dofm

16 ms·

361.
by dofm 2mo ago
If there isn't an appropriate metaphor in the Silmarillion he's not going to have thought about it.
362.
by dofm 2mo ago
You now have me testing it with reasoning turned off, which I have never bothered much with on any other local models because it's rarely worth it. The result appears to be almost as good as Qwen 3.6 35B A3B on medium thinking mode. It
363.
by dofm 2mo ago
AFAIK that is initially only for the iPhone?
364.
by dofm 2mo ago
It’s also a little bit snarky, almost. The stuff it thought during the car wash puzzle made me laugh.
365.
by dofm 2mo ago
Yes. Though the chat template doesn’t tell LM Studio to offer the little dropdown. You can bodge the template in the load parameters. Unsloth Studio / Desktop has it working really well with their version of the weights.
366.
by dofm 2mo ago
Unfortunately in xhigh reasoning effort it will burn through 2K tokens before it has even finished its bullet point overview. It really is intense and obsessive. You might need ten times more! Your strategy would likely help in medium reaso
367.
by dofm 2mo ago
> The reason thinking seems to work is that it's just trying to find an optimum outside the local optimum, and the thinking trace helps find it. Yes, I think I finally have an intuitive sense for that. But surely on a longer prompt
368.
by dofm 2mo ago
Unfortunately in xhigh thinking it goes down rabbit holes in such an extreme depth-first way, that whenever you choose to cut it off, there is a very good chance it will not have got round to musing on even half of the prompt! It doesn’t re
369.
by dofm 2mo ago
I mean creativity there in the LLM sense (temperature driving more creative solutions), not the human sense, and in the context of its consistency, configurability and being amenable to analysis. The watermarking approach makes that non-rep
370.
by dofm 2mo ago
His complaint about the ChatGPT app is that it’s a shitty non-Mac-ish Mac app. Complaining about shitty non-Mac-ish Mac apps to people who hate shitty non-Mac-ish Mac apps is more or less how he became a full time writer.
371.
by dofm 2mo ago
> This is some audiophile thinking that because your RNG is “pure”, you get text with an expansive soundstage or whatever. You are projecting that onto me, and I cannot tell you how comically poorly aimed it is.
372.
by dofm 2mo ago
I don’t disagree about open weights (though the enabling aspect there is actually open source inference, right?) But it feels to me like you would need a hell of a lot of text to bury even a simple account ID. The nudges they are talking ab
373.
by dofm 2mo ago
It’s fully possible I didn’t explain it very well in the first place, but he is making a wider point. The point I made (quite briefly) is that watermarking is only feasible because for good writing it is necessary to use T>0, or the writ
374.
by dofm 2mo ago
This is not it, no. He is not using AI and it is not I think remotely in his nature to surrender that control. He is engaging with this on principle. Again I am not sure I agree with him, but then it’s a hypothetical because I am not going
375.
by dofm 2mo ago
> Does the author think he is currently getting T=0 output from Claude? Is he under the impression that T=0 produces the "best" writing? No and no. I am not sure I agree with his point but I know he is not ill-informed on eithe
376.
by dofm 2mo ago
One fascinating small indication is the relatively underreported machinations in the last couple of weeks where the USA has been selling billions of its Euro holding, without any co-ordination with the ECB, to buy Yen, so that the Japanese
377.
by dofm 2mo ago
It is that but it's also the opposite of that at the same time — he wants big businesses to own things but he thinks the government should have a cut. It's the state as protection racket. He's argued for it for a long time, n
378.
by dofm 2mo ago
Trump may be untouchable (and he is decaying too fast to really experience what consequences could reach him) but I think someone may go to prison for that scheme. It's just astonishing.
379.
by dofm 2mo ago
And a-pox-on-both-their-houses, vote-suppressing voices like Glenn Greenwald, Bill Maher, and countless others.
380.
by dofm 2mo ago
(Inexpert ramblings follow) Part of the problem of this is likely that the deep meanings of words you might use in chat to describe a business problem or task that you wish to see implemented are essentially inseparable from scenarios in wh
381.
by dofm 2mo ago
It’s very hard to convey! If it wasn’t, they would be better at it. What I am talking about is the sort of neutral non-formal British default writing that any of us, or indeed people on the island of Ireland would feel was likely to have co
382.
by dofm 2mo ago
I don’t think this is working that well. The unslopped content has a whiff of automatic translation. In the same way that it is impossible to get an LLM to sound authentically British.
383.
by dofm 2mo ago
Many of us are old enough to remember “medireview”, which emerged from different automatic replacement reasons.
384.
by dofm 2mo ago
I have a sneaking suspicion that Qwen is fine-tuned on youtuber test cases (like Luke's Dev Lab, where Qwen 3.8 27B has just done almost eerily well). Part of my suspicion is drawn from the thinking trace I got when I tested the car wa
385.
by dofm 2mo ago
Yeah — you definitely get more ticks! I think everyone I know who lives in the Highlands or walks in it has been bitten, and that is one of the Lyme risk zones. But weirdly almost all of the peak risk areas for Lyme (and the encephalitis di
386.
by dofm 2mo ago
> 1986 is forty years ago. Epidemiologically recent (and also dofm-recent because I'm old). But the graphs typically show that Scotland ballooned in the last twenty years, from below a hundred a year into the thousands. Apparently i
387.
by dofm 2mo ago
> Perhaps the most interesting thing is that RISC-V shows just how ISA agnostic people are To the extent that Raspberry Pi shipped a microcontroller that can literally be either RISC-V or ARM (indeed, one of each at the same time I think
388.
by dofm 2mo ago
People are more aware of ticks because of dog ownership, I think, but I don't know if this is a real risk because of the widespread use of tick repellent treatments for dogs (which are their own little eco-disaster). COVID lockdowns ch
389.
by dofm 2mo ago
Nothing really to do with public healthcare. It is literally that there is almost no medical or social experience in the UK of actually seeing developed Lyme disease up close, because it has historically been so unusual, so there are no pro
390.
by dofm 2mo ago
Not everyone is in a position to take that approach so freely, alas. So one more tool in the toolbox is a very good thing.

More ›