nacs

4 ms·

1.
by nacs 5d ago
Apple product won't be the cheapest but it is a full package (CPU, RAM, VRAM/GPU, fast-storage, etc). If you look at the pricing of a full (x86) AI workstation you'd need around the nvidia GPU, you'd approach $10k easily
2.
by nacs 5d ago
Good to know thanks. That's with CPU offload to a DDR5 6000 RAM though which is around $3-4k at least.
3.
by nacs 5d ago
5090 has 32GB VRAM. It'd be silly to buy the 18k model to run a tiny model like Qwen 27B. You use models like GLM Flash and Qwen Next which won't fit on a single 5090.
4.
by nacs 5d ago
People don't buy Sparks and M5 Ultras to run a 27B model - you buy it to run an MoE model like Qwen Next which this M5 excelled at.
5.
by nacs 5d ago
That's a dense model. Of course it will do worse. Now try running that Qwen 3.8 Next model on the 5090 and tell me what TPS you get (hint: it's near 0 since it doesnt fit the 32GB VRAM on 5090 vs the 256 in OPs M5).
6.
by nacs 7d ago
It was yes, lots of words in the input to describe state of the game-world at each checkpoint.
7.
by nacs 7d ago
Easily. It took around 4GB VRAM on my nvidia card so a 48GB unified mem should have no issue. < 1s inference
8.
by nacs 8d ago
Tailscale funnel works well but you can only have 1 per Tailscale instance (1 per machine) at the https endpoint so if you want to have a few apps, the others will go on non-https, custom-port URLs.
9.
by nacs 15d ago
Intended to be publicly accessible internet content, I have no problem with. But are you ignoring the literal scanning (and 'burning down' of books) that Anthropic has been found guilty of? Or the torrenting of pirated content en-
10.
by nacs 15d ago
Claude/OpenAI etc have taken and continue to take literally all data from the internet, printed books, image, audio, and video humans have created in all of existence without permission to train their models. But when same AI company g
11.
by nacs 15d ago
Openrouter lets you pin or blacklist providers or specify provider per-request as well.
12.
by nacs 15d ago
Did you miss the part of this article where the benchmarks specifically call out Fireworks as one of the worst in their tests: > Fireworks scored 46% on TAU, a 30 point gap Another surprise was DigitalOcean being bottom of barrel too. Co
13.
by nacs 17d ago
And of course it just renders an HTML document (applies some CSS to the text content) and then uses the embedded browser's printing.
14.
by nacs 27d ago
I gave it another go with the hint above about the tetherd log: tail -f ~/.local/state/tether/tetherd.log I killed that persistent process (sudo killall), rm /run/user/1000/tether/tetherd.
15.
by nacs 27d ago
I ran into similar issues on a mostly stock Ubuntu 24. When I start the gtk app, it has a scan devices button but it never finds my phone even though I've paired via bluetooth and its on same LAN. The iOS app just has a spinning icon t
16.
by nacs 1mo ago
I thought so too and scrolled the page looking for a last name. Then realized the ending was not the same. RIP
17.
by nacs 1mo ago
What's the point in having it use an offline LLM if it still requires an internet connection to send all your chat messages through the "Groupmind" paid service? Seems like the real purpose of this project was to plug Groupmi
18.
by nacs 2mo ago
He had thinking mode set to "low" fortunately.
19.
by nacs 2mo ago
It's looks like they literally cloned the UI and graphics from the Mini Metro/Mini Motorways games (not just the map style).
20.
by nacs 2mo ago
The vast majority of companies are still putting most things in the cloud and will continue to do so and this far outnumbers the must-be-on-premise companies. Sure there will be self-hosters but hosting AI models will always be more of a ch
21.
by nacs 3mo ago
Reminder that by using Bitlocker, you're using a closed source encryption for which Microsoft will happily hand out your recovery key on request. https://www.forbes.com/sites/thomasbrewster/2026/01/2
22.
by nacs 3mo ago
UBlock Origin is also giving me the warning on that domain.
23.
by nacs 3mo ago
Only the 31b is Gemma. All the rest - including 9B - are Qwen 3.5/3.6: https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B/blob/m...
24.
by nacs 3mo ago
Pi Coding Agent would be great also.
25.
by nacs 4mo ago
Ghostty-web exists and is even API compatible with xterm.js (same engine that powers Ghostty): https://github.com/coder/ghostty-web/ In fact, it looks like wterm's 12KB plugin doesnt offer full term emulation
26.
by nacs 4mo ago
Node-based editors are not new. The Arc homepage is clearly vibe slop but with 75 nodes and backend supposedly coded in C, it looks to me like it would have taken a couple of months to get here at least, certainly not a release within 1 wee
27.
by nacs 4mo ago
Lastpass has had multiple large breaches, especially after LogMeIn bought them out
28.
by nacs 4mo ago
This feels like deja-vu with Lastpass. LogMeIn buys Lastpass, multiple massive breaches occur[, people move to Bitwarden].
29.
by nacs 5mo ago
Just checked since you made me curious. With 1 PHP and 1 nodejs project in 2 windows, here's the usage (on Ubuntu Linux): Zed is using: - CPU: 4.7% - RSS: ~2.1 GB The bulk of it is language servers (TypeScript/tsserv
30.
by nacs 5mo ago
I only installed it today for the first time but yes it does have a very prominent button to completely disable all AI.

More ›