tristor

7 ms·

1.
by tristor 5d ago
> Treating AI as if the output is factual is just a misuse of the tools, I think. Especially because it's very often wrong. AI tools lack the necessary context of the specific situation and the surrounding technology environment to
2.
by tristor 6d ago
> And, of course, if you're simultaneously paying for Netflix, Disney+, Hulu, HBO Max, Apple TV+, Paramount+, Peacock, YouTube Premium, and Spotify every month, then you're either (a) really into consuming this kind of content
3.
by tristor 6d ago
Well, they basically are. But that's not alone a good enough argument against SDVs.
4.
by tristor 7d ago
> It is quite absurd (but categorically true) that the quality of audio that folks were generally listening to in 2005 was better than what we are doing today. I don't believe this statement is true, as someone who cares about audio
5.
by tristor 8d ago
As someone who has worked in this space longer than a lot of people currently in industry have been alive, VMware predated KVM by a number of years. VMware Server was released in 2001... KVM was announced in 2006 and entered the Linux kerne
6.
by tristor 8d ago
Every time I see this posted, I am tempted to do something similar but then I remember all the friends I used to have for LAN parties have moved different places across the country, so it'd get used at most once a year (we still have a
7.
by tristor 9d ago
That's because the folks who are unemployed, unfortunately, are largely those who already were further down the curve. We're destroying the minds of those who were supposed to move society forward and gaining nothing elsewhere.
8.
by tristor 9d ago
> It's like: everybody is just mentally exhausted, mortal. Very much this. I think a lot of it is that the "slack" of our lives is being eliminated little bit by little bit to improve "productivity", and in so d
9.
by tristor 14d ago
> Back when turbochargers became common on passenger cars, there was even advice to be sure you let the engine idle for a minute or two before shutting it off, so that the turbocharger could cool. Otherwise the oil in the turbocharger wo
10.
by tristor 15d ago
Yes, Q3 = 3bit quant, so that's not comparable to a 4-bit quant.
11.
by tristor 16d ago
I assume you mean Qwen 3.8-27B? Yes, you can run this in 32GB of RAM, but it's very context limited. With KV cache compression and other techniques, it's better now than in the past, but I'd still want more RAM, personally.
12.
by tristor 16d ago
Unfortunately, while cybersecurity is a real thing with real concerns and there are some amount of highly competent people, it's mostly, as a field, full of box-checkers and grifters. That lack of nuance or even attempt to understand
13.
by tristor 16d ago
I've been experimenting with local models on an M5 Max MBP w/ 128GB of RAM since March of this year. Generally I've had very good results. Where things were lacking initially was with tool calling and the need to rely on to
14.
by tristor 16d ago
> If I had to pick a product, I'd say an affordable 32GB mac would be the sweet spot for running local models that function well like Qwen 3.8. 32GB is not enough RAM. I don't even own a device with less than 36GB at this point
15.
by tristor 20d ago
This is a really stupid decision. As someone with mixed feelings about our current immigration situation (which feels at least somewhat out of hand), it's also /very/ clear to me that immigrants in highly-skilled roles have
16.
by tristor 20d ago
> Why would a person who is supposedly having rare skills come to a country and spend their best years here without reasonable guarantee of permanency? For the same reason I temporarily resided elsewhere: An opportunity to expand my hori
17.
by tristor 20d ago
> The guy at my company has a house, wife, and kids in the US; I agree that this is cruel, but I also think that many people coming to the US on an H1-B visa aren't taking to heart the very clear terms of that visa. An H1-B is a &#
18.
by tristor 21d ago
Writing does help you think... but only if you write. Dumping a bunch of stuff into a "second brain" is not you writing. Keeping a journal is you writing. Those serve similar functions, but keeping a journal is more intentional
19.
by tristor 21d ago
Looks like it errors out in LM Studio using the Unsloth quants, apparently the Unsloth team has already posted patches for llama.cpp to support this.
20.
by tristor 22d ago
> There is already at least one company out there that showed its possible so others will follow IMO. Are your referring to Taalas / chatjimmy ? I definitely think we'll see an ASIC-like approach in the future, especially for e
21.
by tristor 22d ago
I considered doing this or buying a house that already had an ADU/casita when we were house shopping, but I have a very different situation with being mostly home alone during the day and a dedicated room in the house for an office. I
22.
by tristor 22d ago
RDMA is buggy and Thunderbolt only delivers 1/10th the throughput of native connectivity. 1TB of Unified Memory w/ 1.2TB/s of bandwidth with marginally ~$30k cost is a different story than 1TB of sorta Unified Memory w/
23.
by tristor 22d ago
I would probably spend up to $30k if I could get 1TB of Unified Memory, because it would allow me a guarantee to run pretty much any local model I want, including >1T parameter models with reasonable quants. I wouldn't be surprised
24.
by tristor 22d ago
It's a few different things. I've gone down a hardcore rabbit hole on this because I also find this exceptionally annoying. I now sometimes experience lag when typing into a terminal, where I am typing faster than characters app
25.
by tristor 22d ago
Kinda. Strix Halo does 256GB/s of memory bandwidth, and is significantly slower than an M5 Max (614GB/s). Feels like intentional market segmentation?
26.
by tristor 22d ago
I wish they were offering 1TB of Unified Memory for the M5 Ultra. I already have an M5 Max MBP w/ 128GB of RAM for running local models, and while there's a /few/ models that I can run in 512GB that I can't run in
27.
by tristor 23d ago
Nice. I hope we'll see more support for jxl across the web. It's so good at this point that it's my primary output/archival format for all of my photography.
28.
by tristor 26d ago
I found that Claude / Sol were basically useless when approaching various CTFs that allowed AI tools, but running Qwen and DeepSeek locally and GLM via OpenRouter worked fine. It's blatantly obvious to anyone that even does tange
29.
by tristor 26d ago
At the risk of sounding arrogant, I will out myself as being incredibly smart and also incredibly unhappy in a general sense (although content with my life). The answer is very simple and mundane: I observe and understand problems that ha
30.
by tristor 29d ago
I've had some rough moments with my Framework 13, but overall I appreciate it because when something like this happens you have a way out other than binning the entire laptop. I agree with the author that Framework should have a bette

More ›