3. LatentMAS LatentMAS introduces a framework enabling language model agents to collaborate directly within a continuous latent space rather than relying on text-based communication.
@dair_ai
-
HunyuanOCR: Lightweight 1B Parameter Vision-Language Model
By
–
2. Lightweight End-to-End OCR HunyuanOCR is a commercial-grade, open-source, lightweight vision-language model with only 1B parameters designed specifically for OCR tasks.
-
INTELLECT-3: 106B MoE Model Achieves State-of-the-Art Performance
By
–
1. INTELLECT-3
— DAIR.AI (@dair_ai) 30 novembre 2025
A 106B-parameter Mixture-of-Experts model (12B active) trained with large-scale reinforcement learning, achieving state-of-the-art performance for its size across math, code, science, and reasoning benchmarks.https://t.co/8tphkxnVzy1. INTELLECT-3 A 106B-parameter Mixture-of-Experts model (12B active) trained with large-scale reinforcement learning, achieving state-of-the-art performance for its size across math, code, science, and reasoning benchmarks.
-
Top AI Papers of the Week: Latest Research Highlights
By
–
Top AI Papers of the Week (Nov 24 – 30): – LatentMAS
– INTELLECT-3
– OmniScientist
– Lightweight End-to-End OCR
– Evolution Strategies at Hyperscale
– Training LLMs with Reasoning Traces
– Cognitive Foundations for Reasoning in LLMs Read on for more: -
Swarm Intelligence: The Collective Learning Secret Revealed
By
–
Swarm intelligence has a secret. The standard approach treats swarms as collections of independent learners. Each agent makes decisions, learns from outcomes, and coordination somehow emerges. More agents, more complexity, more mystery. But what if the swarm itself is the
-
LLM Benchmarks Misleading: Fixed Prompts Underestimate Model Capabilities
By
–
Lots of LLM benchmarks are misleading us. The problem isn't the benchmarks themselves. It has more to do with the prompts. Current evaluation frameworks use fixed prompts that systematically underestimate what models can actually do. Pay attention to this one, AI devs! This
-
Learn to Build with AI Agents Academy Discount
By
–
No better time to learn how to build with AI agents. Our academy is now 30% off for Pro subscription and live cohort-based courses. Use code BLACKFRIDAY at checkout.
-
NVIDIA Research: Efficient AI Beyond Model Scaling
By
–
Banger paper from NVIDIA. Bigger models aren't always the answer. However, the default approach to improving AI systems today remains scaling up. More parameters, more compute, more cost. But many tasks don't require the full power of a massive model. This new research
-
Multi-Agent Communication Costs: Beyond Token Overhead
By
–
Multi-agent systems are powerful but expensive. However, the cost isn't in the reasoning itself. It's in the communication. Agents exchange full text messages, consuming tokens for every coordination step. When agents need to collaborate on complex problems, this overhead adds
-
Building Multi-Agent Systems with Vision Capabilities
By
–
LLMs can't see. How can we build effective multi-agent systems with vision capabilities? Building multimodal models from scratch is expensive. Training joint vision-language architectures requires massive compute, specialized datasets, and careful optimization. But there's