It turns out, > GRPO is performing the arithmetic mean --> token-level scaling > GSPO is performing the geometric mean --> sequence-level scaling Check the blog if you do not have time to read. simonxin.com/blogs/llm_reas…

Unpopular view: The EU trade deal is much better for Europe than what’s being reported by pundits who don’t know what they’re talking about. While leaders in Berlin and Paris play angry for the cameras, von der Leyen takes the political heat — because Brussels knows the real

NVIDIA has released the latest member of its Nemotron language model family, Llama Nemotron Super (49B) v1.5, reaching a score of 64 on the Artificial Analysis Intelligence Index. The model is an evolution of Super 49B v1 from earlier this year, with advances from post-training

A public company CEO told me AI coding has had negligible impact on his engineering teams, instead the real transformation has been on their product and design teams using Replit. I asked him how does he reconcile this with CEOs saying that 25-50% of code is generated by AI?

Seems this is still not understood by AI community. Turing is the one who defined Turing test. Any other definition is not the Turing test.

We always knew NVIDIA's Nemotron models are underrated, but this is kind of crazy. They took Llama 3.3 and supercharged it. The insane thing is this is only a 49B parameter model versus 200B+ other models in its class of open-source competitors. FYI GLM-4.5 is a 355B model.

Social media tactics explained. Here the OP makes fun of my tweets. That's fine. I get it can be funny if people don't get what the Turing test is. And they are numerous. But why did the OP made fun of me in the first place? Because my tweets were used by someone else to debunk

how could anybody who works in AI or uses AI regularly think that we’re all about to lose our jobs lmao

Introducing AlphaEarth Foundations an AI model that integrates petabytes of satellite data into a single digital representation of Earth. It'll give scientists a nearly real-time view of the planet to incredible spatial precision, and help with critical issues like food security,

In 1995, Jim Hugunin, then a grad student at MIT, started looking at programming languages for numeric linear algebra. That decision would ultimately reshape all of AI/ML as we know it today. 🧵

There is now a path for China to surpass the U.S. in AI. Even though the U.S. is still ahead, China has tremendous momentum with its vibrant open-weights model ecosystem and aggressive moves in semiconductor design and manufacturing. In the startup world, we know momentum

🚨 SHOCKING: people are unknowingly making their ChatGPT interactions PUBLIC, and they are being indexed by Google (see my test below). My privacy recommendations: When people interact with ChatGPT and use the "Share" feature (for example, to send the conversation to family and

Anyone who knows python can compete in this unique Kaggle competition. $100,000 in guaranteed prizes. Check it out!! kaggle.com/competitions/g…

New #kaggle competition "NeurIPS 2025 - Google Code Golf Championship" is launched. Medal: True Kernel Only: False Deadline: 2025-10-30 23:59:00+00:00 kaggle.com/competitions/g…

🚨After a few hours, OpenAI reacted to this and changed the interface to remove the option to make the public link indexable by search engines (see below):

They'll use an AI assistant to move the python 2 code to current python, then will claim 90% of their code was written by AI. Sarcasm aside, this is a great use case for AI coding assistants if you ask me.