armcat

13 ms·

181.
by armcat 2y ago
I tried one of their "distill" versions on HF Spaces: https://huggingface.co/spaces/Aratako/DeepSeek-R1-Distill-Qw... . It seems to suffer from the same old repetition and overthinking problems. Using the
182.
by armcat 2y ago
This is brilliant! I spent several years with in QA and even built an ML system for categorizing faults. It's incredible how far we've come though, well done! Do you mind telling us a bit about the AI tech/stack that is used
183.
by armcat 2y ago
Nice! That was the first thing I did as well, ask ChatGPT. The only one not attempted from the list above is the compressed air method. Thanks!
184.
by armcat 2y ago
Great suggestion, that's the one she hasn't tried, I'll let her know!
185.
Ask HN: Separating Two Glass Containers
4 points by armcat 2y ago|8 comments
186.
by armcat 2y ago
This is awesome, and simplifies lot of my workflows when using their APIs directly. I also want to give a shout out to Outlines team, https://github.com/outlines-dev/outlines , they've been doing structured outputs
187.
by armcat 2y ago
My understanding is that Perplexity AI doesn't "just Google it". They have their own indexing/crawling engine called PerplexityBot. They also have their own ranking engine, but they use ranking signals from both Google a
188.
by armcat 2y ago
Perplexity doesn't actually use Google Search directly. It uses its own indexer and crawler called PerplexityBot [1]. It then uses a mixture of Google and Bing ranking signals to "help" with some of its own search result rank
189.
by armcat 2y ago
Amazing concept! I am a heavy user of such tools, and typically build my own bespoke search. I tried out Find AI across a number of different queries, and I think the general issue here is one of coverage. For example, when searching for &q
190.
by armcat 2y ago
Thanks for the link to Conv-KAN!. I had a quick look and I have a few points. Firstly, they use a different implementation compared to the original KAN paper, i.e. they use efficient-kan ( https://github.com/Blealtan/eff
191.
Claims from Kolmogorov-Arnold Networks (KAN) research may be false(linkedin.com)
10 points by armcat 2y ago|2 comments
192.
Agent Hospital that simulates the entire process of treating illness(arxiv.org)
5 points by armcat 2y ago|0 comments
193.
by armcat 2y ago
Bit off-topic but I love the artwork on that site, and especially the parallax on the image at the top.
194.
by armcat 3y ago
One reason is that overall there are more PyTorch based ML projects out there, which translates to larger exploration space and wider support base. Around the beginning of 2021 PyTorch overtook TensorFlow as the ML framework of choice, see
195.
by armcat 3y ago
"Open source" or "open weight"? Because there is a distinction. Many have previously provided open weights (or what they call "open model" now): Mistral, LLaMA, Falcon, etc. There are not many open "source
196.
by armcat 3y ago
I've worked with EXACT this type of problem and for me RAG works perfectly well - it may seem "clumsy" as you put it in terms of trying to engineer or optimize the indexing, augmentation, and retrieval techniques, but it&#x27
197.
by armcat 3y ago
RAG approaches should work quite well for the examples you mentioned. It's a matter of how you approach the retrieval part - you can opt for a larger recall on retrieval, and leverage the large context window for the LLM to figure out
198.
by armcat 3y ago
It looks like it touched down, mission control picking up a faint signal, trying to refine it now.
199.
by armcat 3y ago
Aldi is big in Australia too, around 600 stores across the country.
200.
by armcat 3y ago
I went a bit meta on this and supplied the trademark report to ChatGPT, prompting it to summarize. It did so as follows: The US Trademark Office rejected the application, deeming "GPT" merely descriptive of the features, function
201.
by armcat 3y ago
Since the original motivation from the OP was to understand maths in ML papers, has anyone tried the free Maths for ML book, https://mml-book.github.io/ , and if so what are your thoughts?
202.
by armcat 3y ago
This is an awesome collection, thank you! For lots of the conferences there are different "tracks", and each one has "the best paper award". For example, at NeurIPS 2023, there was a "main track", and also a &q
203.
Gaia – Zero-Shot Talking Avatar Generation(arxiv.org)
2 points by armcat 3y ago|0 comments
204.
by armcat 3y ago
Interesting, so they are using single precision (fp32), so 2.7B x 4 Bytes = ~10 GB. With CUDA overhead and room for context, you would need at least 12GB VRAM. They could use half precision and half that VRAM requirement and save costs for
205.
by armcat 3y ago
Looks like the embedded slideshow is what is "free". They want you to purchase the actual book.
206.
Deep Learning – Foundations and Concepts (Chris Bishop)(bishopbook.com)
236 points by armcat 3y ago|36 comments
207.
by armcat 3y ago
Those impressive demos, e.g. the cup shuffling seem to have been "staged". The end results are correct, but the method of getting them is nowhere near as fluid and elegant as in the demo. They used a series of still images with ca
208.
by armcat 3y ago
Not necessarily, it would be just RAG, the use the standard Bing search engine to retrieve top K candidates, and pass those to OpenAI API in a prompt.
209.
by armcat 3y ago
Everything on social media (and general news media) pointed to Ilya instigating the coup. Maybe Ilya was never the instigator, maybe it was Adam + Helen + Tasha, Greg backed Sam and was shown the door, and Ilya was on the fence, and perhaps
210.
by armcat 3y ago
The final codebase, yes. But ML is not like traditional software engineering. There is a 99% failure rate, so you are forgetting 100s of hours that go into: (1) surveying literature to find that one thing that will give you a boost in perfo

More ›