ofou

why not just use linux/?

The UK is becoming 1984.

I want a simple USB-C earpods with noise active cancellation. How is that hard for Apple?

privacy is a right, until you don't have a voice to say so.

this is just https://github.com/tinygrad/tinygrad

Demoniac move by Spotify

The model released is called Latam-GPT, the first large-scale open-source language model developed specifically for Latin America and the Caribbean. Its size is 70 billion parameters (70B). It is a continuous pre-training (CPT) adaptation of Meta’s Llama 3.1 70B base model, trained on a regionally-curated corpus of ~300 billion tokens that emphasize Latin American languages, cultures, and contexts.

Press release: https://cenia.cl/2026/02/10/latam-gpt-la-primera-ia-regional...

Presentation (in Spanish): https://www.youtube.com/watch?v=FdLzAiQizhA

HF: https://huggingface.co/latam-gpt

Github: https://github.com/latam-gpt

btw, you can change the hashing algorithm in git easily

This is great news for the community. We should measure a bit more closely how tool usage behaves in the wild. The gains you can get from different providers are not marginal.

Sometimes it's not just raw intelligence, but engineering as well

UTF-8 should be a universal tokenizer

the internet's own boy :( I'll be always deeply touched by him

Shadow libraries maintainers deserve a Nobel prize for their contributions to humanity. Satoshi would be proud.

This is stupid because a UTF8 is a tokenizer that covers all Unicode with a vocab of only 256 (yes, without a K). This is the only way of scaling the bitter lesson with tokenizers. Also, with architectures that span +1M context windows, it’s no longer an argument/issue the reduced context windows.

I'd submit a PR with this idea to improve coverage of agents

Is this Garcon [1], or a new tool?

[1]: https://transformer-circuits.pub/2021/garcon/index.html

No luck so far with: When does the BB(6) halt?

considering the amount of bots in HN, not really that much

mmm... like a whole Harry Potter series

My most sincere love to all shadow libraries out there, you're doing god's work.

You gotta love these guys, they're really pushing the open source frontier for all of us, thanks for sharing

https://www.emergentmind.com also offers Deep Research on ArXiv papers (experimental)

This is one of the reasons I've been advocating to use UTF-8 as a tokenizer for a long time. The actual problem IMHO are tokenizers themselves, which obscure the encoding/decoding process in order to gain some compression during training to fit more data in for the same budget, and arguably gaining some better understanding from the beginning. Again just a lack of computing power.

If you use UTF-8 directly as tokenizer, this problem becomes evident once you fit it into the context window. Plus, you can run multiple tests for this type of injection; no emoji should take more than up to 40 bytes (10 code points * 4 bytes per code point in the worst case). This is an attack on tokenizers, not on UTF-8.

Plus, Unicode publishes the full list of sequences valid containing the ZWJ character in emoji-zwj-sequences.txt

Who would have known that BitTorrent, shadow libraries, and seeders will help to train the best AI models out there, that adds a whole new meaning to a "seed".

It’s called DeepSeek. The founder just confirmed a few days ago that he got the data from Anna's to train on, I think for their latest vision model.

This is a wonderful submission to Anna's archive [1]. I really love people pushing the boundaries of shadow source initiatives that benefit all of us, especially providing great code and design. Can't emphasize enough the net plus of open source, BitTorrent, and shadow libraries that have had in the world. You can also make the case that LLMs wouldn't have been possible without shadow libraries; it's just no way of getting enough data to learn.

Just thank you.

https://software.annas-archive.li/AnnaArchivist/annas-archiv...

I find quite interesting they're releasing three compute levels (low, medium, high), I guess now there's some way to cap the thinking tokens when using their API.

Pricing for o3-mini [1] is $1.10 / $4.40 per 1M tokens.

[1]: https://platform.openai.com/docs/pricing#:~:text=o3%2Dmini

competition is beneficial for all of us, this is great

Where can I get the actual tokenizer data?

Nevermind, it's here

https://api-docs.deepseek.com/quick_start/token_usage

By the way, I think this is a fantastic reading list for creating AI products, and especially for staying updated on the latest in the AI space. However, it feels a bit scattered and might be hard for beginners to follow, IMO.

I read your book, The Coding Career Handbook, we need something similar for AI Engineering! I really enjoyed it. Thank you for creating and sharing such high-quality multimodal content :)

Dive into Deep Learning is implemented using various libraries such as PyTorch, NumPy/MXNet, JAX, and TensorFlow.

Here’s an example: https://d2l.ai/chapter_natural-language-processing-pretraini...