hadlock

6 ms·

31.
by hadlock 21d ago
>I don't get the second brain thing. I don't keep a second brain, I just read and think a lot. Yes, but writing also helps you lens your thoughts in a completely different way. This is why so many people journal, or blog, or vl
32.
by hadlock 21d ago
the important thing is that Qwen 3.7 27B will run unlimited jobs on my consumer grade laptop at 60 tokens/second for free, forever, in about 1-2 years
33.
by hadlock 21d ago
set thinking to minimal and use these jinja templates: https://huggingface.co/froggeric/Qwen-Fixed-Chat-Templates we went from 62% completion to 92% using a claude code harness
34.
by hadlock 22d ago
Most of the erosion is due to oil and gas companies dredging poorly thought out canals to move drilling equipment through the marsh, causing unslowed erosion for ~70+ years.
35.
by hadlock 22d ago
The oil and gas companies brought in huge massive drilling rigs in the 1930s-1960s. The marsh was too soft/soggy to bring in the equipment overland, so they just... dredged huge canals in the soggy marsh a couple feet deep. You can see
36.
by hadlock 23d ago
It turns out you can train a 1b model at almost 1000 tokens/s on a m5 max laptop. As a personal experiment, I've been asking Sol for synthetic training data and synthetic agentic training data (model distillation in it's pure
37.
by hadlock 24d ago
Maybe it's just me, but between the extremely thin font and layout design, I find this extremely difficult to parse. Overuse and improper use of italics is confusing as well.
38.
by hadlock 25d ago
I guess you haven't tried ratatui then. All my cli offer a ratatui dashboard mode, and about 50% of the dashboard mode end up as a status dashboard down the road
39.
by hadlock 25d ago
MacOS: what is my purpose Me: you exist so I can launch iTerm2, Chrome, and VS Code MacOS: oh, my god I can't ever imagine using walled garden graphics api in 2026. Particularly for work tooling
40.
by hadlock 26d ago
I suspect due to model distillation, their need to stay on top for IPO value is more important than absolutely crushing their competition, who will simply harvest their output and distill it for training data on a ~3 month delay. Also they
41.
by hadlock 28d ago
When the market settles down, most of the money is going to be in bulk inference, for batch jobs. In 5 years everyone's laptop will be able to run Qwen 3.8 27B for coding tasks, but businesses will still need to run inference 24/7
42.
by hadlock 28d ago
I was surprised when I went for an extended visit to Colombia - outside of beef cuts (an entire rack of ribs, bit butcher cuts, not a single steak or a dozen hamburger patties) frozen foods are virtually non-existant. If you wander into any
43.
by hadlock 29d ago
OpenAI will have pretty good future free cash flow...until GDDR7 prices are no longer inflated by scarcity. Which could be 24-72 months.
44.
by hadlock 29d ago
Prior to 1970? Almost all of them. Or enough to be statistically significant.
45.
by hadlock 29d ago
>They don't perform as well as winter tires in the winter, and A lot of people don't live in climates where winter tires are even sold; particularly California, Texas, Florida, three of the most populous states. Many people in
46.
by hadlock 29d ago
Two of my cars wear specialty tires, one is a citroen that uses specialty Michelin Pilot-X tires designed exclusively for Citroen (and a handful of Renault) of that era, the other wears very limited production tires designed to go on spoke
47.
by hadlock 29d ago
>Register is not a serious publication Do you have a source to back this up? The register has been covering IT since before most people here were born, and literally invented BOFH.
48.
by hadlock 1mo ago
NIN and iD (re-?)released a vinyl record set for the 30th anniversary a couple weeks ago.
49.
by hadlock 1mo ago
I benched Qwen 3.6 35B-A3B against Qwen 3.8 27B with the same parameters, thinking set to low. Despite 35B having 9x fewer active parameters, it benched only 2.34x slower. The 35B got only 50% more agentic tasks done per hour.
50.
by hadlock 1mo ago
Strangely Qwen 3.8 Max isn't on their list, at all.
51.
by hadlock 1mo ago
I didn't have time to run this as well, but we're getting a 99% agent completion rate across all tasks and 98% task decision that matches the human selected option(s) on Qwen 3.8 27B @ NVFP4. There may be a difference between FP8
52.
by hadlock 1mo ago
Thanks this bumped my agent success rate from 67% to 92.5% (!!!)
53.
by hadlock 1mo ago
Turning off thinking mode you need a different tokenizer and a whole host of settings. It's more like switching your vehicle from gas to diesel, than turning your headlights from lo to hi.
54.
by hadlock 1mo ago
For high concurrency, using the blackwell's native native W4A4 MLP compute path, nvfp4 is something like a 1.2-1.5x performance increase over FP8. We're doing data enrichment (so, tasks completed successfully + tokens/second)
55.
by hadlock 1mo ago
This is the version we'll be testing on our rtx 6000 today! Thank you
56.
by hadlock 1mo ago
At 14,000t/s that's effectively a motor cortex for an android, you no longer need to train the robot to walk, it has a general idea for how to walk (baked into the 1b model), and then just corrects based on sensor input, in real t
57.
by hadlock 1mo ago
Much like spaceflight, aerospace or nuclear engineering, you need to retain local talent for national defense purposes. Being 70% as good is still way better than being 100% dependent and heavily leveraged by your opponents.
58.
by hadlock 1mo ago
I'd love to have been a fly on the wall of the Waymo offices when this news broke
59.
by hadlock 1mo ago
It's a forum, not an engineering conference, but here is a demonstrated 10 point difference between two top harnesses anyways: https://artificialanalysis.ai/agents/coding-agents#harness-c...
60.
by hadlock 1mo ago
When it comes to quality of outcome, since at least Feburary, the harness has almost equal, if not more weight than the model itself. It's no longer "which model is the best?" it's "which model + harness is the best

More ›