Remember: K2 is *not* a reasoning model. And very few active tokens in the MoE. So it's using less tokens, *and* each token is cheaper and faster.
LLMS
-
Behavior differences between Grok 3 and Grok 4 Heavy
By
–
That’s Grok 3. This thread is about Grok 4 Heavy. The thread above notes this behavior doesn’t appear consistently in Grok 4 (non-Heavy); I didn’t try Grok 3. See the post below. Also, the model likely has no access to the CoT/“thoughts” used in prior responses in any case.
-
Separate Few-Shot Examples from Main Prompt for Better Performance
By
–
My main issue with this prompt is having it build few-shot examples in the main prompt. It is cognitively already doing a lot of heavy lifting and I think any examples should be generated in a separate pass. Otherwise it's not bad.
-
Framework for Recreating Grok Heavy Functionality Across Models
By
–
Releasing this tomorrow: A framework for recreating Grok Heavy functionality.
— Pietro Schirano (@skirano) 14 juillet 2025
In this video, I'm using Claude 4 Heavy, but it works with any model (GPT, Kimi, Gemini). pic.twitter.com/v5zLxTltN3Releasing this tomorrow: A framework for recreating Grok Heavy functionality. In this video, I'm using Claude 4 Heavy, but it works with any model (GPT, Kimi, Gemini).
-
LLMs can leverage external tools like chess engines for task performance
By
–
Y en realidad sí podría porque si se acepta usar software como el que usaría la Atari para ganar consistentemente, nada impide al LLM buscar en internet, instalarse un módulo de ajedrez en su sistema y usarlo como herramienta. Es el problema de creerse los argumentos de Marcus.
-
Model Refusal Rate Analysis and Prompt Engineering Support
By
–
Nope, it cannot be training because this behavior is tracked and the refusal rate is the lowest in its size category. I can help more if you send me the prompts.
-
@testingcatalog — 2025-07-14
By
–
And Mistral Let's see how soon we will start hearing about Chinese mega clusters for training closed models
-
MIRIX: Multi-Agent Memory System for LLM-Based Agents
By
–
MIRIX: Multi-Agent Memory System for LLM-Based Agents Yu Wang, Xi Chen: https://
arxiv.org/abs/2507.07957 #ArtificialIntelligence #DeepLearning #MachineLearning -
Anthropic unveils curated MCP connectors directory for Figma, Notion, Stripe, macOS
By
–
BREAKING 🚨: Anthropic released Connectors directory with loads of curated MCPs.
— 🚨 AI News | TestingCatalog (@testingcatalog) 14 juillet 2025
Figma, Notion, Stripe and loads of other connectors are now available, including desktop-specific MCPs for the Claude macOS app. https://t.co/ZxXRobIo1p pic.twitter.com/m9vEzy8jwwBREAKING : Anthropic released Connectors directory with loads of curated MCPs. Figma, Notion, Stripe and loads of other connectors are now available, including desktop-specific MCPs for the Claude macOS app.