Diegstroyer
Will VincentOpenAI acquiring Astral is sending shockwaves through the Python and AI programming communities.
I believe it opens up a 4th way to solve a historic problem: how to fund open source.
TechTrav@wsvincent I have no idea how to feel about this. wow. something tells me it can't be good. I was just starting to like uv.
at least is wasn't Atlassian!
@techtrav I would say Charlie and the Astral team have handled everything extremely well so far. But... hard to know.
Eric Matthes@wsvincent I think your main points are reasonable, but this acquisition was frustrating to me because I really liked the pyx approach.
Astral building a high-quality private package index has a much higher potential to align with maintaining a high-quality open source package management tool.
Astral being acquired by an AI company that has repeatedly acted with disregard for its negative impacts on society makes me expect AI bloat in uv.
@ehmatthes Or premium and future tools only being for OpenAI and companies. I share your caution. It's not just an acquihire.
Jeff Triplett@ehmatthes @wsvincent I hope this gets the PSF / PyPI to finally move on to the "high-quality private package index." They barely dipped their toes in the water then made no forward movement on private orgs and private hosted options.
These are proven and reasonably successful/straightforward business models that even a non-profit can manage.
The rest will sort out, but I don't see a world where OpenAI charges for UV.
@webology @wsvincent I agree, but they get to fill it with whatever pushes and nudges to their projects as they want.
That almost always leads to more friction for most users, and I don't see OpenAi keeping the tool as simple and clean as it has been so far.
That said, I'm going to keep on using it until I see a specific reason not to.
@ehmatthes @wsvincent Re: I don't see OpenAi keeping the tool as simple and clean as it has been so far.
Is that based on Codex or their Python APIs? Just curious where this comes from.
Having used their Python API, it's amazingly clean to work with. I just think Pydantic AI is a better tool because it doesn't force me to use a single provider.
OpenAI's API/REST endpoints have remained clean too, and there are probably >50 providers using their shape as a standard.
@ehmatthes @wsvincent Maybe it's Codex's changes? I haven't found it to be unwieldy yet. Cladue Code and Codex are moving quickly, but I get annoyed that one feature isn't in either. So maybe Codex is less clean.
@webology @wsvincent That's based on their overall impact on the world, and grandiose claims about making everything better.
But I'm happy to hear about your experiences that give reason to think that some tools might remain more useful than problematic.
@ehmatthes @wsvincent That's fair. I have no loyalty to these companies, but I don't mind the open source credits they have given out for years either.
Mario Munoz@webology @ehmatthes @wsvincent Ok fine, I'll jump in with a hot take 😅
I don't think it's unreasonable to anticipate anti-user patterns to emerge over the long term. (I think the GitHub/Msoft acquisition is a fair proxy--though OpenAI is a less stable company, which could accelerate enshittification)
In the near term, they will want to bank on user's goodwill and trust--so I don't see a lot of changes coming immediately. But it's highly possible/likely that intentions will change, whether it's through feature bloat, licensing changes, monetization, user privacy breaches, etc...
I'm glad for the developers. They've done great work, and I think it has been positive for the Python ecosystem.
It's notable, however, that at least on Masto, the overall response to this news is negative. OpenAI is not a trustworthy company... (not to anthropomorphize a company... so, rather, Altman and his sycophants exhibit questionable ethics and viewpoints, and their leadership can't be trusted...)
@pythonbynight @ehmatthes @wsvincent Re: It's notable - I don't think it is. Mastodon is overly negative towards any AI/LLM related. I didn't see the rage on other platforms as I saw flexed here.
I think Charlie and crew built a mountain of goodwill and moved Python forward. I'm willing to see what comes of it, or pull an Eric and switch to the next-best community fork, as he did with VS Code. Except I still think that product is hot garbage 🤣
@webology @ehmatthes @wsvincent Fair on the Masto pulse-check, as it does lean negative.
I also agree that Charlie et al. have done great work which has moved Python forward...
It still doesn't distill my point about OpenAI ... Many of us still use Msoft, Google, Apple products despite questionable leadership, so it's not likely a death knell to the tooling and their use...
But there are reasons that some of us try, as best we can, to de-platform or to lessen our dependence on these products, even when they are essentially "free" (i.e., gmail, facebook/IG, twitter, etc...)
And given the financial volatility of where we are today (I know we may differ on our prognostication on what the future holds re: bubble pops and all), it makes me wonder how OpenAI leverages their stake in Astral should their financial situation change drastically.
But on the whole, yes, I'm hopeful a community-led alternative provides a best-case dismount point should it come to that.
@pythonbynight (culling the list a bit)
My take is that LLMs are really useful and I see years ahead of them getting better.
That said, when local models reach the frontier models' current level (I believe they will by the end of the year), they will have achieved "good enough" status to be useful.
In the end, open source wins, and I never thought UV or Ruff would be the last time any of us switched tools. They are just the best we have today. 🍿
@webology If (when?) local models reach usability that nullifies or minimizes dependance on frontier models, that seems to be a financially ruinous scenario for a company like OpenAI... That's actually one of the scenarios where I see a large potential for harming users that are tied to such tooling. (Must mention, I'll be glad if the harms are abated or minimized!)
And that's a fair point on tool hopping... I'll likely go back using pdm, which I used (and liked) before uv--not necessarily because I'm running for the hills, but I also kind of like trying out different tools anyway.
@pythonbynight I don't think it ruins them. Happy to talk more about that outside of here, but they are minting a business that can spin out other billion-dollar businesses. They are still very niche outside of our world and industry, but that's all about to change.
Charlie Stross@wsvincent What happens to Astral when—not if—the LLM market bubble pops and OpenAI goes bust?
(My guess: within 12 months, possibly a lot sooner. The Gulf crisis is going to crash semiconductor manufacturing in Taiwan and SK by choking off the supply of He for fab lines, never mind the LNG strikes. At which point, no new Nvidia NPUs.)
@cstross They will benefit from the federal bailout of OpenAI I imagine.
@wsvincent If things break as bad as I expect the feds won't be bailing *anyone* out—it'll be far worse than 2008, more like 1929 in severity.