vova_hn2

9 ms·

151.
by vova_hn2 6mo ago
> I'm sure with a 3.8k stars repo you'll find competent people willing to continue the work. Oh yeah, I'm sure you will find lots of competent people. Like Jia Tan, for example. I've heard he is very competent.
152.
by vova_hn2 6mo ago
> Moleskine have not provided evidence of completely human-made artwork for the covers > but they also have not credited an artist or provided any proof of human creation What kind of "evidence" or "proof" would sa
153.
by vova_hn2 6mo ago
I think that BPE visualization is slightly misleading, because it seems to imply that the "old" (smaller) tokens are thrown away and replaced with longer tokens, which is not the case. In fact, it is purely additive process: we it
154.
by vova_hn2 6mo ago
The demo on the main page suggests "Paris Travel Overview / Visiting Notre Dame", so I tried a couple of cities and places that I've actually been to. If often identifies some of the points of interest correctly, but the
155.
by vova_hn2 6mo ago
While the engineering behind this achievement is really impressive, it doesn't feel that important in the grand scheme of things. We had machines "beating" humans in physical tasks for a very long time. No one would be impres
156.
by vova_hn2 6mo ago
Thanks for the image, I will see their faces in my nightmares.
157.
by vova_hn2 6mo ago
How can it result in a higher level of control? I don't see why the "judge" should have access to anything except one tool that allows it to send an "accept" or "deny" command.
158.
by vova_hn2 6mo ago
I doubt that you would install this application without even reading the README, so I don't understand how citing literally the second paragraph of the README helps. > it's critical to know whether it's vibe coded Strictly
159.
by vova_hn2 6mo ago
Is anti-AI sentiment somehow exempt from the "shallow dismissals" rule?
160.
by vova_hn2 7mo ago
> I backed up the risks with a link. The problem is that your link doesn't really backs up your claims. #1 and #5 deal exclusively with people with various kidney conditions. #2, #3, #4 and #6 deal with issues that are only tangenti
161.
by vova_hn2 7mo ago
> A weight-based recommended daily allowance (RDA) of 0.8 grams per kilogram of body weight. > However, for the average healthy person (who is not an elite athlete or heavily involved in body building) it's probably best to keep
162.
by vova_hn2 7mo ago
You just googled or asked a chatbot to find you an article and haven't actually read it, have you? #1 literally says > For people with healthy kidneys, higher protein intake is generally safe. #2 addresses the issues of some specifi
163.
by vova_hn2 7mo ago
There is basically no downsides to eating too much protein and there are a lot of potential problems if you eat too little. Of course, you can carefully design other meals to be more protein-heavy, just because of your weird idea of a havin
164.
by vova_hn2 7mo ago
> Unprocessed fiber-containing carbs are fairly healthy. I never said that oats are unhealthy or that carbs are unhealthy. I said that a meal without protein is not the best idea. My critique was not directed at a particular food item bu
165.
by vova_hn2 7mo ago
Okay, but it doesn't address the issue of having a whole meal with barely any protein.
166.
by vova_hn2 7mo ago
Oats are just carbs. > You can enrich it with bananas or goji berries for extra nutrients. "extra nutrients" are more carbs. > it’s healthy I don't think that a whole meal without a good protein source can be considered
167.
by vova_hn2 7mo ago
It is possible of course, but I see no reason to believe it.
168.
by vova_hn2 7mo ago
> From a naive perspective, more usage should mean more revenue since customers pay per token. They are not limiting usage through the API with per token payment. They are limiting usage through a subscription that gives you a lot (but t
169.
by vova_hn2 7mo ago
> This is a library showing you how to train your own Claude Code end-to-end. What does it even mean? Claude Code is a so called "harness" - a thing that builds a context for LLMs, calls LLMs, executes tool calls etc. It uses v
170.
by vova_hn2 7mo ago
> I still remember that people used to enter full questions in Google I think that, in the early days of internet search, entering full questions actually produced worse results than just a bunch of keywords or short phrases. So it was a
171.
by vova_hn2 7mo ago
> Has basic reading become a Herculean task? I find LLM slop much harder to read than normal human text. I can't really explain it, it's just a feeling. The feeling that it draaaags and draaaaaags and keeeeeps going on and on a
172.
by vova_hn2 7mo ago
> instead of talk like a caveman you could turn off reasoning, with probably better results This is not how the feature called "reasoning" work in current models. "reasoning" simply let's the model output and the
173.
by vova_hn2 7mo ago
I don't know about token savings, but I find the "caveman style" much easier to read and understand than typical LLM-slop.
174.
by vova_hn2 7mo ago
Yeah, I don't think that "I'd be happy to help you with that" or "Sure, let me take a look at that for you" carries much useful signal that can be used for the next tokens.
175.
by vova_hn2 7mo ago
The main difference in the current (theatrical) permission model is that the agent is blocked on waiting for your approval. So you can't just launch it and go do something else, because when you return you will see that nothing is done
176.
by vova_hn2 7mo ago
> yeah it kinda works but not really well enough I mean, at some point it was true. I remember that around 2023, when I first encountered colleagues trying to use ChatGPT for coding, I thought "by the time you are done with your bac
177.
by vova_hn2 7mo ago
I think that those permissions are largely security theater anyway. It would be better if an LLM coding harness just helped you set up a proper sandbox for itself (containers, VMs etc.) and then run inside the isolated environment unconstra
178.
by vova_hn2 7mo ago
I was almost sure that they do it client-side for a joke, but if you check the browser console, you can see that they actually make a request! You can even make the same request with curl and it works! Although, making an HTTP request manua
179.
by vova_hn2 7mo ago
It's "Showing 1 - 25 of 58891 posts" now. HN tells me that your comment was posted 6 minutes ago, which gives us approximately 37.23 posts/second rate.
180.
by vova_hn2 7mo ago
1. Is the given tok/s estimate for the total node throughput, or is it what you can realistically expect to get? Or is it the worst case scenario throughput if everyone starts to use it simultaneously? 2. What if I try to hog all resou

More ›