Failure modes on the quest to general, open-source LLMs. Expect pivots to specialized models.
As someone who spent a couple of years in the blockchain space, particularly during the last hype cycle your points resonate with me and remind of something very similar happening in that industry as well. Every company trying to prove they are a capable team would build their own protocol, sometimes the entire new L1 chain. The formula is the same, you need to raise a gargantuan amount of money to fund your protocol and build an ecosystem around it. The moat is very hard to keep, because the next capable group of engineers is doing the same the next day for as long as the VCs keep funding these companies on promises to find traction / business use case in the future. The technology is cool, but it turns out that building strong defensible businesses is hard and sometimes it has little to do with the lines of code or the performance of a particular model.