skhameneh

5 ms·

31.
by skhameneh 9mo ago
Also, a less efficient option with less “tech” would be a Silverado WT (work truck), those have a lot of range and sometimes you can find them for good used prices. Those are a hit or miss for deals because the MSRP is on the higher end.
32.
by skhameneh 9mo ago
For cheap, the 2018 and later Nissan Leafs (old tech) or an Ariya (new tech, nothing notably exciting. Maybe also check out Ioniq 5, EV6, Equinox, etc. FWIW, my wife drives a Mach-E and I drive a Fisker Ocean. The Mach-E is very comfortab
33.
by skhameneh 9mo ago
This does read/interpret a bit odd, because the Hummer H2 doesn’t strike me as a reliable vehicle and I’ve generally heard of them to be cost sinks (completely disregarding the horrible efficiency). Why not start off looking at the che
34.
LLM Optimized Engineering Principles(github.com)
1 points by skhameneh 9mo ago|1 comments
35.
by skhameneh 9mo ago
I’m sharing a prompt/template/document created to weight context with engineering principles. The goal is to weight initial context with principles to improve software development when using LLMs.
36.
by skhameneh 9mo ago
I didn't have the patience to click through after visiting a few pages only to find the depth lacking. About an hour ago or so I used Opus 4.5 to give me a flat list with summaries. I tried to post it here as a comment but it was too
37.
Recursive Language Models: the paradigm of 2026(primeintellect.ai)
5 points by skhameneh 9mo ago|1 comments
38.
by skhameneh 10mo ago
> I can't stop any megacorp from training on my code, and it's futile to try. I do not like this take and I do hope you reconsider repeating such. This very much reads as accepting a lack of any claim to reasonable privacy an
39.
by skhameneh 10mo ago
I have no doubt your intentions were good and I am a fan of open-source myself. Unfortunately, since this was written, it’s become more widely known and commonplace for large corporations to disproportionally benefit from open source. And
40.
by skhameneh 10mo ago
ik_llama is almost always faster when tuned. However, when untuned I've found them to be very similar in performance with varied results as to which will perform better. But vLLM and Sglang tend to be faster than both of those.
41.
by skhameneh 10mo ago
There is plenty that can run within 32/64/96gb VRAM. IMO models like Phi-4 are underrated for many simple tasks. Some quantized Gemma 3 are quite good as well. There are larger/better models as well, but those tend to rea
42.
by skhameneh 10mo ago
In theory, it’s only sufficient for pipeline parallel due to limited lanes and interconnect bandwidth. Generally, scalability on consumer GPUs falls off between 4-8 GPUs for most. Those running more GPUs are typically using a higher quant
43.
by skhameneh 11mo ago
At first glance, it appears to have lower negative pressure and a much higher filter surface area than most air purifiers. AFAIK PC fans are designed to move air at low pressure while quality air purifiers are designed to handle more pressu
44.
by skhameneh 11mo ago
> i think people that were promoted for building systems that turned out bad, should be demoted Nope, in the same vein of "lording" over others, they become the expert of knowledge of bullshit. The environments that allow suc
45.
by skhameneh 11mo ago
I'm convinced it'll be in the form of emergence and it won't immediately be known when it's started. It may very well already be here, but the feedback loops are excruciatingly long and expensive.
46.
Collection of LLMs that run well in 32gb VRAM(huggingface.co)
3 points by skhameneh 11mo ago|0 comments
47.
by skhameneh 11mo ago
There’s been community commentary that many of the GPT models are a tad overfitted WRT benchmarks. Benchmarks are not representative of end user experiences. That’s not to say the benchmarks aren’t useful at all, but are only useful as a
48.
by skhameneh 11mo ago
STM32 boards and PlatformIO. ESP32 is quite popular (as seen by other suggestions) but I find the quality of Espressif, hardware/software/support, is widely varied. FWIW PlatformIO works with Arduino and ESP32 (and will give you a
49.
by skhameneh 11mo ago
Miele... Expensive, works great (when it does), but absolute shit customer service in the US. One of my largest local retailers cut back on Miele in general, they had a high return rate. Miele makes a number of great products, but their cu
50.
by skhameneh 11mo ago
I've been a proponent of upstreaming fixes for open source software. Why? - It makes continued downstream consumption easier, you don't have to rely on fragile secret patches. - It gives back to projects that helped you to b
51.
by skhameneh 11mo ago
The very first thing I saw from Apple that, IMO, Jobs would have vehemently stopped was the two-toned back on the iPhone 5. That said, the iOS 26 release is abysmal. The only redeeming thing for me has been the enhancements to Stage Manage
52.
by skhameneh 11mo ago
I've been under the impression most inference engines aren't fully deterministic with a temperature of 0 as some of the initial seed values can vary. Note: I haven't tested this nor have I played with seed values. IIRC the i
53.
by skhameneh 1y ago
That's still very limiting when comparing to commercial models. To be truly competitive with commercial offerings the bar is closer to 4-8x that for one node . That said, maybe a quantized version of GLM 4.5 Air, but if we're talk
54.
by skhameneh 1y ago
For tools... I find Aider to be the most reliable and least "invasive" (from following instructions and nothing more standpoint), but it's also the most hands-on. Cline is a nice step up in hand holding, but does require a
55.
by skhameneh 1y ago
XAMPP seems to still be alive and maintained. https://www.apachefriends.org/ I haven't used it in over a decade, but I'm glad to see it's still kicking.
56.
by skhameneh 1y ago
12 memory channels per CPU and DDR5-6400 may be supported (for reference, I found incorrect specs when I was looking at Epyc CPU retail listings some weeks ago), see https://www.amd.com/en/products/processors/
57.
by skhameneh 1y ago
> ChatGPT has largely replaced Google in my everyday use This. Organically replacing a search engine (almost) entirely is a massive change. Applied LLM use cases seemingly popped up in every corner within a very short timespan. Some cha
58.
by skhameneh 1y ago
And consists of 8 GPUs
59.
by skhameneh 1y ago
I wouldn't be surprised if power consumption is a starting point due to things like permitting and initial load planning. I imagine this as a subtractive process starting with the maximum energy window.
60.
by skhameneh 1y ago
Before reading your comment I did some napkin math using 600W per GPU: 10,000,000,000 / 600 = 16,666,666.66... With varying consumption/TDP, could be significantly more, could be significantly less, but at least it gives a startin

More ›