🔹 **DeepSeek-V4-Flash:** 284B total / 13B active params. Your fast, efficient, and economical choice.

On a 150-class benchmark, the surrogate fully replaces the teacher

One estimate puts Llama 3's information compression at just 0.07 bits per token meaning the model has only a hazy recollection of most of what it trained on

Eight out of eight models detected Mythos's flagship FreeBSD exploit, including one with only 3.6 billion active parameters costing $0.11 per million tokens.

We present Attention Editing, a practical framework for converting already-trained large language models (LLMs) with new attention architectures without re-pretraining from scratch.

MinerU2.5, a 1.2B-parameter document parsing vision-language model, achieves state-of-the-art recognition accuracy with computational efficiency through a coarse-to-fine parsing strategy.

the trained 4B model exceeding GPT-4.1 (49.4 percent) and GPT-4o (42.8 percent) despite being 50 times smaller

In 23 months, the same capability that needed 1.8 trillion parameters now fits in 4 billion parameters. A 450x compression.

Gemma 4 outcompetes models 20x its size

个人学习可能取决于他人行为的主张突出了将学习环境视为一个涉及多个互动参与者的系统的重要性

Better contribution workflow: We will be using GitHub’s contribution tools and features, essentially moving MDN from a Wiki model to a pull request (PR) model. This is so much better for contribution, allowing for intelligent linting, mass edits, and inclusion of MDN docs in whatever workflows you want to add it to (you can edit MDN source files directly in your favorite code editor).

“My feeling at the time was this was a good idea,” Dr. Wachter says. “The trend toward our system being pushed to deliver better, more efficient care was going to be enduring, and the old model of the primary-care doc being your hospital doc … couldn’t possibly achieve the goal of producing the highest value.”