SwellJoe

11 ms·

151.
by SwellJoe 2mo ago
"used to" in your sentence already answers your curiosity. ROCm runs most things just fine without any ceremony or difficulty. I have a couple of the same cards as covered in the article, bought before they got expensive (I'd
152.
by SwellJoe 2mo ago
Pi can only log what the model shows it. Many models keep their thinking traces hidden and only provide a hash or something to recover it on subsequent resumes. DeepSeek shows CoT traces, and is maybe the best model that does so, I think? K
153.
by SwellJoe 2mo ago
Asking them what they think is happening is actually not reliable, though? They don't always know how they came to a conclusion after the fact. It is probable that it's roughly similar to the path they took to get there, since i
154.
by SwellJoe 2mo ago
DeepSeek is more fun, because it's cheap-as-free, Good Enough, quite fast. I use it for all API stuff, automated runs, testing of the security auditing harness and benchmarks I'm working on, etc. Kimi K3 is smarter, though. At lea
155.
by SwellJoe 2mo ago
Reasonix (which has been seemingly the most recommended harness for using DeepSeek, as it is designed around maximizing caching in DeepSeek) is now a Go app, but still installed via npm. Which feels ugly, but I guess everyone has npm alread
156.
by SwellJoe 2mo ago
I hate nodejs and the npm ecosystem more than most, but Java, really?
157.
by SwellJoe 2mo ago
Which are unavailable with the leading American models. You can't look at the complete traces of OpenAI or Anthropic model agents, as they are encrypted (there's been discussion of a couple of different ways to expose those, but t
158.
by SwellJoe 2mo ago
"Every run is traceable Everything the model sees is recorded in an append-only session log: system prompts, reasoning, tool calls and results, subagent scheduling, and every context injection. In the Trajectory view, you can inspect t
159.
by SwellJoe 2mo ago
Linus actually mostly deals only with trusted subsystem leads. They may choose to deal with less trusted contributors, but I wouldn't say the Linux kernel development process is low trust. Many of those folks have worked together for d
160.
by SwellJoe 2mo ago
I guess I have occasionally used screen or tmux to help someone out before, but it was a few minutes, and not a regular occurrence.
161.
by SwellJoe 2mo ago
I stopped picking Fable because it refuses based on guardrails so often. I do a lot of security related work, and Fable just won't do any of it. So, I don't bother. Unfortunately, Opus 5 also refuses quite a bit of security work
162.
by SwellJoe 2mo ago
Counterpoint: I've been using the Unsloth Gemma 4 quants extensively since very soon after release (on ROCm and Apple Silicon, I don't have any Nvidia hardware big enough), pretty much every quantization down to 4 bits (the QAT is
163.
by SwellJoe 2mo ago
Ah, now I hate all my coworkers because I don't want to share a keyboard. Incredible insight.
164.
by SwellJoe 2mo ago
I've been working on a similar-ish set of tools for my old synths and have been pleasantly surprised to find that synth manuals used to include the full SysEx spec for the synth. Most synths/controllers today don't come with
165.
by SwellJoe 2mo ago
I remember the fad. I was around for the Extreme Programming era. But, I've never worked anywhere that pair programming was the norm, nor have I known any developers that wanted to pair program, in the general case. My personal experie
166.
by SwellJoe 2mo ago
Yeah, YC also has their own "multiplayer agent", qm ( https://github.com/yc-software/qm ), though it doesn't seem to be focused on code, which makes a little more sense.
167.
by SwellJoe 2mo ago
I think Zed is an excellent editor (fast!) with a pretty good AI agent built-in, but I have no desire to do multi-player development in my editor. Never have had any such desire. Coding is a single-player game and I can't think of a si
168.
by SwellJoe 2mo ago
Opus 5 doesn't really even speak coherent English. I'm not sure what's going on, but it can't explain anything. It still does an excellent job with code and writing tests and code review and creating and completing a pla
169.
by SwellJoe 2mo ago
That's not how it reads, to me. But, I guess one could generously interpret it that way. However, if it is a scientific experiment, then changing a primary variable in the middle seems like bad thing. I don't actually think it&#x2
170.
by SwellJoe 2mo ago
It still ends up being one or two orders of magnitude cheaper per task on benchmarks.
171.
by SwellJoe 2mo ago
The "safety" guardrails in Anthropic and OpenAI models are becoming a noticeable problem for security work. And, the reason I'm keeping my Kimi subscription even though it's not a great deal; Kimi subscriptions are quite
172.
by SwellJoe 2mo ago
Because an LLM does not get smarter and it does not learn. It is fully cooked when it launches and only gets smarter with more training of some sort, a long, expensive, process that doesn't get phrased as "we thought the model wou
173.
by SwellJoe 2mo ago
As an aside, the bigger S 2.1 runs faster than the dense Qwen 3.6 and Gemma 4 models on the same hardware (assuming the same hardware is big enough to run it) and definitely feels smarter, and more capable of long tasks, but it doesn't
174.
by SwellJoe 2mo ago
Anybody ever tell you that you write like an LLM?
175.
by SwellJoe 2mo ago
Dumping $100,000 into it, and trying to save a few bucks on the model. Penny wise, pound foolish.
176.
by SwellJoe 2mo ago
You're anthropomorphizing the LLMs, just like most writing from non-technical press, including this article, does. LLMs do not function very much like human brains, and when you try to map the pieces of an inference system on to human
177.
by SwellJoe 2mo ago
People just really don't understand that LLMs do not "remember" anything. They have no memory. They can't have memory. The AI companies bolt a database on the side with instructions for the model to query the database fo
178.
by SwellJoe 2mo ago
> I use it to write short sci-fi stories. Life is not only about being an SDE. What you do in the privacy of your own home is none of my business, but I would hope you don't inflict those stories on others.
179.
by SwellJoe 2mo ago
I agree on desktop/laptop, but on mobile there are images that appear under the text making it hard to read.
180.
by SwellJoe 2mo ago
"Stealing" is a strong word to use for looking at the words produced by models built from the collective commons of the world. And, honestly, being able to see how LLMs make decisions is critical to trust and security. I consider

More ›