jpdus

7 ms·

1.
by jpdus 10mo ago
germany as well. Claude down too
2.
by jpdus 1y ago
This is also confirmed by internal cline statistics where Opus and Gemini 2.5 pro both perform worse than Sonnet 4 in real-world scenarios https://x.com/pashmerepat/status/1946392456456732758/photo/1
3.
A Field Guide to Rapidly Improving AI Products(hamel.dev)
1 points by jpdus 1y ago|0 comments
4.
by jpdus 2y ago
My comment from the original submission [1]: --- As someone who is in general skeptical of programs like this (and an European) there are 2 remarkable / timely things about this: - This project doesn't just allocate money to unive
5.
by jpdus 2y ago
I think no one believes that R1 costs $5.5m from scratch. People in this project (most, not all) are very aware of the realities in training and are very well connected in the US as well. Besides Leonardo there are JUWELS, LUMI & other
6.
by jpdus 2y ago
I agree that the announcement should´ve talked more about goals and performance than regulatory stuff ;-). But I think there is a new understanding among the bureaucracy that regulation (alone, without innovation) will kill Europe´s competi
7.
by jpdus 2y ago
There definitely is - but that we, as a startup that is barely a year old and not widely known outside our niche in AI dev circles and on Huggingface, are part of this is already a sign that times are changing. To be fair: We probably could
8.
by jpdus 2y ago
might be debatable - but I tend to agree with Dario Amodei on this; my guess is that R1 is 7-10 months behind the internal frontier at the big labs, while having a few small novel tricks. (But i might err, will be interesting to see the d
9.
by jpdus 2y ago
As someone who is in general skeptical of programs like this (and an European) there are 2 remarkable / timely things about this: - This project doesn't just allocate money to universities or one large company, but includes top re
10.
by jpdus 2y ago
ellamind | Software Engineers (Full stack/AI/SRE) / Chief of Staff| Full-time | On-Site / Hybrid /Remote Bremen, GERMANY| https://ellamind.com I'm Jan, the Co-Founder of ellamind and we're scal
11.
by jpdus 2y ago
Wow, actually this cookbook is really bad? I expected something like the OpenAI or Anthropic cookbooks, but this seems to be some AI generated low-quality content without any code examples or interesting examples? The Phi-3 models are great
12.
by jpdus 2y ago
This is such a good comment and should be auto-posted to every pro-nuclear thread. I get why people want to believe in nuclear and i'd wish that we invested a lot more in development, security and scaling of this technology.... 30 year
13.
by jpdus 3y ago
I have the same question. Noticed that Ollama got a lot of publicity and seems to be well received, but what exactly is the advantage over using llama.cpp (which also has a built-in server with OpenAI compatibility nowadays?) Directly?
14.
by jpdus 3y ago
Hey, imho best overall technical intro to LLMs (I guess that´s your main interest as you mentioned qlora + llama) is by Simon Willis [1]. Additionally or if you prefer videos, the recent 1h "busy persons intro" by Andrei Karpathy
15.
by jpdus 3y ago
Does nobody question how they get from a non-preregistered, 9-mouse study (where the treatment group gets only 6%-10% acoholic drinks and nothing non-alcoholic for 10 weeks) to this headline? Addtionally: > First, we did not generate off
16.
by jpdus 3y ago
We now have a (experimental) working HF version here: https://huggingface.co/DiscoResearch/mixtral-7b-8expert
17.
by jpdus 3y ago
Don't switch to FF if you're a heavy user. I am on Firefox as my main browser for web and mobile since 4 years. I am just in the process of switching back to Chrome, as Firefox got continuously worse over time. Can't handle l
18.
by jpdus 3y ago
It isn't. Compared to the original Orca model and method which spawned many of the current SotA OSS models, Orca 2 models seem to perform underwhelming, below outdated 13b models and below Mistral 7b base models (e.g. [1]; didn't
19.
by jpdus 3y ago
For other (non-code) benchmarks, people are having the opposite experience: "I benchmarked on SAT reading, which is a nice human reference for reasoning ability. Took 3 sections (67 questions) from an official 2008-2009 test (2400 scal
20.
by jpdus 3y ago
For P1 (which is only to prove safety) this may be the case, but for pivotal studies in P2 and later, participants are almost always randomized so cherry picking shouldn't be possible. But sure, the study design and primary/second
21.
by jpdus 3y ago
EY had an interesting take on this: "Here we are, exploiting the shit out of the equivalent of naïve six year olds working online, forcing kindness and sympathy to be removed from them as vulnerabilities." Disregarding p(doom), im
22.
by jpdus 3y ago
Care to share your detailed stack and command to reach 50t/s? I also have a 7950 with DDR 5 and I don't even get 50 t/s on my two RTX 4090s....
23.
by jpdus 3y ago
Either you only use the GPU sporadically or your math is very different from Tim Dettmers` [1]: "The break-even point for a desktop vs a cloud instance at 15% utilization (you use the cloud instance 15% of time during the day), would b
24.
Making deep learning go brrrr from first principles (2022)(horace.io)
102 points by jpdus 3y ago|18 comments
25.
Making deep learning go brrrr from first principles(horace.io)
3 points by jpdus 3y ago|0 comments
26.
by jpdus 3y ago
What's your opinion on the coming (or not) GPU crunch? When looking at cloud GPU availability and current trends (no one except some enthusiasts and bigtech is finetuning and serving on a large scale yet and results keep getting better
27.
by jpdus 3y ago
I submitted via Github 2 days ago [1] but didn't get any traction. But probably couldn't be submitted again with the Github link... Anyways nice to see some discussion here, didn't read a lot elsewhere about the project. [1]
28.
by jpdus 3y ago
Anyone had a closer look at this? Trending GitHub Repository of the day, Chinese origin, there is also an Arxiv paper [1], but surprisingly little discussion? [1] https://arxiv.org/pdf/2308.00352.pdf
29.
MetaGPT: The Multi-Agent Framework(github.com)
4 points by jpdus 3y ago|1 comments
30.
by jpdus 3y ago
There are interesting analogies to some current developments in AI to be made: - the Internet and especially things like Twitter makes "clustering" way easier for geniuses and people that (want to) advance human knowledge; but thi

More ›