as an ethical guy, and doing things that are broadly positive for everyone even if they're against your own personal self-interest is basic ethicality.
Balinares··on AI researchers debate how close we are to recursive self-improvement
How about connecting the sadist to the caring environment where they might heal and, in time, learn empathy.
Balinares··on I resigned from Anthropic today
Humans need food. Food does not just appear in supermarkets.
Balinares··on "We Have to Assume That the Internet Will Go Offline in the Next Few Years"
Nvidia is party to a number of circular funding deals that would land them in an exciting situation if the other involved parties, i.e. frontier US labs, were to bite the bullet, for instance if the market in general broadly switched to open weight models as better value for the dollar.
Balinares··on GLM-5.3-Flash
Thank you, I know about those. And I'll stick with this quant. Normally I'd be with yout there, but Qwen 3.8 is turning out to be good at self-correcting, and the free memory I can use for extra context is worthwhile.
Balinares··on I Don't Think I Can Stay in Tech
OP is calling blatant fascism fascism. I am utterly baffled by your position.
Balinares··on I Don't Think I Can Stay in Tech
Disagreeing with fascism seems like a reasonable position to me.
Balinares··on True Rate of Unemployment
For real. I'd pay good money for a concierge service to find me the right person.
Balinares··on Sovereign Tech Agency invests €500k in Flatpak
The idea of democracies is that the government is answerable to its people, though the idea may seem quaint in this day and age.
Balinares··on GLM-5.3-Flash
I've found it tends toward long thinking loops even for simple tasks (and any quantization seems to increase their length), but those do exit eventually, unlike with Qwen 3.6.

I use the Unsloth UD_Q2_K_XL GGUF with default parameters, along with that custom template linked elsewhere in the thread, and no K/V cache quantization.

Balinares··on GLM-5.3-Flash
Interestingly, I've found the output of Qwen 3.8 27B to be competitive with Opus (4.6-ish anyway, not quite 4.8), but the experience is very different.

Where Opus has seen it before and knows how to do it, Qwen knows how to work it out. It turns out it's surprisingly capable at working things out. The obvious drawback is that it takes tokens and time.

All the same, getting to run something this capable locally is momentous, and suggests to me that streaming tokens from colossal data centers might not be the long term path forward.

Balinares··on Unsloth Dynamic 3.0 GGUFs
Wow, interesting. What KV cache quantization do you use?
Balinares··on Unsloth Dynamic 3.0 GGUFs
I anecdotally observed the same.

Interestingly, it also seems to tend toward self-correcting, which makes lower quantizations borderline usable. There'll be more faffing around, but still converging toward a solution. I wonder if that's a deliberate product of its RL.

Balinares··on Cerebras CS-4
Qwen 3.8 27B beats Opus, Fable and GPT 5.6 by a comfortable margin on the AA-Omniscience Hallucination Rate benchmark.
Balinares··on Qwen3.8 27B scores 52 on Artificial Analysis
And once again, Qwen 3.8 27B beats Opus 4.6, what the hell.

It's both funny and a bit terrifying and I still can't quite believe it. It runs decently on a gaming PC! Opus 4.6 came out only 6 months ago and was then broadly considered the new SOTA by a comfortable margin! How in hell did they package capability in the ballpark of a Feb 2026 frontier SOTA into 27B?!

More importantly, what's the point of building monster-scale data centers on unprecedented amounts of debt when a more than good enough model runs on a GPU from a couple years ago?

The coming months are going to be exciting, that's for sure...

Balinares··on Qwen 3.8 27B is excellent, but it defaults to overthinking things
Reproducing the post's money quote here because it's absolutely the crux of why the Qwen 3.8 release is seismic IMO:

"The models at this size continue to get better at an impressive rate. We don’t need to spend half a million dollars on datacenter-class hardware just to run a competent model."

Balinares··on Qwen 3.8 27B is excellent, but it defaults to overthinking things
Worth noting that the default GGUF template sets the reasoning to xhigh. You can use the Froggeric template to set reasoning to medium instead: https://huggingface.co/froggeric/Qwen-Fixed-Chat-Templates

Worth noting as well that the weights come with an MTP layer that seems particularly accurate, to the point it can give you up to 6-8 correctly predicted tokens consistently enough to be useful. Which obviously boosts its speed enormously.

I find it difficult to believe how good this model is. It feels like it's lagging heavyweight frontier models by less than a year, and it runs on your PC.

Balinares··on GIMP Development Update
Haven't those existed in Krita forever?
Balinares··on Qwen 3.8 27B
That's a reasonable point, so let me qualify: at a glance, it seems Qwen 3.8 27B can trade blows with Opus 4.6 on coding tasks, where trading blows doesn't necessarily mean it's a clear winner or even an equal, but does mean it'll at least hold its own and land a punch or two.

(Which I still think is bonkers, FWIW.)

Opus 4.6 is an especially interesting comparison point, I think, because it was a step change; IMO it's when LLMs became serviceable for coding. Yeah, pre-4.6 models did output code, and that code often superficially worked; and bringing it up to production standards still generally meant rewriting it entirely. Opus 4.6 is when that changed.

From my early tests, it's looking like the public benchmarks are not misleading, and Qwen 3.8 somehow got there too, by and large. I've got a few personal tests. One is a mid-complexity one-shot, purposefully underspecified. Beyond a few minor bugs that it could easily fix once pointed out, Qwen 3.8 largely aced it. There are a bunch of things I'd improve, but that was true of Opus 4.6's output too, and by and large the code is clean and well structured. Also worth noting that I'm running Qwen 3.8 fairly aggressively quantized to fit in VRAM; I'd expect tighter results still from the full weights.

Another test I ran is a variant of a common puzzle with an additional structural constraint that makes the usual solution inapplicable, so the model has to actively turn away from the well-known solution and construct a new one that takes the constraint into account. I've never seen a home model pass that test. Kimi K3 passes it, GLM 5.2 passes it (painstakingly). Qwen 3.8 struggles a lot... but does arrive at a correct solution. First time I see a home model do so.

I haven't yet tested it on long multi-turn scenarios. In my experience, that's where pocket models are weakest against heavyweight ones, especially when quantized.

That aside, it does seem like Qwen 3.8 can, in fact, trade blows with Opus 4.6. I don't know yet if it could replace it, and my money would be on no, but I may well be wrong about that considering how weirdly capable it is.

Interestingly, Qwen 3.8's MTP layer is uncannily accurate too. It still gave me good results up to 6 to 8 predicted tokens, which boosts its speed so much it's competitive with Qwen 3.6 MoE. So that's another bizarrely impressive thing about it.

And given all of the above, I do think that the trillions of dollars invested into OpenAI and Anthropic are becoming harder and harder to justify.

Balinares··on Qwen 3.8 27B
I wonder if Anthropic and OpenAI possibly missed the window to go public. A 27B open-weight model trading blows with the SOTA from just half a year ago is not great news for trillion-dollar investments...
Balinares··on The Strongest El Niño Ever Forecast and the Hunger It Will Leave Behind
I wonder if that's what deer feel, yeah. See what's coming, see it right there, and yet just... don't quite believe it.
Balinares··on The Strongest El Niño Ever Forecast and the Hunger It Will Leave Behind
I find I can't even process that models from only 6 months ago, whose predictions were already somber, turned out to be overly optimistic by a factor of 50% and climbing. And life appears to just go on for now, at least for my privileged arse and hopefully yours also, even as we watch in real time a rising tide biting into the base of the metaphorical sandcastle we're sitting on. Goodness.
Balinares··on Hyperspace
What makes it a (crystal clear IMO) dark pattern is that the purportedly "free" feature set gives the user no value whatsoever. It only exists to induce sales. Inducing a high-bar commit (the purchase) by first securing a low-bar commit (install + scan) is an old trick. Don't fall for it.
Balinares··on A Growing Threat: The Anti-Rights Movement in the UK (By Amnesty International)
Amnesty was indeed threatened with lawsuits from anti-rights orgs that thereby did not in fact beat the anti-rights allegations.
Balinares··on Microsoft Edge is about to lock out older ad blockers, just like Chrome did
In case your assumption here is that TFA is about Edge disabling adblockers: it's not. It's just removing ManifestV2 support. ManifestV3 is still supported. Despite what one might assume, MV3 does not seem to measurably diminish the effectiveness of adblockers; if anything, it improves it. (Source: https://petsymposium.org/popets/2026/popets-2026-0027.php)
← PreviousPage 2 of 12Next →