pama

49 ms·

691.
by pama 3y ago
A lot of things become simpler, but hard problems remain. In the cell, a structure of a protein is the result of its sequence and all the other environmental conditions (cofactors, ligands, binding partners, lipids,…), and furthermore, the
692.
by pama 3y ago
I will update my language that mimicked the original comment. However it is not simple discussion. Here is a snippet related to Japan’s law, updated from ca 2018 with AI systems in mind, and clarified recently. I personally find it totally
693.
by pama 3y ago
Japan is already taking action regarding copyright law and the AI world is noticing: https://www.natlawreview.com/article/japanese-government-ide... https://www.cliffordchance.com/insights/resource
694.
by pama 3y ago
How does the recently proved zero-one theorem for graph neural networks affect such works? https://arxiv.org/abs/2301.13060
695.
by pama 3y ago
It is great that the FDA does these random checks and can identify supplement and food items that contained erectile dysfunction drugs in them. It is a tough call to try and completely eliminate such products from all the marketplaces, esp
696.
by pama 3y ago
The chatGPT model is not reproducing the contents nor making it available at scale. I, the user, ask the model to read the website and other websites and come back with a useful summary to me. This is not training data thus not violating c
697.
by pama 3y ago
If the future AI can indeed cure disease my mission of working in drug discovery will be complete. I’d much rather help cure people (my brother died of melanoma) than protect any patent rights or copyrighted text.
698.
by pama 3y ago
Please recall how much the world changed in just the last year. What would be your expected timescale for the solution of this particular problem and why is it more important than instilling models with the ability to logically plan and an
699.
by pama 3y ago
Treating computers like humans in this one particular way is very appropriate. It is the only way that LLM can synthesize a worldview when their training data is many thousands of times larger than their number of parameters. Imagine scali
700.
by pama 3y ago
This is not answering the GP question and does not count as a satisfactory ranked citation list. The first one is particularly dubious. Also you didn’t clarify which statement was based on which citation. I didn’t see “dog” in your text. To
701.
by pama 3y ago
There is no proof so far that openAI has violated any copyright. I understand the public’s emotions but recall that these large language models can read all the public info and make useful generalizations.
702.
by pama 3y ago
No. It is a technology that can massively accelerate human progress—I don’t buy the theory that OpenAI used NYTimes content that was not freely available to them. If you read all of the public internet you probably have lots of snippets o
703.
by pama 3y ago
[edit: they have since opened a comment section to the article.] It is unfortunate that the NYTimes don’t allow reader comments to this article. I like some of the NYTimes content, but in this case use of chatGPT is infinitely more valuable
704.
by pama 3y ago
And this is the zoology paper: https://arxiv.org/abs/2312.04927
705.
by pama 3y ago
See also this resource if you’re interested in these models: https://news.ycombinator.com/item?id=38719675
706.
by pama 3y ago
This page has materials that help understand these modern architectures that may replace transformers for many tasks over the next couple of years. From the team that included the amazing people that came up with flash attention, hyena vari
707.
by pama 3y ago
That concept breaks down with the multiple thousands of default key bindings in a typical Emacs setup (source: go into an org buffer press C-h b, then go to a Python buffer and press it again, and so on)
708.
by pama 3y ago
This blog post describes “Based” a mamba-adjacent architecture with huge context that achieves similarly low perplexity in NLP. The request came up in: https://news.ycombinator.com/item?id=38717236
709.
Building blocks for AI systems: links to modern state space model research(github.com)
2 points by pama 3y ago|1 comments
710.
by pama 3y ago
This blog about “Based” came out just before neurIPS: https://hazyresearch.stanford.edu/blog/2023-12-11-zoology2-b...
711.
Based: An Educational and Effective Sequence Mixer(hazyresearch.stanford.edu)
1 points by pama 3y ago|1 comments
712.
by pama 3y ago
1. Attention is quadratic with context length; RNN with gating (LSTM, GRU, etc) are linear, as are all these new architectures. Early RNN used gating to avoid exploding gradients, these new ideas use theory from dynamical systems that guara
713.
Demystifying Diffusion Models(developer.nvidia.com)
3 points by pama 3y ago|0 comments
714.
by pama 3y ago
ELO takes a while to establish. It does not sound likely that the newer GPT3.5 is that much worse than the old one that has a clear gap to all the non proprietary models. In the immediate test, GPT-3.5 clearly outshines these models.
715.
by pama 3y ago
17000000 Deaths per billion flights for the space shuttle is not safe. This is more than 2,000 times more dangerous than skydiving, which is also not very safe.
716.
by pama 3y ago
The additional link is indeed very useful, however, with one exception, it is limited to inference, ie code for evaluating a model using somebody else’s trained model weights. The original link has code for training (which can also be used
717.
by pama 3y ago
There is no limitation in copying the work of an author for personal enjoyment without permission. If I use my neurons to read and create an exact copy I can retell to myself, or I write it on pen and paper, or I type it on a personal comp
718.
by pama 3y ago
It does not sound like what she did helps advance the development of AGI that is broadly beneficial. It simply helps slow down the most advanced current effort, and potentially let a different effort take the lead.
719.
by pama 3y ago
Absolutely. There simply hasn’t been a better/faster implementation outside the existing TeX ecosystem so far. Returning to the comment that I replied to: everybody could rewrite/extend Fortran; it will still be Fortran.
720.
by pama 3y ago
I'm guessing that unless the board caves to everything that the counterparties ask of it, MSFT lawyers will very soon reveal to the board the full range of possible legal actions against the board. The public will probably not see man

More ›