But annoyingly sometimes this is great advice. It's not that dissimilar to YC's make something people want, but it relies on you having taste/needs/wants that are an addressable market. Which is sometimes true.
Sounds to me like he made a couple unfortunate moves:
- Propose a 5 year plan (lol)
- Don't lead the incident with leadership, but with blame
- Speaking about problems more than solving problems (hard to do)
- Lack of relationship building
On one hand, I empathize with Chris. On the other hand, this sounds to me like he just didn't know how to perform in this environment. And that's totally fine! Not everything should learn how to perform in the bureaucratic knots -- startups are simpler in this way. And there's a reason the big guys lose their edge over time, and then some exec in a board room is faced with -10% YoY loss without any truthful VPs around the table.
Given the model performance is thus affected by a k-nearest neighbor, but those algorithms are proving not great for baseline vector search, how well will this actually work?
It seems mostly like a vertically integrated vector DB + existing LLM call, but correct me if I'm wrong. There are of course some performance gains with that, but the holy grail of "understanding" at unlimited length still seems unsolved.
Chromium is great. Why exactly should they innovate their first? A v1 should take whats available and not seek to reinvent the wheel.