Kyle Kelley (@KyleRayKelley) header
Kyle Kelley (@KyleRayKelley) profile picture

Principal Engineer @AnacondaInc, nteract lead, maker of document engines and runtimes 🚲 🏫 🚶🏻🌲🏙️ 🌊, prev Netflix, Rackspace

Santa Cruz, CA
Joined December 2012
Photo by Kyle Kelley (@KyleRayKelley)Photo by Kyle Kelley (@KyleRayKelley)Photo by Kyle Kelley (@KyleRayKelley)Photo by Kyle Kelley (@KyleRayKelley)Photo by Kyle Kelley (@KyleRayKelley)Photo by Kyle Kelley (@KyleRayKelley)Photo by Kyle Kelley (@KyleRayKelley)Photo by Kyle Kelley (@KyleRayKelley)Photo by Kyle Kelley (@KyleRayKelley)Photo by Kyle Kelley (@KyleRayKelley)
As much as I want more AND BETTER housing, I have to hand it to Greg Abbott, the Texas GOP, and their wealthy voucher backers. They managed to alienate rural Texas by going after public schools. Don’t mess with Friday Night Lights.
> Be James Talarico > Co-sponsor statewide single-stair bill > Win Texas by 12 pts Clearly, the path to victory in America is through YIMBY single-stair reform.
3
267
Kyle Kelley retweeted
> Be James Talarico > Co-sponsor statewide single-stair bill > Win Texas by 12 pts Clearly, the path to victory in America is through YIMBY single-stair reform.
> Be James Talarico > Co-sponsor statewide single-stair bill > Win senate primary Clearly the path to Washington is through YIMBY reform.
17
134
9
2,432
35,261
$429 is how to tell the Bay Area too many are traveling.
5 different airlines, the exact same $429 round trip LAX-SFO. Is this what airline competition is supposed to look like?
1
1
268
Models are bad at subplots. IOW, bad at subtext. Not just in narrative fiction. All conversations with Language Models even in coding. 1. They miss genuine subplots 2. They invent subplots that aren't there Paradoxically, making the subplot explicit kills it.
What are some things you predict a frontier AI system will *not* be able to do one year from today?
1
1
1
177
Humans can give a nod towards the subtext. We can acknowledge what's happening without saying it outright. Models can't help themselves. They insist on reading between the lines out loud (if they even notice it at all).
58
All of the frontier labs' models are bad at subplots.
137
This talk on Design has me a little irked as a housing nerd. youtube.com/watch?v=GLvFTMtw… There's a lot more going on than just "buildings look same." There's irony in opening with Levittown, PA then treating sameness of newer buildings as lack of intention.
187
I'm surprised the 4o people haven't run out of stamina. That or it's one big astroturf to keep noise running.
3
401
Does @OpenAI have an OSS program that will let me use the Daybreak models? Somehow while investigating a possible issue with celld I'm getting flagged. cc @rough__sea @kevinwhinnery
1
2
297
Damn. All those data science tutorials are screwed.
BREAKING: A new penguin species has been identified for the first time in over 100 years! 🐧 Genetic research revealed that gentoo penguins are actually 4 distinct species, including a newly described species found on the remote Kerguelen Islands. bit.ly/4ACbG1h
6
2
25
6,026
Fixed up @nteractio's rendering inside of @ChatGPT Codex. Interactive tables are back in action.
1
6
234
Amusingly during the ChatGPT Plugins session.
1
51
Am I crazy in thinking this would be faster and on device with SBERT and a local vector store?
jev understands me on an emoji level
1
3
323
Kyle Kelley retweeted
celld v0.5.1 is released~ - Dynamic Workers can enforce CPU time and subrequest limits from WorkerCode.limits - Support for Tail Workers - KV accepts a ReadableStream value in put() - Compaction can read retained LTX bundles after their segments leave the in-memory index, so a slow cell can continue draining its history. - The /evict/<cell> operator route reports a refusal, cancellation, or failure instead of returning success before the eviction completes. - Node.js compatibility improvements github.com/denoland/celld/re…
8
16
1
234
11,580
A\: "We're all going to die" VP looking at AWS Bedrock bill: "We're all going to die"
2
240
Depressing if the labs dress up not collaborating on a pacing proposal as antitrust instead of creating industry standard bodies. How can either party really trust each other though? On a level between individuals, sure. Between organizations, not really. Needs a third party.
First step is for industry leaders OpenAI and Anthropic to stop feuding and work on a pacing proposal together. They'll cite antitrust, but that's fake -- antitrust prohibits certain agreements, but not from jointly developing a proposal. Bringing in USG before there's a concrete proposal will likely result in something dumb (see: our pre-release testing program)
189
Got my first pytorch PR merged!
1
1
15
739