supermatt

6 ms·

31.
by supermatt 2mo ago
The problem is that some people think all AI-assisted writing is prompting "write me a post about X", so they see LLM prose and assume there has been no thought or effort put into it.
32.
by supermatt 2mo ago
Yes, I did. That was the whole point of the post.
33.
by supermatt 2mo ago
Because some people are just ignorant and downvote anything they either aren't willing or able to understand :)
34.
by supermatt 2mo ago
I submit pretty much the opposite of this, but it didn't gain any traction: https://news.ycombinator.com/item?id=49332291
35.
by supermatt 2mo ago
It is a part of the rust codebase: https://rustc-dev-guide.rust-lang.org/offload/internals.html https://github.com/rust-lang/rust/issues/131513
36.
by supermatt 2mo ago
This is really nice! is it accessible somewhere to play with?
37.
by supermatt 2mo ago
I haven’t tried any larger models, but a 12B model on my Ampere A5000 gets around 4-5x the aggregate throughput with concurrency. I have the maximum context configured to 32k, but my actual requests are usually around 2-4k tokens. No idea h
38.
by supermatt 2mo ago
Can you please try and see how many tokens you get with some form of concurrency. Pretty much ALL the benchmarks I've seen on the more accessible cards are just single request.
39.
The Message Isn't the Medium: In Defense of LLM-Assisted Writing(claude.ai)
2 points by supermatt 2mo ago|2 comments
40.
by supermatt 2mo ago
Indeed. I LOVE reading LLM content. I find that they tend to present things in a way that keeps my attention - something I have struggled with my whole life. I also use LLMs to assist with managing my own thoughts and helping make things co
41.
by supermatt 2mo ago
How can they be so lazy with updating their prompts (or is it just a case of these not being current)? Surely the prompts are an integral part of tuning their offering?: “Above Opus sits Anthropic's new Mythos tier. The first Mythos-cl
42.
by supermatt 2mo ago
> just block the entire countries network range(s) Why not just block all the inbound connections you don't need? Is there a particular reason your firewall policy needs to be xenophobic?
43.
by supermatt 2mo ago
So the concern is that the agent will discover a novel VM escape, exploit it and take control of your whole machine instead of working on its prompted task? That seems rather far fetched.
44.
by supermatt 2mo ago
Why are 3rd party hosted VMs safer than your own?
45.
by supermatt 2mo ago
Definitionally not? "Microblogs, for example on Twitter..." https://dictionary.cambridge.org/dictionary/english/microblo... "Some popular social networks such as X (formerly Twitter), Threads, Tumbl
46.
by supermatt 2mo ago
Honestly, I don't really care that much. The project looks like fun, and I wish you the best with it, but I don't really post on microblogging platforms myself. I just wanted to give a bit of feedback as someone who reads them - a
47.
by supermatt 2mo ago
Sure, but I don't care that much. It is a Show HN, so I was giving feedback.
48.
by supermatt 2mo ago
For perspective: you just commented 404 chars. Your first paragraph alone was 346. My comment on the about page was that it needed more than 280 chars to communicate what the app is about - even the title + intro exceeds the limit. If it is
49.
by supermatt 2mo ago
But clearly not the microbloggers on twitter/x, Bluesky, etc, who use the patterns I mentioned.
50.
by supermatt 2mo ago
A limit that worked well for who? Almost every twitter/x post submitted to HN has people labelling their posts as "X of Y" threads, using images as text, or using pro account posts. You can see it yourself: https:/&#x2F
51.
by supermatt 2mo ago
The 280 char limit seems arbitrary, given you cant even keep the about page within that limit: https://textlog.cc/about If you are going to limit it to short posts, maybe the length of a short post is the length to limit it
52.
by supermatt 2mo ago
I absolutely loved Burn::Cycle on the Philips CDi, where it was the definitive version (apparently the visuals were much better than the PC): https://www.youtube.com/watch?v=zoLw7SEmLYk The gameplay itself was a bit rough,
53.
by supermatt 2mo ago
> NATS control-plane events: stream leader election / R3 quorum re-form during pod churn. Whats is wrong with this? I work with durable tracing (not using NATS, or k8s though) and this looks almost exactly what I would expect to see
54.
by supermatt 2mo ago
How does "shooting" on 70mm relate to the projection of the 70mm? I have absolutely no idea how any of this works, but are there no CGI effects in the movie or how does the CGI get on the film?
55.
by supermatt 2mo ago
> Marble makes tools visible and easy to select for each task Isn't that what current agents already do? e.g. codex: https://ibb.co/Y71qJJZb
56.
by supermatt 2mo ago
Of course - its only a few years old :) But I think the biggest problem is that no-one directly tells you how to use the technology effectively. Mostly because no-one really knows, hence the reason for all the competing harnesses, agents, e
57.
by supermatt 2mo ago
I did something similar in 2020 when we constructed our sauna/guesthouse. Used a program called "LiveHome3D" to build out how what we were looking for, then exported as an OBJ and imported it via Google Poly into "Tilt B
58.
by supermatt 2mo ago
Few pointers: Dont try and handle the entire project in context. Use a well structured filesystem layout for your code with a few lines in an AGENTS.md describing the layout and core architectural requirements (no more than that, as per the
59.
by supermatt 2mo ago
I didn't get the reference, but it looks like im going to have to watch that series now :D
60.
by supermatt 2mo ago
There was an article a few years ago called "Lost in the Middle: How Language Models Use Long Contexts" https://arxiv.org/abs/2307.03172 From my experience this holds true to this day. It was one of my core o

More ›