WASDx

4 ms·

1.
by WASDx 6d ago
The model is only 0.3B params so probably not much.
2.
by WASDx 10d ago
I tried calculating historical "intelligence per cost" recently but stopped when I realized intelligence is not linear. For any meaningful "x per y" you can just double "y" if you have a half as efficient syste
3.
by WASDx 11d ago
It works if you neglect code quality and understanding. I've done a few vibe coded projects on my free time but I would feel shame for submitting that kind of code at work where I use LLMs more responsibly.
4.
by WASDx 21d ago
Likely because it uses fewer thinking tokens (that you don't see anyways).
5.
by WASDx 21d ago
Give a task you have to 3 different models and see what actually works for you. There are no good benchmarks.
6.
by WASDx 22d ago
This is like saying mass production is "cheating" against handcraft.
7.
by WASDx 22d ago
Once it figures out a puzzle it could probably be instructed to design a specialized harness for Luna to be able to solve other instances of the same puzzle. Minimum wage workers are not solving novel problems.
8.
by WASDx 23d ago
They explain it here: https://openai.com/index/how-two-settings-tripled-our-arc-ag... TLDR: The official ARC harness throws away old context and reasoning. No real-world harness is this bad, the model has to re-learn t
9.
by WASDx 24d ago
With the contributor pricing being more than 10x cheaper than the standard, that would make it best and cheapest on the DeepSWE leaderboard! It feels fast in my experience too. LLMs keep improving at an insane pace.
10.
by WASDx 24d ago
I'm party using 1.2 to reverse engineer and re-implement an old game binary and it has been quite good and fast. The contributor pricing is very attractive, excited to try 1.3 and see if I feel a difference. 1.2 can get stuck outputtin
11.
by WASDx 24d ago
3.7 high and 3.8 medium are essentially the same on AA intelligence and cost. Output tokens on DeepSWE gives the same picture. So there might be something to it but they have done other things as well. At least the tokens are really fast.
12.
by WASDx 27d ago
If their platform allows any kind of processing as it sounds then anyone could just dump the data. So I don't see why they would not allow downloads for local processing.
13.
by WASDx 29d ago
Then use other providers hosting open models. Companies and individuals already put their whole code base on the cloud. I'm genuinely interested in privacy-oriented use cases where ZDR is not enough.
14.
by WASDx 29d ago
I think the privacy argument that keeps coming up is overrepresented. Certainly ZDR is enough for an absolute majority of use cases? I see so much talk about local inference but I doubt most of it has privacy as a valid argument (not arguin
15.
by WASDx 29d ago
I would guess such a small difference is within the error margin so "roughly equal" I think is fair to say but different models are good at difference things in practice. It is about the same size as GLM 5.3 at about half the pric
16.
by WASDx 1mo ago
The memory and GPU are integrated into the CPU so those can't be upgraded anyways. That's also how the memory can be so fast (shorter physical distance).
17.
by WASDx 1mo ago
Good code is self-explanatory and don't need comments. I can really recommend the book Clean Code, here is a summary: https://gist.github.com/wojteklu/73c6914cc446146b8b533c0988c...
18.
by WASDx 1mo ago
"Qwen3.8 35B-A3B" and 4B/9B variants are already on huggingface distilled by hobbyists.
19.
by WASDx 1mo ago
> Agents playtest my games overnight I'm curious if you want to elaborate. What kind of games? Turn based? Do you just feed it repeated screenshots?
20.
by WASDx 1mo ago
I'm glad someone is voicing this. Overconsumption at that level is not defensible. However if they meant cached tokens so it's not that bad.
21.
by WASDx 1mo ago
That might actually compensate for the overthinking, if it can think really fast. Dense models are easier than MoE to put on silicon. https://chatjimmy.ai/ is getting 16k tps with an 8B model. Extrapolating that gives nearl
22.
by WASDx 1mo ago
On DeepSWE it's now 53% vs 63% which is one of the coding benchmarks I trust the most. DS own measurements also show a more significant increase so I suspect AA might update when they release an article. Surprisingly DeepSWE currently
23.
by WASDx 1mo ago
I needs to be harness+model combination, https://artificialanalysis.ai/agents/coding-agents
24.
by WASDx 1mo ago
Right. I'm also curious about those use cases. I don't want an AI clicking through my mailbox.
25.
by WASDx 1mo ago
I upgraded my workflow a few months ago from "copy-paste things in and out of ChatGPT" to "use an agent that edits my project files and runs tests on its own" and the ergonomics are just so much better and enables automa
26.
by WASDx 1mo ago
This was a disappointed to me. Why would I use pro over flash now? Is there some area where the difference is significant?
27.
by WASDx 2mo ago
If you take 10 turns with a model A, it has to read the cache 10 times and write a lot of tokens (the expensive part). Switching to model B is just prefilling the diff + your new message, which is still just one turn. So total number of tur
28.
by WASDx 2mo ago
And they are all TUI's installed via curl | bash.
29.
by WASDx 2mo ago
At 830tok/s * 1 hour that's almost 3M tokens which is just $0.54 worth of tokens at Deepseeks current output price.
30.
by WASDx 2mo ago
Do you know why they don't just cache the system prompt for everyone? It seems so wasteful not to.

More ›