Notes on LLMs

. LLMs' intelligence is different from human intelligence. Evolutionary forces that forged our homo sapiens brains are not what created LLMs.

. Instead, LLMs' training goal is to learn humanity's collective thinking patterns. This fundamentally makes them imitators. This doesn't necessarily mean they are better or worse just that their mental faculties are destined to be different.

. OpenAi's conservativism and hamstringing of their models may be leaving some space for edgier competitors.

. In general, no one knows what exactly will happen. We could be at the beginning of an exponential lift-off or we may meet a slow and anticlimactic AI winter.

. If relative stagnation occurs, smart iterative engineering alone will be enough for massive disruption. We won't need GPT-5 for that.

. A couple of years from, I expect to see another breakthrough in LLM architecture, let's call it XXX. It will be a step change from GPT. Another year before it gets launched publicly.

. XXX architecture will make even the calmest normies shit their paints. Governments will be pressured to act swiftly.

. Five years from now: the first AI ban/strict regulation. Fear of falling behind might result in a half-measure. Progress will not halt but researchers might need a license and government oversight.