Research
5 articles on this topic, newest first.
2026
- Not all AI models are LLMs: seven model families and how they differAutoregressive LLMs, reasoning models, diffusion LLMs, Mamba hybrids, mixture-of-experts, world models and robot VLAs, explained by what each one predicts and how it spends compute.14 min
- JEPA, explained: Yann LeCun's bet against next-token predictionWhat a Joint Embedding Predictive Architecture is, how I-JEPA and V-JEPA 2 work, why LeCun thinks it beats LLMs, and what the evidence shows so far.16 min
- Fast and slow: System 1 vs System 2 AI models, explainedWhat people mean by System 1 and System 2 AI, how reasoning models like o1 and R1 learned to think slowly, and why frontier models now decide for themselves.16 min
- AI wrote a working genome. Here is what that actually meansGenome language models designed 285 bacteriophage genomes. Sixteen came alive and killed E. coli. What the result means, and the safety fight around it.12 min
- Your AI has amnesia: why smart models still can't learn on the jobContinual learning, explained. Why frontier models forget everything between sessions, what catastrophic forgetting is, and the research trying to fix it.12 min