A 14B model just beat a 671B model on math reasoning.
Here’s how Microsoft’s rStar2-Agent achieves frontier math performance in 1 week of RL training
- by “thinking smarter, not longer.” 🧵
Here's what I learnt from the paper “Small Language Models are the Future of Agentic AI” by NVIDIA Research:
Relying on LLMs to build prompt-based agent systems is not the future.
Instead, consider building your SLMs and integrating them into agentic AI systems
It will make
That's the Mohammed bin Rashid Al Maktoum Solar Park. The world's largest single-site solar park. They want to be the leader of renewable energy in next era.
Here is my quick note on this paper for people who want to learn more:
- Energy-Based Transformers (EBTs) enable transformers to “think” during inference, not just predict. It simulate System 2 reasoning by refining predictions step-by-step through gradient descent.
Why This
I guess one surprising thing I found about GLM 4.5 is how quickly it processes a quite challenging prompt and starts generating.
I tested various game/animation prompts through AnyCoder, and the generation speed is overall fast, with high accuracy.
e.g. in this example I saw
Most current AI agents cannot perform fully autonomous tasks like "taking a request, browsing the web, and physically cooking a meal"...
Then this paper introduces "Embodied Web Agents," a new AI paradigm that interacts with the real world and web information simultaneously.
While roy is getting a lap dance, some are actually hustling and focus on the future of AI.
It is for SLM and LLM,
🌟Small Language Models are the Future of Agentic AI ⬇️
Here's what I learnt from the paper “Small Language Models are the Future of Agentic AI” by NVIDIA Research:
Relying on LLMs to build prompt-based agent systems is not the future.
Instead, consider building your SLMs and integrating them into agentic AI systems
It will make
Yesterday, I made and deployed a cool NFT collection on the Base blockchain.
It is called "StarCeiling" - surreal, inspiring, chill style. (just like my profile avatar)
What do you think about this style of design?
If you like it or want one, I'd be happy to know.
What’s good? Guys! Happy to join in this community and hope we could grow and support one another.
Hope we could connect here. Let me know and I will follow back!
Most current AI agents cannot perform fully autonomous tasks like "taking a request, browsing the web, and physically cooking a meal"...
Then this paper introduces "Embodied Web Agents," a new AI paradigm that interacts with the real world and web information simultaneously.
