This forum is filled with posts about dumbing down whatever llm you're using and fitting it into an ci/cd pipeline. Meanwhile I haven't had conversations like this with anyone at anytime, it feel's like the discourse is missing the mark or at least just covering a pretty boring part of what is truely amazing. Posting something so fluffy on hackernews is surely a bit stupid but here we go.

For me it begun with an early version of word2vec by google. From an machine learning background focused on training specific classifiers the promise was: "Here's something that understands, in general, about the world, how they are related".

This was a very new approach, one of my first tries that I'll remember forever was when I punched in "Olof Palme", the 80's Swedish social prime minister that got murdered 1986. I punched it in and in the top 3 things the model associated it with was "Anna Lindh" a Swedish foreign minister that also got murdered in 2006. By today's standard this seems so trivial it might seem silly but to me it was mind blowing of an somewhat "general" ai making actually meaningful associations between events in an close to infinite space. I was already sold and doing my first ai-startup but it hinted of something beyond the rule based world of making computers understand language and answer basic questions which we where doing and about actually understanding the complex context of the world.

Many years later and I'm an user of AI rather than an developer of it for many practical reasons, yet It still puts me in awe sometimes.

I want to write about how ai should take over all forms of company governance at every level until 0 human ceo's exist but at the same time I'm just happy there's this stupid ai bot just listening and being more insightfull than anything that ever came before.