Hi, I'm James. Eternal dilettante and purveyor of nonsense, much of it about #Python or #physics. I work for a computer vision company whose customers actually care about results, so the current """AI""" craze is slowly melting my brain. I boost more than I post. TANSTAAFL
does the linked-to procedure also apply to investigations by the DOJ, or is there an analogous mechanism for subpoenas issued for their investigations? That specific code seems to be about appeals to the MSPB, and while there is to my knowledge no public information about the subpoena in this case, it seems strange that this would be the context for the DOJ's request
@jbzfn@encthenet I'm open to the idea of notification here, but wondering what the mechanisms of this are. IANAL, but my understanding is that such a subpoena may come along with a gag order that explicitly forbids notification. Should PyPI defy such an order?
As far as your "eventually", that sounds more like a call to action for general legal advocacy work than it does like something PyPI has any leverage over right now. I'm all for it, but it sounds like a separate topic.
I learned today about an LLM[1] with a large context window and the demo in the announcement is "we fed an entire book into this, then changed one sentence in the book and fed THAT in, then asked it to identify the difference, and it got that done in 22 seconds"
I am reeling from the sheer quantity of biz brain that leads one to brag about making a diff tool that costs O($1) per query to run and takes O(1 min) to evaluate
[1] which I shall not name because true names hold marketing value
The dedication page in the paperback of Forever Peace:
This novel is for two editors: John W. Campbell, who rejected a story because he thought it was absurd to write about American women who fight and die in combat, and Ben Bova, who didn't.
@recursive Haldeman seems like a mensch. I still need to read that one, I got it confused with "Forever Free" and ended up reading that one in its place
I'm looking for a list of file formats that use ZIP (like DOCX, EPUB and JAR) which are ZIP files with standardized contents. Are there more formats that sit on top of ZIP like this?
@edsu Python's "wheel" format (.whl) is the example that leaps immediately to mind, with the contents standardized by PEP 427 https://peps.python.org/pep-0427/
I know also that the multiphysics simulation software COMSOL has a proprietary format (.mph) that is a ZIP with some standardized content, although I'm not entirely sure how formally/publicly the contents of the archive are communicated (as I recall, it's mostly a bunch of XML, mesh data, and maybe some Java source)
can't believe people keep saying they can use AI for code. out of morbid curiosity i've tried a couple times and i've never gotten anything usable out of it. feels like anything for which there isn't just a copy/pastable stack overflow answer it won't help with, and if there's an SO answer then you can literally just google it and copy and paste it. what is this code that people are writing with AI? i really wanna know
@aeva@eniko yup, these tools perform decently if you want it to rattle off some code that looks like something you could dig up on a forum, from a free online book, from docs, etc.
and otherwise, they have underwhelming performance, but will muddle through and generate something possibly-executable that pantomimes a solution
@Doomed_Daniel fancy running into you here in a thread about C++!
Thanks again for porting Classic Doom 3's changes to dhewm3 many moons ago! The email you had to send me to get the licensing sorted out for that is my canonical example of why it matters a lot to pick a license up-front and communicate that clearly 😅
@Doomed_Daniel yea, I think you're right. I guess RAGE was the first 'real' test-case for that, and maybe the only one since another acquisition happened after that, and has certainly done no favors to the old practices.
I am delighted to announce that I have recently accepted a position on the Board of Directors for the Haskell Foundation.
The Haskell Foundation is an independent, non-profit organization dedicated to broadening the adoption of Haskell, by supporting its ecosystem of tools, libraries, education, and research.
I look forward to working with the community to help make a better tomorrow :)
I have a base class A that is a factory for Bs, and A takes a type parameter because of covariance or whatever. but this seems to stop me from writing a function that takes the base class as an argument (see works() and does_not_work() near the bottom).
anyone know what I'm supposed to do here? thx
@regehr does declaring does_not_work() as a template<class T> that takes an argument of type A<T>& do what you'd wanted, or does that not make sense in situ?
A friend is trying to decide whether to join a startup that plans to train LLMs on the writing of particular individuals so people can chat with their loved ones after they're gone. I said, "That's pretty fucking macabre." My friend said, "That's what I thought, but the founder asked me why it was different from re-watching old family movies and I don't have an answer." My friend has lost people to AIDS, cancer, and dementia and still misses them; I honestly don't know what to think…
@gvwilson the difference to me seems to be that watching an old family movie is watching something that actually happened, recalling something that was, and LLMs for this pupose are by-definition synthesizing an impersonation, with the goal being a sufficiently-convincing (or healing, or whatever) simulacrum.
What the rest of that chain of reasoning/ethics looks like involves a lot more moving parts, but the distinction seems like a fairly direct one to me.
@gvwilson on the other hand, I lost my father 2 years ago and the grief is still very fresh. I think of him every time I see the idea floated, and there's certainly some ambivalence there.
I think I'm open to the platonic ideal of the concept in a SF sense, and not really at all open to the reality of what for-profit implementation of it will end up (has already ended up) being
@glyph@gvwilson yea, I think even if we allow the technologists the full power of the magic wand here (i.e. let's just assume we have really rich training data, etc.) there are still a lot of things left to consider.
And in reality, the technology doesn't get to rely on the magic wand. Result will probably be a lot of backfill when building a fine-tuning corpus, and/or maybe just outright lying and aiming for a generic corpus that pushes emotional buttons (a la psychic grift)
@universalhub eastbound train on Framingham/Worcester commuter rail announcing to us that there is a water rescue at West Natick Pond preventing train passage, passengers will be deboarding at Framingham in favor of a bus (~7:45 PM)
Has anyone ever written a sci-fi story based on the notion that the ideal future military is a decentralized insurgency, and in its success, it spells the end of centralized state power entirely?
@gsuberland this is a good way to put it! my only encounter with it was a relatively frustrating one, porting some MicroPython code demoing talking to an e-ink driver to Lua, and I was grumpy that I couldn't find a way to tell the watchdog to just piss off and let my code monopolize the system
but then, I guess that's kind of a good thing for an RTOS to strongly discourage!