rfoo

29 ms·

361.
by rfoo 2y ago
Well, 010 Editor also checks all three. So: 4. Unlike 010 Editor, it does not take you $150 and is FOSS so you can easily patch it to do whatever you need.
362.
by rfoo 2y ago
You can fill the form here: https://support.anthropic.com/en/articles/8241253-trust-and-... and eventually you get unbanned. Unfortunately, by the time Anthropic's next model is out you are likely to be banne
363.
by rfoo 2y ago
It wasn't unique to Korean. 15-20 years ago Chinese banking website did exactly same things. Maybe they got the idea from SK, don't know. It only ended with widespread adoption of mobile phones and Chrome.
364.
by rfoo 2y ago
Well, it would be depressing if we have to rely on The Shadow Brokers-style "whistleblowers" in the future.
365.
by rfoo 2y ago
Not without occasional problems. It shows how much original team cares about the product though. Feels like if running Keybase was already financially viable at that time they won't join Zoom.
366.
by rfoo 2y ago
This is the third time Anthropic releases something interesting and I try to login to claude.ai and get banned by an "automatic review". Thankfully they are not Google, if you follow the appeal instructions they actually un-ban yo
367.
by rfoo 2y ago
Dunno. It could also just mean the so-called "Quantization-aware training" where your weight, activation and gradient is still bf16 and just before use it gets quantized to int8 in the same way you'd do it during inference. T
368.
by rfoo 2y ago
I thought we are having oversupply right now.
369.
by rfoo 2y ago
I agree. However, it is not a clear cut what's fair and what is "gaming the benchmark" in this setup, for example: - can I train on my own private training set (which is harder)? - can I pretrain on The Pile or something simi
370.
by rfoo 2y ago
> You're saying the publicly available problem set isn't indicative of the distribution of the test set? Yes. From https://arcprize.org/guide : Please note that the public training set consists of simpler
371.
by rfoo 2y ago
That's the point. /s
372.
by rfoo 2y ago
First, LLMs are not AGI. Never will be. Can we talk now? > if given the entire test set. I don't want the entire test set. Or any single one in the test set. The problem here is ARC challenge deliberately give a training set with di
373.
by rfoo 2y ago
... and we know that if we really want to nail it we'd better just pay someone else to create 1,000,000 more harder problems for training (without looking at any in test set, of course). i.e. make the training set distribution similar
374.
by rfoo 2y ago
DeepSeek V2 was largely dismissed on HN due to being large and require non-trivial (i.e. vanilla llama.cpp or even vLLM does not run it) software to run (my guess). And their hosted API is heavily censored because of China. Let's see w
375.
by rfoo 2y ago
VSCodeVim/Vim misses some of my favorite text objects and most importantly, it still feels laggy even on a 7950X :(
376.
by rfoo 2y ago
Sorry, I can't. My employer doesn't publish this kind of metrics, either. What I posted was definitely just some very rough number off my brain.
377.
by rfoo 2y ago
IMO the relevant benchmark for now is a mixed stream of requests with 50 (20%), 500 (50%), 2000 (10%) and 50k (20%) input tokens, ignore EOS and decode until you get around 300 output tokens.
378.
by rfoo 2y ago
"GPU has fallen off the bus" is an actual error message nvidia.ko prints to dmesg in this case :p
379.
by rfoo 2y ago
> will scale faster * Only in USA, both intentionally and not.
380.
by rfoo 2y ago
Which languages? IIUC Go and Rust resort to OS IPC based plugin system mainly because they refused to have a stable ABI. On the other hand, at $DAYJOB we have a query engine written in C++ (which itself uses mostly static linking [1]) loadi
381.
by rfoo 2y ago
> integrated memory Yes. The cost of bonding memory to their chip is mostly the same for 8G / 16G / 32G / practically any number.
382.
by rfoo 2y ago
> to reason Should we expect current LLMs to be able to reason at all? As the paper said GPT-4 maybe make a better scanner. The problem is too many people think all we need is better scanners, which is not new.
383.
by rfoo 2y ago
There's certainly smaller and even better models for OCR. But the whole "point" of LLM (forget it, it's not AGI) is you don't need to make many specialized models and cursed pipelines anymore, to solve a definitely-
384.
by rfoo 2y ago
> but it also get devalued relative to USD when these coins inevitably get sold to pay expenses Good point. It invalidates the "good" part but does not make it doubly bad IMO. And for ETH, well, I don't think it's abo
385.
by rfoo 2y ago
Honest question, just curious what's your definition of "different arch", is a model with GQA=8 different than one with GQA=16 in your mind? I guess no? How about GQA vs MHA, or GQA vs MLA? If anything attention-like is same
386.
by rfoo 2y ago
A100s, (old) A800s, H100s or (old) H800s. How did they get them? Who knows, but it's there.
387.
by rfoo 2y ago
Indeed, I guess you can say I hate wealth. Wel, but at least in PoW you burn actual money (and in the end, actual resource) proportioned to your profit to keep the network running. In PoS you burn nothing.
388.
by rfoo 2y ago
Yeah, PoW is bad too. But I'm happy to pay a tax to those who burnt energy to keep the networking running and converted USD to the native token, proportioned to their effort. I'm less happy to pay someone a tax just because they a
389.
by rfoo 2y ago
> Participating in staking is fully permissionless, stakers are not taxing non-stakers. They are being remunerated for their work. That's just a more polite way to say tax. Being permissionless is cool, but it's still tax in my
390.
by rfoo 2y ago
Sure, friends also won't let friends skip the fact that circulating supply of ETH is now decreasing instead of increasing. Also, only ~30% tokens are staked. The 30% who chose to stake essentially tax the other 70% in use. Each of the

More ›