Essential AI

I did a project with some of the Adept people. Very disorganized and chaotic. There were definitely people pulling in different directions. Made it very difficult for me and I'm glad it's over.

You’re not convinced by “our technology will make data analysts 10X faster and give business users the tools”?

They have LLM mentioned on the front page. Whatever they're building it's 10x better with LLM.

Pre-vision, pre-team, pre-product, and pre-revenue -- An opportunity to build the future in this blank canvas with the founding team.

Who promised that AI could/would solve climate change or cancer? The focus at the moment is on the enterprise because that's where its current capabilities are useful.

There is no money in science customers and lots of money in enterprise customers.

Not that I know of. AI is just about making as much money as possible, as quickly as possible.

I'm finding it hard to understand what you are after.

Is it given a researcher, P(second breakthrough paper|first breaktrough paper) for some definition of breakthrough?

Most breakthroughs are discovered accidentally and retroactively, so I'd think that having multiple breakthrough papers is fairly uncommon.

One could argue that someone who has published 3 breakthrough papers is more likely to publish a 4th simply because they must have deep understanding of the field to have published 3 breakthrough papers.

It is obviously not a guarantee, but as with many things in life, previous success does tend to favour future success.

Currently working at Google? It's more likely that you will work for Apple in the future compared to the average developer. Got good grades in physics, you will probably get good grades next term too when compared to your average class mate. Got a startup that focuses on some practical use of LLMs, much more likely to get funding than if it focuses on some practical use of Markov models.

Well whatever it is, it'll be written in Java to be sure.

It means cold sweet investment cash, that's what it means.

Seems like a commercial for a startup on HBO's Silicon Valley.

Have to disagree. LLMs in their current form are not the end all be all of AI architectures. Seems like the co-authors of the original transformer paper are in a good position to find other valuable applications of the results described in the paper. They are also probably in a good position to see what other advancements are on the horizon that can be productized. Not sure why you would write them off as being too late to the game as though the game is already over. It's only over if you think that grabbing mind share on the current LLM hype train is the goal.