Live data from Hacker News

Anthropic invests $1.5M in the Python Software Foundation

discuss.python.org

21–30 of 192 posts

Re: Anthropic invests $1.5M in the Python Software Foundation

#21

This makes sense given how much of the current AI ecosystem is built on top of Python. I hope this helps the foundation improve security for everyone who relies on these libraries.

For anyone who isn’t aware/remembering, this is certainly made with the security of PyPi in mind, python’s main package repository.

NPM is the other major source of issues (congrats for now, `cargo`!), and TIL that NPM is A) a for-profit startup (??) and B) acquired by Microsoft (????). In that light, this gift seems even more important, as it may help ensure that relative funding differences going forward don’t make PyPi an outsized target!

(Also makes me wonder if they still have a Microsoft employee running the PSF… always thought that was odd.)

AFAIU the actual PSF development team is pretty small and focused on CPython (aka language internals), so I’m curious how $750,000/year changes that in the short term…

EDIT: there’s a link below with a ton more info. This gift augments existing gifts from Amazon, Google, Microsoft, and Citi, and they soft-commit to a cause:

  Planned projects include creating new tools for automated proactive review of all packages uploaded to PyPI, improving on the current process of reactive-only review. We intend to create a new dataset of known malware that will allow us to design these novel tools, relying on capability analysis.

Re: Anthropic invests $1.5M in the Python Software Foundation

#22
post #14

Just recently I heard that typed languages are best for agentic programming

AFAICT Python basically is a [statically-]typed language nowadays. Most people are using MyPy or an alternative typechecker, and the community frowns on those who aren’t.

It's a pretty nice best-of-both-worlds arrangement. The type information is there, but the program still runs without it (unless one is doing something really fancy, since it does actually make a runtime construct that can be introspected; some ORMs use the static type data to figure out database-to-object bindings). So you can go without types for prototyping, and then when you're happy with your prototype you can let mypy beat you up until the types are sound. There is a small nonzero cost to using the types at runtime (since they do create metadata that doesn't get dropped like in most languages with a static compilation step, like C++ or TypeScript).

I can name an absolute handful of languages I've used that have that flexibility. Common LISP comes to mind. But in general you get one or the other option.

Re: Anthropic invests $1.5M in the Python Software Foundation

#23

Just recently I heard that typed languages are best for agentic programming

Just recently I heard that they can donate to “typed languages” too, a donation to one language does’t preclude other donations, and given their cash injections they have a few $1.5m’s to spare.

Re: Anthropic invests $1.5M in the Python Software Foundation

#24

Earlier quoted context omitted.

Types are best, period. Whether they are native or hints doesn't really matter for the agent, what matters is the interface contract they provide.

I don’t get this argument because if we put the effort to get it typed, we don’t get one of the best benefits - performance.

But that's not the argument here. Python type hints allow checking correctness statically, which is what matters for agents.

Re: Anthropic invests $1.5M in the Python Software Foundation

#25
post #14

Just recently I heard that typed languages are best for agentic programming

AFAICT Python basically is a [statically-]typed language nowadays. Most people are using MyPy or an alternative typechecker, and the community frowns on those who aren’t.

> Most people are using MyPy or an alternative typechecker, and the community frowns on those who aren’t.

That's not like a widespread/by-default/de-facto standard across the ecosystem, by a wide margin. Browse popular/trending Python repositories and GitHub sometime and I guess you can see.

Most of the AI stuff released is still basically using conda or pip for dependencies, more times than not, they don't even share/say what Python version they used. It's basically still the wild west out there.

Never had anyone "frown" towards me for not using MyPy or any typechecker either, although I get plenty of that from TS fans when I refuse to adopt TS.

Re: Anthropic invests $1.5M in the Python Software Foundation

#26
post #12

I did not know you could make donations with a string attached ("improve security")...

Of course you can. The vast majority of donations of this magnitude come with strings attached, be it how the money is spent, access to leadership/events, etc

Re: Anthropic invests $1.5M in the Python Software Foundation

#27
post #12

I did not know you could make donations with a string attached ("improve security")...

It's super common with non-profits. Obviously they would prefer no strings attached but some light strings are usually not a problem for most non-profits.

And they come in a variety of bindingness. I didn’t notice any details in this link which makes me think this is mostly a handshake deal, but it wouldn’t be at all unusual for there to be some auditing mechanisms on a quarterly/yearly cycle.

For example, Wikimedia just recently claimed that they can’t chase some political project that critics wanted them to because most of their funds are earmarked-for/invested-in specific projects. So it does happen with US-based tech non-profits to at least some extent.

Re: Anthropic invests $1.5M in the Python Software Foundation

#28

Earlier quoted context omitted.

Types are best, period. Whether they are native or hints doesn't really matter for the agent, what matters is the interface contract they provide.

I don’t get this argument because if we put the effort to get it typed, we don’t get one of the best benefits - performance.

The best benefit depends on your problem domain.

For a lot of the business world, code flexibility is much more important than speed because speed is bottlenecked not on the architecture but on the humans in the process; your database queries going from two seconds to one second matters little if the human with their squishy eyeballs takes eight seconds to digest and understand the output anyway. But when the business's needs change, you want to change the code supporting them now, and types make it much easier to do that with confidence you aren't breaking some other piece of the problem domain's current solution you weren't thinking about right now (especially if your business is supported by a team of dozens to hundreds of engineers and they each have their own mental model of how it all works).

Besides... Regarding performance, there is a tiny hit to performance in Python for including the types (not very much at all, having more to do with space efficiency than runtime). Not only do most typed languages not suffer performance hindrance from typing, the typing actually enables their compilation-time performance optimizations. A language that knows "this variable is an int and only and int and always an int" doesn't need any runtime checks to confirm that nobody's trying to squash a string in there because the compiler already did that work by verifying every read and write of the variable to ensure the rules are followed. All that type data is tossed out when the final binary gets built.

Re: Anthropic invests $1.5M in the Python Software Foundation

#30

Just recently I heard that typed languages are best for agentic programming

For any programming really, but I think Python got big due to a) the huge influx of beginners into IT, b) lots of intro material available in Python and c) having a simple way to run your script and get feedback (same as PHP) I say that as someone urging people to look beyond Python when they master the basics of programming.

Python has a terseness that is hard to rival. I think that was a major selling point: its constructs and use of whitespace mean that a valid Python program looks pretty close to the pseudo-code one might write to reason out the problem before writing it in another language.
Post reply on HN