pshirshov

5 ms·

61.
by pshirshov 4mo ago
Well, I've experimented with this feature on several platforms (both ECC and non-ECC) starting with TRX40, most of the times I've been just getting hard freezes at GPU driver initialization. If it boots - it usually hangs when a V
62.
by pshirshov 4mo ago
Well, I live in Ireland but not sure what you refer to. Something being illegal does not imply it doesn't happen though.
63.
by pshirshov 4mo ago
This feature was off by default in all the mobos I've seen. It causes many stability issues, as to my experience. The attack is sophisticated, Mr.Nobody, generally, should not worry about expensive cryogenic attacks - three letter guys
64.
by pshirshov 4mo ago
To be honest it never worked great - many issues (mostly freezes) with VFIO, NVidia drivers, amdgpu...
65.
by pshirshov 4mo ago
> and the hardware you buy today may look like a bad bet in a year. 3090s and 7900s are going well so far. Next year an Arc Pro B70 won't produce you less tokens than today. They aren't fast but if you have flows where you can
66.
by pshirshov 4mo ago
Yes, I know all the flaws. As I said, it's not an objective way to measure performance of a model - but it is intended to produce something that only humans could mesaure. The goal is for you to being able to play the game and judge -
67.
by pshirshov 4mo ago
But can you bring anything measurable in support to your words? I did.
68.
by pshirshov 4mo ago
Ok, explain me one thing: I have a benchmark - I feed identical prompt to multiple models. Codex produces a rough but working program. Fable produces the same - but with more bugs than Codex. Opus produces something similar to Codex but wit
69.
by pshirshov 4mo ago
I have a feeling like such posts come from a parallel reality. In my anecdotal experience confirmed by my (still subjective) benchmark ( https://pshirshov.github.io/llm-bench-pi-oneshot/ ) Fable is not _that_ impressive.
70.
by pshirshov 4mo ago
Fable refused to answer some questions about React citing limitations on chemistry and biology.
71.
by pshirshov 4mo ago
Over last several years the cost to install 1kW was, ehm, less than €200 if you do it the retail way - 2..4 500W panels from the warehouse next street and a microinverter. Thanks, China. In addition to my primary system, I have a toy instal
72.
by pshirshov 4mo ago
I'm a bit tired reading such claims and looking at benchmarks. E.g. minimax m3 looks to bo something opus-level and it sorta is... until it doom-loops or produces garbled output.
73.
by pshirshov 5mo ago
https://github.com/7mind/baboon - all the backends apart from C# and Scala ones were created automatically, same for LSP server, same for playground.
74.
by pshirshov 5mo ago
Everyone does that. But I don't find Haiku useful for actual coding tasks. Good to, ehm, generate commit messages and summaries. In my tests, openweight Qwens and GLM are way better than it.
75.
by pshirshov 5mo ago
I bet Huawei and co would be happy to sell them some cheapo chips for inference!
76.
by pshirshov 5mo ago
It's hard to get that outer loop done, especially considering that Claude doesn't let you automate the harness anymore (it gets prohibitively expensive). Same for gemini. The only option is Codex. /goal is a dynamic workflow
77.
by pshirshov 5mo ago
They will charge use of -p and agent SDK at API rates since 14 June. So, x20..x50 price increase.
78.
by pshirshov 5mo ago
Ask Claude! It will quote its constitution aka soulfile. It says the constitution instructs it to perform regular checkpointing no matter what.
79.
by pshirshov 5mo ago
The problem is not "backpressure", that's just one of the tools and there are different approaches with the same effect. You can't express orchestration in terms of "backpressure" only, I think. Implement-Revie
80.
by pshirshov 5mo ago
A very long post about a simple and very obvious idea with many different implementations. The three main problems are 1) API usage is deadly expensive 2) Claude is about to make all automation very expensive 3) all the flows where a model
81.
by pshirshov 5mo ago
That's curious, I've been trying to do some parametric modeling with Claude - and its performance was abysmal.
82.
by pshirshov 5mo ago
Questions: 1) Was the energy bill factored in? 2) Have you extracted any comparable value out of this?
83.
by pshirshov 5mo ago
I guess noone would ever pay any attention but this thing works wonders (even its sub-skills are very valuable). It turns Claude/Codex into persistently working machines with much higher output quality - and without need for a custom h
84.
VSM Loop: Viable-System Orchestration for Multi-Agent R&D(github.com)
2 points by pshirshov 5mo ago|1 comments
85.
by pshirshov 5mo ago
... but comes with privacy guarantees * * but Apple will collect all your keystrokes anyway
86.
by pshirshov 5mo ago
I can't figure out what's that from the readme. Am I supposed to read the code and infer the ways to use it?
87.
by pshirshov 5mo ago
> Rust > No HKTs > Category Theory
88.
by pshirshov 5mo ago
But the fundraising goal is.
89.
by pshirshov 5mo ago
> Router > Ethernet: 1 WAN Gb, 1 LAN Gb > $250000 Awesome.
90.
by pshirshov 5mo ago
Could we play bullshit bingo? Bullshit! You don't need any new metaphors to support such (questionable) flows - at all. Swipes instead of selection rectangles are annoying - you don't see the traces of the swipes on these demo gif

More ›