If you think an LLM is a thinking machine, I don’t know what to tell you. They are great at predicting tokens, but there is no evidence they are thinking in the human sense. They rely on patterns and probabilities, not understanding. Powerful search engines, certainly, but there is much work ahead to see how hard the limits being bumped up against are.
> As someone that has sold a bunch of LLM enabled software over the past 6 months, I don’t really buy the AI capex turning into huge productivity gains. Everything to date are just chatbots with RAG and API calls. None of them are going to do my laundry or file my taxes.
Without getting hung up on the definition of the word "think", the fact that anyone can feed an LLM a couple of bullet points and generate a pile of convincing sounding marketing copy, is worth something if you're to someone. We can disagree on exactly how much and to whom, but if you're not willing to see that as fact then there's no common ground to base a conversation on.
It's under appreciated that some jobs being trivially done by LLMs doesn't mean that all jobs are trivially done by LLMs.
Low-value-add marketing copy, for example, being automated changes... what? Now LLM-generated copy is the new minimal-cost baseline, and everyone adapts to the new normal.
The real killer feature will be autonomous business planning, but we're a long way from there.
It's hard to take your point of view seriously after that honestly.