31–40 of 43 posts
- Top ML papers of the week: https://github.com/dair-ai/ML-Papers-of-the-Week - Daily Newsletter on AI: https://tldr.tech/ai - Subreddit about open-source LLM advancements: https://www.reddit.com/r/LocalLLaMA/
I leave it to YOShInOn, the smart RSS reader I have been dreaming about since 2005. YOShInOn ingested 2054 articles (not all about A.I.) and picked out 300 to show to me based on a transformer model. I make thumbs up, thumbs down or favorite jugements on the articles in an interface that looks like TikTok or Stumbleupon. One feed it ingests is https://arxiv.org/list/cs/new and it's learned that I really like anything…
I'm really interested in your RSS reader - is it open source or available anywhere?
I leave it to YOShInOn, the smart RSS reader I have been dreaming about since 2005. YOShInOn ingested 2054 articles (not all about A.I.) and picked out 300 to show to me based on a transformer model. I make thumbs up, thumbs down or favorite jugements on the articles in an interface that looks like TikTok or Stumbleupon. One feed it ingests is https://arxiv.org/list/cs/new and it's learned that I really like anything…
I'm really interested in your RSS reader - is it open source or available anywhere?
Earlier quoted context omitted.
Slightly off-topic, but I am just starting to learn about the field, and saw > GPT4 still projects information into some highly non-linear latent space and samples from that Do you have more information/know where I can find things to read about the "highly non-linear" part of that? I've been reading some stuff[1][2][3] about smaller models, and my impression has been that the latent space is shockingly linear for th…
The space must be non-linear, as a consequence of the non-linear activation functions. A neural net is just a big math equation, and deeply embedded throughout are non-linear transforms which necessarily make the entire transform non-linear. Just like the presence of 1/x makes an equation no longer linear (at least with rare exception!). Here is a deeper explanation/visualization: https://colah.github.io/posts/2014-0…
This clarifies things a lot, thanks!
* I work full time, not in AI, and don't have tonnes of free time. It is unlikely I will get to even Masters level, let alone PhD, unless something changes drastically. Therefore I don't need to go super deep into papers and suchlike.
* On the other hand I want some tactile understanding, so doing Andrej Kaparthy's course. Fiddling with some PyTorch, even if I know my model will be shittier than the cheapest of chips OpenAI models, just to get a feel.
^ From this I really got a sense of what an embedding is, and why we have those and not human picked features, for example. I feel this helps me understand stuff I read online a lot more easily. So speeds things up in general even if I am just hustling with various APIs from JS code or whatever.
* Once I understand how transformers work, roughly, I will move on to the FastAI course which I think will give me a broad but shallow view of lots of models, and sort of make the top part of the T shape.
* I really like microsoft's "guidance" api, so I will probably focus on building out quick apps using that to solve everyday problems. There is also YC company doing something similar called Glass.
* I will ignore the latest LLaMA or other fluffy animal derivatives. They come out at a crazy rate, all seem to run on different platforms, some Python, some C++, need a decent machine etc. I feel that is moving to fast to even keep up on the download it and try it cycle. Probably the most I will do is play with them in a HuggingFace interface.
* Generally will ignore all of that stuff you mention i.e.: press releases/research papers/news/blogs/library updates/videos and tweets. Too much of a barrage of information.
* Most important: find a problem to solve.
I see myself enjoying being an applier, a glue coder, rather than the person who keeps training model and writing math hoping that their idea will be the next breakthrough (sort of like an Edison I guess), but kudos to those who do it. At the same time I want to understand a little of what happens under the hood!
Arxiv + google + papers with code is how I do it. As another user pointed out, focus on solving your problem. By doing this, you'll search and then naturally find relevant papers. After a while, a lot of papers become super easy to read because they're conceptually easy to grasp. Once you hit this point it becomes extremely easy to skim the content that you're searching for. Once I identify something interesting in my problem space or adjacent then getting through it is the hard part.
TLDR: Focus on a problem domain and search within it. Don't just shotgun it. If its relevant enough to you it will show up. Use filters like HN and whatnot.
AI/ML researcher + builder. Mainly HN and ML subreddit, and a bunch of newsletters. My overarching goal is to completely stay away from Twitter (which I have stopped going to 2 months ago). I instead rely on the above to bubble up interesting things. Also as others said, find something to build and work on and don’t keep looking sideways (aka Twitter, random news), just keep going. It can be discouraging and depressi…