gradys

M1 Pro Max is only $200 more. I'm tempted, but do we think it will be more power hungry under the same workload than the Pro?

Agreed on Markdown's limited expressiveness.

I think the reason for programmers' attachment to Markdown isn't just about being able to operate on it in a terminal. It's about the general interoperability of plaintext.

Terminal editing is one example, but even IDEs and non-terminal text editors also use plaintext. There's also the fact that version control tools are designed for line-based plaintext. Markdown is especially often used for documentation that gets checked in alongside code, so this is an important consideration.

I would be very interested to hear thoughts on good alternatives to Markdown for technical documentation, even if not plaintext-based.

Are we doing that? Was some new gas tax passed in the US?

Node compatibility mode seems like a big deal. I wonder how much this helps with library compatibility practically speaking. They don't say much about any limitations in the notes, just that it will get better in the future.

Probably not. It's not clear how to scale other methods to make use of so much compute. They top out sooner, with fewer parameters and less compute.

One way to look at why deep learning is having the impact it does is that unlike other ML methods, it's actually capable of making use of so much compute. It gives us modular ways to add more and more parameters and still fit them effectively.

I've wanted to try out sorting neural net dimensions for visualizations via a travelling salesman problem solver over the correlation matrix among the dimensions. Has anyone tried something like this?

This is Spring's "erring on the side of caution" trademark enforcement mechanism, not trademark law itself. I highly doubt that a court would acknowledge that Affinity has some trademark claim over all short strings of Greek letters.

A library to do this that integrated well with typical inference and and training systems would be a great contribution!

Your post is obviously trying to fan flames, but I'll point out that many people in the US, including, I suspect, most Americans on HN, did not actually think "America first" rhetoric was good.

Setting aside that the Economist is not actually a US publication, are you of the view that it's good for Cuba that at the same time that they're rationing food, so much of the country's farmland is in low productivity state-run farms?

You sound very confident about this, whereas copyright lawyers I've read discuss this issue seem much less confident overall, but lean toward thinking this would be fair use.

What makes you so confident that this would not be ruled fair use?

(And for people not familiar - if ruled fair use, it doesn't matter what the license is because fair use is an exception to copyright itself.)

Could you explain why you think training models on copyrighted text is illegal or copyright infringement or whatever else it might be?

On the last point:

1) The vaccine isn't perfect, and herd immunity nonlinearly reduces the risk of everyone including the vaccinated from getting it.

2) Unvaccinated people, who make up the vast majority of the cases, create an environment for the virus to evolve within, possibly mutating into a form that is more dangerous to everyone, including the vaccinated

It's quite possible to do this on-prem and even on-device. TabNine, a very similar system with a smaller model (based on GPT-2 rather than 3), has existed for years and works on-device.

The AI most certainly does not lack the ability to generalize. Not as well as humans, but generalization is the key interesting result in deep learning, leading to papers like this one: https://arxiv.org/abs/1710.05468

The ability to generalize actually seems to keep increasing with the number of parameters, which is the key interesting result in the GPT-* line of work that Copilot is based on.

There's an interesting discussion raging on elsewhere in the thread about the possibility for cultural adaptation that would enable us to cope with statements like "the cancer test was positive, but there's still a decent chance you don't have cancer".

There's also an interesting question, taking your roughly 1:1 true to false positive ratio as an example, of whether the marginal true positive does more good for the world than the marginal false positive.

Side-stepping those questions though, they probably won't give this test to literally every person in the population. How much does the picture improve when you give this only to people over 50? Or only smokers? You could still massively increase effective screening for cancer with a cheap and easy test if you combine it with enough population filtering to increase the true positive/false positive ratio to something like 70-90% rather than the 50%

One way to analyze the two similar function situation is to ask whether a future change in the logic to one would imply that a change should also made to the other.

If so, then by keeping them separate, you introduce the risk that you would forget to update the other. Using a single function and calling it twice, even with different parameters, is a way communicating that it's important that the semantics be tied.

If a change in one wouldn't imply a change in the other, then you really have two separate semantic roles being played that happened to resemble each other in implementation - harmless incidental duplication.

I don't think that, no. Still, different people with different toolboxes will come up with different solutions.

But in doing so, you bring that latent knowledge back into the minds of living people who can rethink the solutions in the face of the modern ecosystem and clean out the cruft that truly has no modern purpose.

It's easy to use templates to convert some given knowledge graph node or maybe subgraph into a paragraph that could maybe serve as a Wikipedia intro paragraph.

It's much harder to generate answers to questions. This calls for jointly choosing what knowledge to use in the answer and synthesizing text that presents that knowledge in a way that actually answers the question. This work is about this more dynamic problem.

This has very much not been my experience in the US.

Would Ethereum be able to make the switch to PoS if it had never used PoW originally? How would they ensure a good distribution of ownership?

Maybe just a Python notebook with a Postgres client library and some helper functions to keep the amount of Python in the main body to a minimum?

What is ridiculous exactly? That it's deflationary? Or the intuition that it should rise in the face of USD inflation (all else equal)?

You'd think an intrinsically deflationary asset like Bitcoin would move differently in the face of USD inflation. Any explanation for why this might be?

Shout out to the attrs library. If you find yourself using namedtuple a lot, it's worth considering despite it being an external dependency.

https://www.attrs.org/en/stable/

I don't have comment on the legacy of supply-side economics and and how successful e.g. Reagan's economic policy was, but it looks to me like we have more of a supply-side issue right now than demand-side.

We have shortages of semiconductors, congested ports, all of the supply issues with early COVID PPE, absurdly low housing supply in the most desirable places to live, and so on.

Why is the no diamond constraint useful here?

Seems like you should sue?

It works much better when using immutable data and mostly side-effect-free functions as is heavily encouraged by Clojure.