moab

> I’d still take HNSW over such approach

Why? Sorry, but your comment is not very insightful without more explanation, and the rest of your comment is just an advertisement for your (admittedly impressive looking) work.

Spot on. There is zero moat and the self-hosted alternatives are rapidly improving (if not better) than Pinecone. There are good open-source contributions coming from bigcorp beyond Meta too, e.g., DiskANN (https://github.com/microsoft/DiskANN).

That's very fair. I'm sorry for being rude and have a great weekend as well.

You don't know what you're talking about. Research and teaching at the upper-division level can and should be two sides of the same coin.

That's not even remotely true (re: the need to use LinkedIn). Just quit.

Do you plan to publish about what is being implemented soon / support ANN Benchmarks or BigANN Benchmarks?

I am not sure what algorithmic "secret sauce" pinecone is implementing beyond being a wrapper around FAISS, but I have got to give it to James that he is doing god's work in helping people understand ANNS and how to apply it.

> including software

highly unlikely. You should be concerned if you are just plumbing between different software stacks, but even that is likely hard to fully automate.

Makes sense. I hope you keep working on this. There is a big gap between what people are doing in the academic (and even publicly-described industrial literature) and what you describe. So I think there is a lot of opportunity to push on this front if you have some ideas.

Totally agree with your last point. But based on my understanding, over the past 5-ish years commodity multicores with multi-terabyte memories have made a big dent in the supremacy of supercomputing, at least for some of the topics you mention (thinking about graph analysis in particular).

> codes that would be effectively computationally intractable if you tried to run them in, say, AWS.

such as?

I mean, that's what I'm wondering about his parents. They must have known, or at least had a feeling. And not looking into that and giving your child some advice not to go down this path, or to be careful, and at the very least, not to get involved yourself, that would have been the responsible thing. Maybe it's the blindness we can all fall prey to when it comes to objectively thinking about our blood.

I'm pretty sure you are. I've never seen M being used for thousand.

How can you feel sorry for them? They're professors of LAW at Stanford. And this white-collar crime was being committed right under their noses. Pynchon would have had trouble coming up with a plot this wacky.

110 years! I wonder how many she'll get out of that for being the first out of the woodwork.

If someone keeps shooting themselves in the foot, at some point you have to throw in the towel and call a moron a moron.

What a hypocrite. Elon is now living in an alternate reality where he is being constantly persecuted, defamed, and under attack. I wish him all the best, but based on our great works of literature (think Macbeth), it's usually all downhill from here. Was Grimes his Lady Macbeth? One can wonder.

As a little twist on the author's badly thought out proposal, what if we simply take the NSF scheme for doling out funding and make it more broadly available. Each year (or every few years) you can submit a proposal about what you plan to do if you're given a UBI, it will go through peer-review, and about 6 months later you'll get a notification if you were selected for funding or not.

Hell, as a PI currently writing a bunch of NSF proposals this doesn't sound half bad to me.

It wasn’t just rich people. What about the patients that were lied to? The employees?

I think it's quite different. For GPUs these archiecture-specific optimizations / primitives make orders of magnitude difference (and also change from version to version). On the other hand, multicore code I wrote ~8 years ago in TBB or Cilk is still extremely fast compared to hand-optimized code customized for current CPU architectures.

To summarize my gripe, it's is that the abstractions on GPUs seem broken, and we still don't have a good model for how to write high performance GPU code that just works and keeps getting faster as GPUs get faster.

That sounds awful and a big reason why I think GPUs will go the way of the dinosaur in the near future.

There are many "extra" features that one needs in practice which are patently not solved, starting with working in a dynamic setting.

Should faculty be paid more? Absolutely. Should Ph.D. students be paid more? Absolutely!! But the blanket statement you make in (1) is wrong and strikes me as awfully close to the extreme left-wing and right-wing mindsets of "the system is fucked up beyond repair, all that remains to be done is to tear it down". The reality is more nuanced than this, and the picture you paint of industry is hardly that rosy, even at silver-spoon companies that invest heavily in R&D.

In previous generations, other forms of popular culture filled the same role, and hordes of young kids conformed in exactly the same way. From the beats and hippies all the way to skate culture in the 00s. I'm sure my parents felt the same way about me skateboarding and wanting to buy crap at the mall.

:-)

It's spelled "shibboleth". You could have googled it.

Plenty of those, and inter-racial Indian marriages among friends who happen to be upper-caste exist. You're either not a first-generation Indian (seems likely) or you're seeing the story you believe in. And FWIW, even my dad's expat friend circle from IITB has people from all different caste backgrounds as extremely close friends, though you might cynically argue that IIT is an even higher caste.

The claims made in this post are in dire need of citations, and don't strike me as believable. And just FWIW, most first-generation Indians in America couldn't give two shits about the caste system.

The plan makes sense. A likely issue for folks currently preparing papers in LaTeX is that they will be put off by the extra work of having to port/write another copy of their document in Nota, but maybe this is intended. You shouldn't just be dumping your paper to the web.

Re documents that would be useful, you could dogfood Nota with 242 to show how powerful it could be to teach students PL theory. A course like 15-312 would be much less intimidating for students if materials were encoded both informally/formally in Nota. You could imagine porting one chapter from PFPL as a small example.

I'm teaching a new course on algorithms in the spring and I plan to try it out to develop some online course materials. Will get in touch with you with any questions / to share what we prototype. You might also want to share this with Umut who has been developing (what I believe is) a KaTeX-based system called Diderot that a lot of CMU courses are using now.

Nice work as usual, Will. How do you plan to convince users that this is not just a research prototype that will be abandoned after being published, and won't suffer from bit-rot 5 to 10 years from now? You talk about this a bit at the end of the longer paper on Nota, but I'd be curious if you have additional thoughts. Seems like we need something like Talos/immutable infrastructure but for the web (does that exist already?).