[croissants]

Previous discussion: https://news.ycombinator.com/item?id=22101066

IMO the confident Tweets about how to become a good researcher and hire good researchers look pretty weird next to the lack of any apparent research papers (or even visible research products) five years later.

[king_magic]

Guy seems & speaks like a charlatan.

[Xmd5a]

>I moved from Sweden to Ghana, West Africa. I started working as a teacher in the countryside, but after invoking the spirit of their dead chief, they later annotated me the king of their village.

[RestartKernel]

For those casually browsing the comments, this is an actual quote from the article.

[emilwallner]

True story, happened when I was around 18: https://imgur.com/fNd55hB

[dkarl]

I knew a guy like this. He was fairly transparent about the fact that he talked a lot of bullshit about himself, in the sense that he didn't keep it realistic and didn't respond to reasonable skepticism the way a normal person would, but a lot of people responded positively to it. The people it worked on, they recognized that he wasn't speaking literal truth and mentally discounted what he said by some large percentage, thinking that they were critically adjusting for his boastfulness. But he painted such a grandiose picture of himself that anyone who subconsciously assumed that 10% of it was true would still be impressed.

He was a weird dude. I think he grew up in a household where outrageous boasting was common and had been doing it so long that it was his authentic self. He desperately wanted people to be impressed with him, and he had mastered a strategy that had a consistent yield, but the people it didn't work on looked down on him pretty severely, and he didn't have any way of earning their respect. He had to trust in the fact that over time he would tend to find himself surrounded by people he could manipulate.

[emilwallner]

I wrote a tldr here: https://news.ycombinator.com/edit?id=42976187

[sepositus]

I stopped reading at this point. Is there some cultural context I'm missing here? What does that even mean?

[gwern]

Isn't he running a company (Palette) which makes heavy use of DL?

[emilwallner]

Indeed, I realized it's hard to compete with PhD students for grants, and subsidising my work with content marketing does not fit my style, and I prefer owning my work and choosing my own research direction. I also want people to use my work, and create solutions that are cost-effective.

So the most logical way was to bootstrap an AI start-up in the area I'm interested, so that's what I'm doing. Unfortunately, it's hard to publish or contribute to open-source, since it becomes too easy to copy, which cuts my margins and ability to fund my research and compute.

Now I spend most of my days doing AI research, and outsource most other parts, really enjoying it :)

[gopher_space]

I think you'd attract a more sympathetic audience if you reframed Emil's Story as "How to get people to leave you alone and let you think." E.g. half of your AI autodidact degree boils down to 'make money with what you\'ve learned', and it's always interesting to see people unpack their approaches to this idea. Your thinking around target numbers would be valuable.

[emilwallner]

I agree, it was written some time ago, and the language is hyperbolic, but many of the key points still hold true. Building a skillset to reach $100/hour consulting gigs was key to work 2-3 days per week, have time for research, and saving money for an ML rig, while living in Paris. Around $5K MRR is sufficient to live off, $10K MRR is more comfortable and allows renting a few extra A6000 Ada and start outsourcing, and $20K MRR to afford full outsourcing to focus on research and renting 8xH100s. That's a good goal to balance research freedom with other responsibilities. After that, it's optional to trade research freedom for higher MRR and more responsibilities.

[croissants]

Sure, so it seems fair for him to offer advice on how to apply AI or start a company based on it, but doing AI research means generating new knowledge about AI itself, and I don't see any evidence of that.

[numba888]

> doing AI research means generating new knowledge about AI itself, and I don't see any evidence of that.

Wouldn't colorization count as research? In vision domain there are a lot of papers like this. Just arranging and rearranging known blocks and getting SOTA result on some datasets. ;)