I'd like to add to your point that private torrent trackers have had invite tree systems for awhile, and usually if your invitee breaks a rule, you get in trouble as well, so you are encouraged to only invite people you trust. The system has worked well for a long time, and some of these communities still thrive because of the trust that is built.
You realize homes are also private property right? You can have a shitty neighbor like the one described that is also enabled by the fact that they're in their own home. That doesn't justify what they're doing, but your argument against stores as "private property" doesn't hold water.
I could, but most people, even the ones who advocate for "homeless rights" don't want to live in a homeless camp. They are fine with letting others though.
The cuDF interop in the roadmap [1] will be huge for my workloads. XGBoost has the fastest inference time on GPUs, so a fast path straight from these Vortex files to GPU memory seems promising.
Can you explain how it’s faster? GPU memory is just a blob with an address. Is it because the loading algorithms for vortex align better with XGBoost or just plain uploading to the GPU?
What you can do if you have gpu friendly format is you send compressed data over PCI-E and then decompress on the gpu. Thus your overall throughput will increase since PCI-E bandwidth is the limiting factor of the overall system.
That doesn’t explain how vortex is faster. Yes, you should send compressed data to the GPU and let it uncompress. You should maximize your PCI-E throughput to minimize latency in execution, but what does Vortex bring? Other than Parque bad, Vortex good.
A (simple) benchmark would be great to figure out where the practical limits of such an approach are. Runtime is expected to grow with O(n*2) which will get painful at some point.
For your wimsey library, using “pipe” to validate the contracts would seem to me to drastically slow down the Polars query because the UDF pushes the query out of Rust into Python. I think a cool direction would be to have a “compiler” which takes in a contract and spits out native queries for a variety of dataframe libraries (pandas/polars/pyspark). It becomes harder to define how to error with a test contract but that can be the secret sauce.
Actually you're almost 100% describing how Wimsey works! It's using native df code rather than a UDF of some kind. Under the hood it uses Narwhal's which converts polars style expressions into native pandas/polars/spark/dask code with super minimal overheads.
If you're using a lazy dataframe (via polars, spark etc) Wimsey will force collection, so that can have speed implications. Reason being that I can't find a cross-language way yet of embedding assertions for fail later down the line.
Most people want unlimited for the convenience. You can very easily go to Fi or a budget carrier for more fine-grained plans like the ones you described.
Looking at your YC history, Rubbrband was initially meant to be "a way to train open-source Machine Learning models in just one line of code." I'm curious what this initial offering was and why you pivoted.
yeah we went through a few pivots actually - Rubbrband was initially a mobile app for musicians. It would listen to your practice sessions and give feedback. We pivoted out of this primarily because of issues with the technology and market.
At what point does it stop making sense to call it a pivot? I was an adopter during your open-source SD training script phase. These earlier efforts sound completely unrelated to anything I'd be interested in. So what are you, as a team, even doing, if your customer and your product are so inconsistent? What is the driving force?
Gainax's financial troubles are well documented in an interview with Hideaki Anno, where he detailed how the company was pretty dysfunctional from the start. The company's president was arrested for tax fraud in 1999, for example. This has been a long time coming.
Ask not why GAINAX bankrupted, ask how the hell it survived so long.
Memes about GAINAX running out of money to finish series go all the way back to Gunbuster, which was their fourth production, and second under the name GAINAX.
The second production, Daikon IV, survives thanks to piracy because they didn't license them music used :V
My favorite "no budget" moment was in Kare Kano, a lesser known Gainax adaptation of the shoujo manga of the same name.
The second half as a whole is clearly made on the cheap, but episode 19 is special, it is all paper cutouts on a stick, most likely from the storyboard, on top of recycled backgrounds.
Gunbuster ran out of money to pay colorists by the end of the series, which was only 4 episodes. Evangelion also clearly ran out of budget by the end of the series, and that was after some blatant cost saving scenes repeatedly appearing in the show, like very very long elevator rides where nobody moves and the only action is a floor counter ticking down, or watching the counter on a Walkman slowly tick by. Almost taunting the audience with the counters on the obvious padding scenes.
The funny thing is, the studio claims last episode of Gunbuster was done black&white on purpose - in fact, that black&white was more expensive the way they did it than otherwise.
Unfortunately I suspect that outside of the people directly involved we won't ever hear properly. The rumours flew around since early 1990s, and there is the part that Gunbuster was originally supposed to be a TV series (AFAIK), and as result the plot was extremely compressed into those 6 episodes that we got ultimately.
And it's not like the studio was not known otherwise for wild swings even when they didn't run out of money... (anyone remembers how Mahoromatic ended?)
reply