jjcm

8 ms·

121.
by jjcm 3mo ago
Cloudflare is obviously more trustworthy/robust here, but if name of the url matters to you, my site non.io [1] allows for named uploads, ie https://html.non.io/solara [2] Somewhat useful if you want a url that isn&#x2
122.
by jjcm 3mo ago
Depends on the domain imo. I work on a design tool - I don't think their political narrative will affect my work.
123.
by jjcm 3mo ago
This feels like a regulation whose effectiveness will expire in the next couple of years (as driverless cars become the norm), but which will set a precedent that this is the norm. This with the EU chat control coming up really set a tone.
124.
by jjcm 3mo ago
Just to add more context to this - he accepted ~$10 million a year while managing a city that has a population larger than New York City. The GDP of the city was 278.9 billion USD in 2025.
125.
by jjcm 3mo ago
These "pull the loop over" kinda knots are delightfully simple, and shockingly secure. The Palomar fishing knot is somewhat similar, and is one of my go-to ones when I need to tie a hook quickly: https://www.youtube.com
126.
by jjcm 3mo ago
A non-trivial amount of their ARR is still from Valve-made games. Counterstrike still nets a bit over 1b per year from just case unboxings, and Dota is in the hundreds of millions. I wouldn't call 8-10% a margin of error.
127.
by jjcm 3mo ago
What's fascinating to me are the Valve comparables here. <500 employees vs 18k at Xbox 17B ARR vs 20B ARR At the end of the day there are two strong differences here. Valve has always been lead by people who were game devs, and have
128.
by jjcm 4mo ago
If you want the core loop of Factorio but with a fresh spin, I highly recommend Dyson Sphere Program. It's my personal fav of the factory genre for the pure scale of it. As a tip, there's full multiplayer support from the mod comm
129.
by jjcm 4mo ago
I'd also look at the qwopus distil if you're using qwen 3.6 27b. It's a nice refinement of the current 27b with slightly better stats. Jackrong has a few different ones available depending on what you're trying to do: h
130.
by jjcm 4mo ago
Neat. I tested it and here were my results: ffmpeg conversion: 1,515 bytes, 0.05s zgif conversion: 1,479 bytes, 90.1s I used cursor to port it to rust as well, and it got the conversion time down to 20s. Still likely not worth it as even wi
131.
by jjcm 4mo ago
IMO the most interesting thing about this is Kimi K2.6, an extremely capable model, can be relatively easily post-trained to allow pen tests. This in its own right proves that the defenses of Fable and others are temporary blocks, and AI ba
132.
by jjcm 4mo ago
I think what I find fascinating about this is it's a native app with no web version... and they still decided to write it in html/js. This is after Microsoft's commitment to rebuild things in WinUI. Don't get me wrong, I
133.
by jjcm 4mo ago
What is the current state of ZFS? I know it had some licensing issues traditionally, despite it being a delight to use every time I've tried it. Is it back?
134.
by jjcm 4mo ago
If you want something price-competitive with composer 2.5, deepseek pro is very cost-effective. Just rig it up in opencode via openrouter.
135.
by jjcm 4mo ago
RE longer prompts, yes. Generally speaking I expand most prompts to be around a full page of text as it is already, so adding more detail in just refines that expanded prompt. That's more for a single screen though. It sounds like what
136.
by jjcm 4mo ago
I have a "brands" system that's exactly for this: https://image.non.io/27e099c5-f0d7-493a-a01f-928d9e42cef5.we... Once you have a few reference screens, you can generate a brand guidelines image, which is a v
137.
by jjcm 4mo ago
I'll share my results / my approach. Here are three designs from the prompt->design thing I'm working on: https://image.non.io/10037610-e35e-44b0-b5c6-69d8fb772109.we... https://image.non.io&#x2
138.
by jjcm 4mo ago
What constitutes a correct answer though? Is something like, "People online say that x y and z because a b c" a credible, correct answer, even if it isn't because of a/b/c?
139.
by jjcm 4mo ago
This is extremely common, unfortunately, to a point where it's a known/expected outcome when you're first creating a brand or product page among those in the biz. If this doesn't work, I'd encourage you to reach o
140.
by jjcm 4mo ago
> Setup a Blog / Static site generator (Pelican), create a simple but stylish theme RE this one, I highly recommend doing image->code as the flow here. Codex's sites feature is doing this under the hood - it's renderi
141.
by jjcm 4mo ago
I think we might be looking at it the wrong way. An individual chat with an LLM is not consciousness, but the entire model itself, over time, might be. Everyone has a different definition of consciousness, but in my mind memory and the abil
142.
by jjcm 5mo ago
Not to undercut the open source nature of this, but what makes this "beautiful"? From a design standpoint, it's basic tailwind. Neutral grey tailwind at that, using Lucide icons. There's nothing wrong with these, but it&
143.
by jjcm 5mo ago
For what it's worth, overall I have a very strong net positive opinion of my time overall at Figma. It was hard, and I spent too much time working, but half of that was probably my fault as well. It's a great company, and I have g
144.
by jjcm 5mo ago
At the time I was a product manager looking after some of the more technically complex features (components, variables, publishing). They were worried that Extended Collections (multi-brand theming essentially) would be delayed further, whi
145.
by jjcm 5mo ago
That's part of the reason why I quit :)
146.
by jjcm 5mo ago
Not quite as I understand it. The ternary approach bonsai uses leverages a FP16 scaling factor that each value in the ternary maps to. You're still using 16 bit multiplication, it's just that the weights are far more compressed.
147.
by jjcm 5mo ago
Decompiling a binary and recreating the source, doing a full line-by-line security audit, always-on agents monitoring state minute-by-minute, etc. I would very easily find ways to hit that level of token usage if it was cheaper/faster.
148.
by jjcm 5mo ago
I did an estimate of that if you're interested: https://x.com/pwnies/status/2028831699736637912 The TL;DR though is that a 10-15b param model baked into an ASIC with the latest fab tech would take around 62W
149.
by jjcm 5mo ago
For an exec, it is, and he is 100% correct. There's a reason why execs get perks - free housing, transportation, personal assistants, etc are all common things at the VP and up level for these large corportations. The whole reason for
150.
by jjcm 5mo ago
I strongly disagree with it not being required. I run a small social news site - AV1 is still prohibatively expensive both for the server and clients for software encoding/decoding. Without hardware encoding, the tradeoff for better co

More ›