
Superintellignce is still decades awayIn 2022 - I thought that we got closer to AGI with LLMs, but it turned out LLMs were more like the engine of intelligence than intelligence itself.In 2023 - I started to understand how important memory is to the system, and the intelligence is a storage problem as much as it is a processing problem.In 2024 - I started to realize that AGI is the wrong term all together and that intelligence can't be solved with scale alone.In 2025 - I started to understand just how many components needs to work together to build a real agent. We started calling them agents because they have many different moving components nothing else described them. RAG became obsolete as it was too rigid to keep up. Not only we needed an LLM, memory & orchestrator, but also perception & RAI.Now in 2026 I am starting to understand just how complex intelligence is, which I would have understood sooner if I took more biology courses in uni. I remember an episode of "love, death & robots" from 2022, in which a human scie
