austin

on one hand it’s kinda impressive that wildfire smoke can get all the way down here

on the other hand I’m getting sick of waking up and feeling like I smoked an entire pack of cigarettes

how would you tell the difference

the seaweed would probably have more flavor I guess

i mean this is basically just taco bell

… would

Fluid compute: How we built serverless servers - Vercel Fluid Compute cuts cold starts and compute costs by up to 95%, scaling I/O-bound and AI workloads efficiently across 45B+ weekly requests.

behold

vercel.com/blog/fluid-h...

reinventing spot instances with a per-second rate

did you read the vercel blog about how they’re binpacking jobs onto lambdas

AI is a Floor Raiser, not a Ceiling Raiser - Elroy An AI assistant that remembers and sets goals

Good article. Instead of trying to build products promising to replace humans with "superintelligence", consider the *actual* shape of language models and their capabilities. We have a great floor-raiser here that can be applied to bridge capability gaps.

elroy.bot/blog/2025/07...

i have no strong opinions about airport lounges but I do strongly believe that there should be a special security line for people who fly more than once every 3 months

does openai use bing for deep research or other search tasks? suppose it could also be copilot

Welcome everyone to the Sickos Committee Podcast.

It doesn’t matter how much money you make today because that’s not why they gave you money to begin with.

This is also why you shouldn’t read PR guys takes about the AI business.

in the event someone reads this far down, the “why” is probably useful. if you’re a VC-backed company, making money is great, but only really as an indicator that you’re on to something. if you pitch investors on “i am creating the magic money machine” and they buy that…

like i said - i would not bet that we’ve capped out on scaling! gpt-5 will be interesting. my ~vibes~ are that we’re getting close to a plateau though

sleepy spuds

timeout tubers

i would personally not make that bet!

if you were OAI and decided to cut over to, idk, 80/20 mix of inference and training then yeah you’d probably be looking at those sort of 10x cost reductions.

you’d only do that if you were sure nobody was going to brute force a better model, since there’s no moat.

tokens are getting more expensive "language models will get cheaper by 10x" will not save ai subscriptions from the short squeeze

i think in some ways this is broadly correct (SOTA models are remarkably inflexible in terms of $/mtok) but I also think it misses the forest for the trees. inference drives today’s dollars, training drives tomorrows.

ethanding.substack.com/p/ai-subscri...

wow someone made a video game that accurately captures the experience of being on bluesky during the invite code era

What game is this?

Terror dj

Oh shit saw @babadooknukem.bsky.social at the lowes

The Future Of Software Is Small — aparker.io August 3, 2025 · 6 min read

been having a think about software and platforms again, wrote it down.

www.aparker.io/post/3lvjepu...

i’m empathetic that you’re going through this but I gotta admit it is extremely funny in a grim sort of way

calendaring remains the worst

Tao's one of the most interesting people to listen to on the AI math thing because he's been incredibly fair in his evaluations with it as a tool.

we built this city on

pro to cols

this is glp-1s right?

this is why I give people who swear by codex side eye

HTTP Cats API for HTTP Cats

i'm really not ashamed to admit that i refer to http.cat whenever i need to remember what a status code means