dofm

17 ms·

331.
by dofm 2mo ago
> Benchmarks are the BMI of model evaluation. That is such an elegant way to put it.
332.
by dofm 2mo ago
Quick tests suggest it’s pretty good at reasoning and tool use (keen to search to check its thinking) and it seems to waste much less time thinking, too. So it feels very fast. But it does not seem to be better than Qwen 3.6 35B at coding.
333.
by dofm 2mo ago
And Denmark should buy Anthropic. “Invade Greenland and we switch it off!”
334.
by dofm 2mo ago
I had assumed it involved a dartboard.
335.
by dofm 2mo ago
So your dealer has cut your drugs, isn't always available when he says he's going to be, and the price is going up?
336.
by dofm 2mo ago
High end porn can just be expensive production values; that is what I assumed. She wanted a proper studio and production company, and the implication is she was positioning it in opposition to the handful of studios that dominate that indus
337.
by dofm 2mo ago
I am fully unbothered about Amodei's wife trying to make high end porn for women, in the same way that I am fully unbothered by Melania Trump having been essentially a glamour/nude model. I know (and have creatively worked) with w
338.
by dofm 2mo ago
One of the more interesting corners of history is the Palm guys settling on honesty as a reaction to that. You make a system that can do enough that you can teach users to meet it in the middle, and you don't pretend it is magic. PalmO
339.
by dofm 2mo ago
The display was really impressive for the era, resistive touch, backlit etc.; this is two years before the Palm Pilot launched. Most people would not have thought "underwhelming", and displays no better than that were commonplace
340.
by dofm 2mo ago
Coupled with the other comment of yours that was flagged, the tone of this is completely unacceptable.
341.
by dofm 2mo ago
Right; the same mechanisms (curation, demand-driven economies of scale, library user demand) that largely stop niche self-published books from ever finding themselves in libraries (well, nationally in public libraries at least; they are oft
342.
by dofm 2mo ago
Some nuances here. There’s a variety of different roadblocks at different levels that will work to prevent pure AI slop books ever turning up in a typical public library of physical printed works; I think they never will. Probably a handf
343.
by dofm 2mo ago
It's already very keen not to overthink. Thinks about it all the time.
344.
by dofm 2mo ago
Trump just threatened to bomb Oman. Oman! It’s not possible to make any predictions about the petrodollar in this context. We know what should happen. But it is like trying to predict earthquakes.
345.
by dofm 2mo ago
I hate that I am no longer certain I don’t need to buy a new machine just in case.
346.
by dofm 2mo ago
Thanks for the reply. Sooo much I have to learn.
347.
by dofm 2mo ago
You've somewhat mischaracterised what I have said. I suggested he as a tech writer has reason to be invested in understanding how they work. It's not really a sustainable position to not understand, is it?
348.
by dofm 2mo ago
Ahh — that was in the context of a suggestion AI-slop-writing I was replying to (quite an accusation for an established blogger IMO). But one of the issues with HN threads is that you can sometimes lose the sense of what you're replyin
349.
by dofm 2mo ago
Yes — I just found out that you can set reasoning level in the prompt — like with Qwen 3.8 27B it is actually really pretty solid at "Reasoning level: low". Ten to thirteen tokens per second on my M1 Max (might be some room to imp
350.
by dofm 2mo ago
> I don't think it's weird to ask that people commenting on x doing y with tech z at least try tech z, no? On what specific basis do you assume he hasn't tried it? He's definitely blogged about the desktop apps, after
351.
by dofm 2mo ago
I still haven't tested MTPLX and the memory of reading about it had drifted out of my brain like so many things have, so thanks! I really must.
352.
by dofm 2mo ago
The Strix Halo / Ryzen AI situation also seems to have improved incrementally recently, both in hardware and software.
353.
by dofm 2mo ago
> One issue I have with Gemma is that they seem to use old architectures that rely on full attention, requiring a lot of RAM for context and quickly degrading speeds as context is filled. Yes, this is something I hope they will change. G
354.
by dofm 2mo ago
Unless, of course, your goal is to actually understand what is going on, regardless of whether that is difficult.
355.
by dofm 2mo ago
> It's only 7GB including the mmproj or 8GB with MTP. Even more impressively it doesn't have a separate mmproj at all — it is fully integrated, and the vision encoder doesn't speak words into the LLM, as it were —- it is d
356.
by dofm 2mo ago
> How have you or anyone been influencing or analyzing the randomness behind the sampling process to create better writing? You're mischaracterising or misunderstanding my point, or I mangled it. I mean it is possible to analyse, c
357.
by dofm 2mo ago
Why? I really don't understand this. Why can't a tech writer take a deep but neutral intellectual interest in something? Isn't it important that some do? Do people have to be stakeholders or clearly on one given team, pro- or
358.
by dofm 2mo ago
Thanks for the reply. It is sort of frustrating and unnerving to me how much of this is dependent on application and usage, but it is occasionally fascinating.
359.
by dofm 2mo ago
Hm — brain jumped tracks a bit there at nearly 4am. I'm talking about budget — I mean limiting it to 2048 tokens. … for one or other of the models I tested at the same time, in llama-server, there is a dropdown that offered options (un
360.
by dofm 2mo ago
> I heard my 12yo uses "AI" as a synonym for crap in a context which had nothing to do with machine learning. The kids are alright.

More ›