According to this research it sounds like Claude uses the same tokenization trick I invented for ULMFiT in 2017! 😀
RESEARCH
-
AGI Self-Improvement Could Rapidly Lead to Superhuman Intelligence
By
–
Most people think that AGI simply means having multiple systems at human level. But once an AI can improve itself, it quickly becomes superhuman. And from there, we have no idea what it might look like. It's like creating an ultra-intelligent mind… that might perceive the
-

ChatGPT Reaches Final in Elite Programming Competition
By
–
ChatGPT a surpassé presque tous les programmeurs d’élite (un seul a résisté). (C'est l'élu, il représente l'humanité ) Dans une compétition centrée sur les algorithmes d’optimisation, le modèle d’OpenAI a atteint la finale le 16 juillet, a pris la tête… avant d’être dépassé
-

Google SensorLM: Multimodal Sensor-Language Foundation Models for Health
By
–
Google Research introduced SensorLM, a family of sensor-language foundation models trained on ~60M hours of data
— The Rundown AI (@TheRundownAI) 29 juillet 2025
They connect multimodal wearable sensor signals to natural language for a deeper understanding of our health and activitiespic.twitter.com/yaiWp8RrtyGoogle Research introduced SensorLM, a family of sensor-language foundation models trained on ~60M hours of data They connect multimodal wearable sensor signals to natural language for a deeper understanding of our health and activities
-

Artificial Analysis Intelligence Index methodology critique benchmarks
By
–
I like that Artificial Analysis is open about how they evaluate models and makes data public, it is a real service. However, I see folks citing their Intelligence Index as a metric without realizing it is an average of the same correlated, semi-saturated benchmarks everyone uses
-

GLM-4.5 Dominates 12 Major AI Benchmarks Over Competitors
By
–
Benchmarking Brilliance
GLM-4.5 excels across 12 major benchmarks, outperforming competitors like Claude 4 Opus and DeepSeek R1. Its open-source nature and advanced capabilities position it as a formidable force in the AI landscape. -

Zai Unveils GLM-4.5: Advanced AI Model for Autonomous Reasoning
By
–
🚨A Chinese research lab, Zai, has unveiled a powerful new model GLM-4.5
— Futurepedia – Learn to Leverage AI (@futurepedia_io) 29 juillet 2025
Not just another AI model—it's a leap into the next era of autonomous reasoning, coding, and agentic intelligence.
With a hybrid architecture and 128K token context, it's engineered to think, plan, and… pic.twitter.com/NVZP1dNg1fA Chinese research lab, Zai, has unveiled a powerful new model GLM-4.5 Not just another AI model—it's a leap into the next era of autonomous reasoning, coding, and agentic intelligence. With a hybrid architecture and 128K token context, it's engineered to think, plan, and
-

GLM-4.5: Efficient 355B Parameter Language Model
By
–
GLM-4.5: Power Meets Precision
Boasting 355 billion parameters with 32 billion active at runtime, GLM-4.5 delivers top-tier performance at a fraction of the cost. Its Mixture-of-Experts design activates only the necessary parameters, ensuring efficiency without compromising -
AI Factual Gullibility: Testing Model Resistance to False Claims
By
–
I would love to see more work on AI factual gullibility. Minor falsehoods are easy, but I have been trying to convince models that The Bronze Age was a hoax (tin deposits weren't located anywhere close to copper, etc.) and so far it hasn't come close to working on modern AIs.
-
Context Rot and Token Efficiency in Large Language Models
By
–
Yeah context rot is about bad tokens (mistakes etc) getting into the context and causing poor performance later on – what Theo is describing is more models straight up wasting tokens