deepdarkforest

10 ms·

61.
by deepdarkforest 1y ago
CT-ART 4.0 is the gold standard. Again, not fully free, but it has some very instructional features, like playing against alternate moves, solving mini versions of a puzzle, playing the opposite side etc. Used it for years.
62.
by deepdarkforest 1y ago
you misunderstood. its 300ms just for the apply model, the model that takes your coding models output (eg sonnet) and figures out where the code should be changed in the file. Cursor has its own, and claude uses a different technique with s
63.
by deepdarkforest 1y ago
Glad to hear quality comes first! Then I assume you have some public benchmarks like the ones you mention that are reproducible? I could only find this graph https://docs.morphllm.com/guides/apply but there is no menti
64.
by deepdarkforest 1y ago
> 1) Raw inference speed matters more than incremental accuracy gains for dev UX—agree or disagree? I know you are trying to generate some controversy/visibility, but i think if we are being transparent here, you know this is wrong.
65.
by deepdarkforest 1y ago
It's not silly. I would bet 99% of the users don't care that much to do that. A hardcoded regex like this is a good first layer/filter, and very efficient
66.
by deepdarkforest 1y ago
Cursor raised 900M, are losing market share to claude code(resorting to poaching 2 leads from there [1]), AND they're decreasing the value of their product? Huge red flag. They should be able to burn cash like no tomorrow. Also, the PR
67.
by deepdarkforest 1y ago
From a cash flow perspective of course it makes sense to sell the future before you have it as working product. It just needs a great salesman or narrative to keep it going, im not arguing that. > that feedback loop is very important f
68.
by deepdarkforest 1y ago
> the demand side, the basic problem with humanoid robots is that they're mostly useless right now ... ... to square this circle, ... we will provide free hardware and software upgrades until we are able to make the robot fully aut
69.
by deepdarkforest 1y ago
If you actually click on the link, it mentions this both in the abstract, and a detailed comparison of evidence in a whole table.
70.
by deepdarkforest 1y ago
The main problem with these approaches is that most sites now are useless without JS or having access to the accessibility tree. Projects like browser-use or other DOM based approaches at least see the DOM(and screenshots). I wonder if you
71.
by deepdarkforest 1y ago
the difference is simple, addiction is the compulsive/habitual action of something that does not have positive effects, more likely negative than neutral. For example, being "addicted" to gym is fine, because even if you enjo
72.
by deepdarkforest 1y ago
This might not get a lot of traction because it's very technical, but i wanted to say a massive well done for the effort. 20k words on anything this specific is not a joke. I wish i would put this level of commitment to anything in lif
73.
by deepdarkforest 1y ago
What irks me about anthropic blog posts, is that they are vague about details that are important to be able to (publicly) draw any conclusions they want to fit their narrative. For example, I do not see the full system prompt anywhere, only
74.
by deepdarkforest 1y ago
By competitive, i mean no.1 in LM arena overall, in webdev, in image gen, in grounding etc. Plus, leading the chatbot arena ELO. Flash is the most used model in openrouter this month as well. Gemma models are leading on device stats as well
75.
by deepdarkforest 1y ago
> Sundar is a really uninspiring leader I understand, but he made google a cash machine. Last quarter BEFORE he was CEO in 2015, google made a quarterly profit of around 3B. Q1 2025 was 35B. a 10x profit growth at this scale well, its un
76.
by deepdarkforest 1y ago
remember, gaps in the market sometimes exist for a reason. Forget AI. How many open source, community driven and privacy first browsers have made serious money? Brave is a decent example but their business model is actually complicated, it
77.
by deepdarkforest 1y ago
This is definitely a winners take all market. Kudos for giving it a shot, but imo browser projects are just too big for a team of 2/3. Plus, google has already demoed at IO the first hint at this. IMO you just cannot move fast enough t
78.
by deepdarkforest 1y ago
you will run into the same problems the extension approaches have (Like nanobrowser etc). Which is if i have to supervise constantly for non reversible actions, then im no more efficient(actually less i would argue) than just doing the task
79.
by deepdarkforest 1y ago
For easy/format stuff for specific journals it will be useful. But please, please for the love of god don't try to give actual feedback. We have enough GPT generated reviews on openreview as it is. The point of reviews is to get d
80.
by deepdarkforest 1y ago
Terrible stuff and a reddish flag. First of all, gpt signs all over the blog post, reads like a bottom of the barrel linkedin post. But more importantly, why double and triple down on no RAG? As with most techniques, it has its merits in ce
81.
by deepdarkforest 1y ago
wouldn't that be just 3rd party llms.txt?
82.
by deepdarkforest 1y ago
just wanted to say this was the most relatable take i have read so far, and i've read a lot. Exact same experiences. And you didnt even touch on the MCP's that enable these things to go wild as well. I think our takes are not bein
83.
by deepdarkforest 1y ago
1. Oh yes right. I remember trying it out thinking it was going to be brittle because of analytics etc but it filters for those surprisingly well. 2. We are working on https://www.launchskylight.com/ , agentic QA. For the
84.
by deepdarkforest 1y ago
Very cool. 1) How do you deal with timings? If a step includes clicking on a link or something that needs loading, then if you just fire off the generated playwright code at at once, some steps might fail because the xpath is not there ye
85.
by deepdarkforest 1y ago
Not sure if this can work. We played around with something similar too for computer use, but comparing embeddings to cache validate the starting position is super gray, no clear threshold. For example, the datetime on the bottom right chang
86.
by deepdarkforest 1y ago
Looks super interesting. I have couple of very annoying workflows that i'm looking to automate but no API's. Can i install my own apps?
87.
by deepdarkforest 2y ago
Love when someone that actually knows their stuff does a deep breakdown like this. Super useful I wonder though if all it matters is the last punchline. The profit margin vs competitors. If llms truly get commoditized and do not benefit fro
88.
Show HN: Picocrowd – AI Teams That Click, Browse, and Work Together(picocrowd.com)
3 points by deepdarkforest 2y ago|0 comments
89.
by deepdarkforest 2y ago
Very clean writeup. On the attention sinks, you mention they enable "infinite-length sequence processing". What does that mean exactly in practice? Isn't deepseek still capped at 128k?