presheaf

That's great. Why don't they produce the next one if they're so good at it?

Can't be done. The state is the one that produces the chips (Intel is just another branch of the government). Software is just a layer on top that can always be subverted with hardware by someone who knows the sequence of operations that will give them access.

I figured. Good luck on becoming one.

Not everything is indexed. This isn't documented anywhere.

Going down this path then you might as well become a cyborg.

We can solve these problems because "Either a given technology is possible, or else there must be some reason (say, of physics or logic) why it isn’t possible". Knowledge is what allows us to develop solutions to all our problems.

This is very much like the law of excluded middle, (there is a solution) ∨ (there is no solution). The point here I think is that there is no reason to not try to solve problems because either the problem is solvable and we make progress or it isn't and we figure out why which is again a kind of progress. This is certainly a reasonable perspective but Hamming and Heisenberg have quotes that provide more nuanced perspectives.

Hamming: Just as there are odors that dogs can smell and we cannot, as well as sounds that dogs can hear and we cannot, so too there are wavelengths of light we cannot see and flavors we cannot taste. Why then, given our brains wired the way they are, does the remark, “Perhaps there are thoughts we cannot think,” surprise you? Evolution, so far, may possibly have blocked us from being able to think in some directions; there could be unthinkable thoughts.

Heisenberg: What we observe is not nature itself but nature exposed to our method of questioning.

Money should not be the primary goal, but rather a by-product of pursuing your passions.

This is very wise.

Watch a lot of movies. It's a surprisingly effective way to learn.

This is interesting. Recently had the idea of making up an instruction set for bit-strings so that I could generate a bunch of programs w/ the instruction set to compress bit-strings where short instruction sequences would get high scores and longer ones would get low scores.

The tricky part is designing the feedback loop to properly train the instruction generator and like in this paper I needed to also include some non-differentiable stack operations. It's surprisingly hard to find work that combines neural networks and compression algorithms even though they seem like an obvious fit. This also allows for downstream tasks that are not possible with just vector spaces because the network that can compress bit-strings must be encoding some non-trivial features of the data set and can be used to augment downstream tasks with differentiable compression.

This is very cool.

That's a good point but that's the case with all fraud because the digital tokens have to be converted back into fiat at some point. FTT was created without much oversight because no one had enough experience to properly audit anomalous activity. Presumably someone can trace all the fraudulent FTT transactions because it's all public but it doesn't seem like anyone has actually managed to do that. Everyone knows the money is gone but no one knows how that happened and through which FTT transactions.

Just curious how the dominoes will fall but it does seem like the people that invested in cryptocurrencies underestimated the risks of corruption. It looks like programmable money and clever financial engineering makes embezzling money a lot easier.

Good point.

Looks like Paul Graham's friend was correct: https://mobile.twitter.com/paulg/status/1594446009010212865.

This is my usual recommendation: http://www.aosabook.org/en/index.html.

One nice aspect of z-lib was that they would provide recommendations for books on the download page. I found a bunch of books on category theory that I wouldn't have found any other way.

Not sure what they were using for the recommendations but the algorithm was much better than other book recommendations, e.g. Amazon's "other people also bought this".

AI training as a service. Seems like whatever expertise OpenAI has developed for managing workloads for AI is now being rolled into Microsoft to be offered as another Azure service. Are there any existing companies that offer a service like this? If someone doesn't want to manage the distributed training infrastructure but they have a large data set and want to distill it into a model to be used for inference then seems like if someone figures out how to automate the process then they can make a lot of money. Lots of enterprises have a lot of data and would prefer to use a distilled model of the data instead of having to manage the entire infrastructure for the training loop.

Seems like a viable startup opportunity.

I remember a discussion in the mid-eighties about the future of artificial intelligence in which for some reason I was asked to participate. The advocates of what was then called “hard A.I.,” to the dismay of their opponents, were painting a triumphalist picture of the future of computer intelligence. As the discussion went on, all semblance of logical argument was given up. Eventually, everyone realized that Jack had not said a word, and all faces turned toward him. “Well,” he said after a pause, “some of these developments may lie one hundred Nobel prizes away.” His felicitous remark had a calming effect. The A.I. people felt they were being granted the scientific standing they craved, and their opponents felt vindicated by Jack’s assertion. - Gian-Carlo Rota

It means they have significant market share in the relevant sector they're operating in. For example, it would be fair to say that Tableau is a leader in enterprise analytics and data management.

I'd say learn about software architecture patterns and how to manage complexity. A good starting point is http://www.aosabook.org/en/index.html. Most backend engineering is about managing complexity and that book has a lot of good information on how to do that.