mpoteat

Reading the question, I thought the answer was six as well. Such a question is so misleading that I would hope that an AI answers as it has in this case, because not all problems are fully logically specified, and you need to remain robust against the question being posed incorrectly.

In other words, I would posit the probability of you having meant for the answer to be 5 low, because the question itself becomes trivial. For a system that needs to deal with many people asking questions, this robustness is helpful in my view.

In other words, it's not totally clear that time has not passed between sentences in your prompt. So the second instance of "now" could be a different "now".

As well, you should be creating regional servers to minimize latency for folks in other geographic regions. Can't beat c!

Let's spend multi-million dollars a year on a team of highly specialized FPGA engineers writing assembly and HDL so that we can save 5k a month. Feature velocity will be 100x slower as well, but at least our application is efficient.

I think that this may make sense for some applications, but I also think that if you can utilize software abstractions to improve developer efficiency, it reduces risk in the long run.

You don't need to prove anything in court to send someone a cease and desist. It's often cheaper to settle.

There does exist the concept of 'originality' in copyright law, which I was erroneously conflating with length.

This is the code of the future to ensure your code remains original: https://twitter.com/TylerGlaiel/status/1611115741809627139

If this lawsuit succeeds, I have a startup idea that I think would be effective.

Create a for-profit copyright registry for code snippets that are long enough to qualify for copyright protection. You can be the canonical owner of the copyright for a given piece of code! For a premium fee, we can generate and submit a patent on your behalf as well.

Once I have a large corpus (perhaps millions of entries of code, most one or two lines long), I can automatically scan new respositories and send cease and desist letters for violating my client's copyright. Even if a piece of code is very common, that doesn't mean its unoriginal, it just means that there are many people violating its copyright after all. According to the logic of the folks in this thread at least.

You'll have to pry my meta type garbage out of my cold dead fingers. :)

I agree with you - while neat types are cool from a mathematical perspective, they impose a huge maintenance burden. Most times, it's much smarter to accept a bit of type unsafety than to try to achieve perfection.

Pretty much the only exception is in library code where you can use advanced types to provide improved developer experience or prevent user error.

You can actually accomplish this in Typescript, I linked a blog post in one of my other comments.

The key is mapping through all of the arguments and enforcing that argument N's output is a subtype of argument N+1's input.

You can implement a type-aware map, reduce, filter, etc. if you use point free semantics with higher kinded types:

https://code.lol/post/programming/higher-kinded-types/

I have to admit it's a bit heavy, but this is totally representible in today's Typescript if I understand your ask correctly. It would be great if it was a more obvious / automatic inference, though.

But you can type 'pipe' correctly in Typescript, without resorting to overloads:

https://code.lol/post/programming/typesafe-function-composit...

It's even more powerful if you restrict yourself to point free semantics - then you can use higher kinded types to do generic-level variadic composition.

As well, optional generics are trivially achievable in TS via default type parameters, unless I'm not understanding that section of the article.

Thanks! :)

Hey there folks - over my vacation I decided to write 'hkt-toolbelt' to scratch an itch I've been having with writing type-level Typescript.

This package allows you to do higher order 'map' operations on the type level, with a legible and extensible interface.

So are you saying Pareto distributions follow linear relationships? Are you sure about that?

I would read up on the topic if I were you. Statistics can be unintuitive sometimes, and basing your argument on supposition can be unreliable. I'm not sure about the correct answer here, but most folks get e.g. the Monty Hall problem incorrect if they rely on their intuition.

For a product at T = 0, why would it be optimal to assume its lifespan would be T = 0 as well, rather than the average age of all products?

Neural networks operate on floating point values. Are you saying that e.g. 64 bits of resolution isn't enough to accurately model a neuron? The idea of significant figures in physics puts a limit on what is useful in practice.

I could be wrong, but I feel that digital vs analog is a pretty meaningless distinction in this case, as past a certain level of resolution in practice this is noise.

This doesn't even get into the fact that we know the universe is quantum in nature and that systems have a finite amount of possible states. The universe does not have a "infinite" state resolution in the way you are talking about, merely a large one.

I believe that neural networks do perform meaningful computations - they are universal function approximators, and certain variants of transformer networks have been proven to be Turing complete.

What is your evidence that neural networks do not perform "meaningful" computation?

If the model responded, "I have not visited New York because I am a computer", would you consider that to be a good step?

If you require physical embodiment, that is fair - but I don't think there's any principled reason to expect a multi-modal neural network could not be embodied with appropriate sensors or actuators.

Transformer models are universal function approximators. They do not "search" - in the same way that a calculator does not "search" when you ask it to perform an operation.

Thanks for the question.

The point I am making is that if the capabilities of multi-modal neural models continues to improve, it will become harder to convince folks one way or the other.

The reason I went with that scenario is that I thought it would be compelling and evocative, as well as a logical extreme of AI agency in human lives.

I personally find the entire question meaningless for the most part, although it's really interesting the divide in peoples opinions on the topic.

Why not?

Imagine a look up table of every possible intelligent conversation under 10,000 words. We then construct an agent which merely looks up a given prompt in the table.

Why should this agent not be considered intelligent? Does the method of computation matter? Perhaps only "biological computation" counts in your view?

What if the most expedient way to accurately continue the text sequence is to temporarily "embody" as an agent that would have generated the prior text? This would involve encoding emotions, performing reasoning ability on behalf of the agent, and keeping track of its memories.

Do you find the above scenario plausible?

Suppose a child was raised by a more advanced version of large scale language model, with the ability to detect and act on its environment, as well as form memories.

Further suppose that this child develops a deep emotional connection with the AI.

Do you think that child would find your argument convincing that the AI is "not really a person"?

Out of curiosity, what is your evidence that modern large-scale language models are "incapable of belief"?

What constitutes belief in your view?

Why does this article on code practices unnecessarily mention NSFW topics, which now precludes my sharing this (otherwise well written and insightful article) with my coworkers?

The subheader of the website mentions they specialize in professional Go consulting, but I'm kind of turned off by their unnecessarily risqué metaphors. Does not seem too "professional".

I have indeed seen codebases where it seems like the programmer was being charged per source code byte. Full of single letter methods and such - it takes a large confusion of ideas to motivate such a philosophy.

Indeed, in the limit of maximal abstraction, i.e. semantic compression, code becomes unreadable by humans in practice. We can see this in code golf competitions.

Perhaps this is pedantic, but that doesn't look like good procedural code to me. Instead, it looks like a set of mathematical equations.

I think there is intrinsic value in simple, procedural code that has state and produces side effects - it can even be typed! However, I think FP maximalism is bad for code readability. How might you write to a MongoDB document in Haskell, for example?

A common example in this space is how to specify a baking recipe. Do you list it as a series of steps to perform on the world, or do you reductively assert what a cake is from base principles?

I think Haskell developers need to realize that pure functional programming is intrinsically more difficult than other kinds. It requires a much higher ceiling for abstract thought.

What I'm essentially trying to assert is that you have to be smarter than average (whatever that means) to read and write Haskell code. I believe this accounts for the majority of the perceived efficiency. The cost is that you are excluding people who are not mathematically talented.

This is also my practice. Any code that is complicated enough to require a special comment is also complicated enough to be its own function.

On the contrary, I find standard JSDoc and its variants to be an excellent tool for internal documentation. With hovering support in modern editors, it allows context explanation in a very streamlined and human way.

The author mentions for preconditions to just “read the code”. I consider this bad advice. If using an external library, would you rather hover over the method and see its conditions, or would you rather crawl into the third party source code?

I recommend that you treat the internal structures of your code as reusable third party libraries, and not assume that anyone will be familiar with it or how it’s used.

Often my JSDoc comments take up more vertical space than the code itself, sometimes even with ASCII tables or example usage code. I believe this is one of the best approaches to documentation, especially paired with a automated documentation site generator tool.

Code is read much more than it is written. You need to think like a writer and consider your audience.

There is none! Variance could be arbitrarily high I suppose. In the limit, if you assume a uniform distribution, I don't think that a single sample will give you any information about the underlying distribution characteristics.

However, for the sake of the argument, variance can be arbitrarily high and still provide increasingly lower amounts of Bayesian evidence. As long as the variance is not unbounded (i.e. as long as the distribution is Gaussian, which I think is a reasonable assumption in nature), you can extract _some_ evidence from a single sample.

My view is that we can reason about organisms living in space because we are one of them. We have a sample size of one, but assuming that properties of different space-faring organisms are distributed normally, it stands to reason that the first sample collected is near to the mean of the underlying distribution.

In other words, imagine a normal distribution with very, very low variance (say, 0.0001). The mean is unknown to you. If you take one sample, your sample will very likely be close to the mean.

Hey David, I really feel for you. These PR-esque situations are never fun, but I think you’ve done an excellent job of handling this.

I believe there is a bias people have of viewing (somewhat ironically) things in terms of a “David vs. Goliath” lens. Everyone loves rooting for the little guy, even when the little guy is objectively incorrect.

Hope your weekend goes better than it has been so far! Retool is an excellent product by the way :)