Machines Don't Press Play
For decades, video's been optimized for one consumer: a human pressing play
But when training a world model, the consumer is a machine, not a human
Check out @ngates_ 's deep dive on training with videos and the infra opps to help you move faster
We've been working on a major schema migration at @PolarSignalsIO that has unlocked some pretty wild improvements. Apart from a 50% (!!) reduction in storage size thanks to logical compression in @vortexdotdev, there has been a dramatic improvement in query performance.
@GDSTeam I don't think this is how you do http redirect - Location: redirect:/onelogin/there-is-a-problem. But also seems like this is an unhandled error elsewhere allowed to propagate to the user
@mitchellh Have you had the agent try to modify your library code to make the fuzzer happy yet? The objective function might have been underspecified :D
Vortex is BOTH 38% smaller and has 10โ25x faster scans than Parquet+ZStd for TPC-H SF10
We implemented BtrBlocks-style cascading compression in Vortex that recursively tries codecs like ALP, FSST, and bit-packing, letting the data pick the best encoding
https://t.co/6Nn4rrN00j
@mitchellh This is a cool feature, definitely love the ease of creating useful themes. It has surprising behaviour when combined with some appliactions. Vim theme (https://t.co/L1xGyQZOuo) I use ended up being needlessly washed out/pastel like. Also couldn't distinguish accents in fzf.
Well, it has happened. Ghostty on macOS now has search. This is now in tip releases, I'm certain there's a bunch of polish and bugs to work through over the next months, but more than enough time for it to be rock solid for 1.3 next year. GTK to follow soon. Gobble, gobble. ๐ฆ๐ฆ
We're building the data infrastructure that AI actually needs.
Current systems were built for humans reading dashboards. But an H100 can consume 4 million images per second.
The future isn't human-scale. It's machine-scale.
Introducing Spiral: Data 3.0 ๐
1/8
Hey @duolingo how would I have both gotten the friends quest boost and failed the friends quest at the same time? Sounds like someone could use a refresher on transactions and consistency models.
We are hosting our inaugural London Systems event on March 26th, inspired by the analogous events in NYC & SF.
We will feature talks from Mistral and Spiral, with plenty of time for mingling over drinks and pizza.
RSVP here: https://t.co/gRyNlasYeR
Is your data where it needs to be? We explore the critical design decisions surrounding data layouts when creating file formats. https://t.co/PQiKxWdE5U
1/ We (@SpiralDB) are putting together a London systems meetup, to give a chance for (software) engineers to share and discuss their work on the underlying systems and tools that power most of our work - databases, compilers, VMs etc. and we're looking for speakers! >>
At @spiraldb, we recently implemented an initial version of FSST for fast, random-access string compression. I wrote a blog post to walk through some of the concepts.