Max Caldwell

@0xSero Looking to replicate this on my m4 24gb rn. Codex is on it.

Holy moly we doing it

@LottoLabs I want to hear more about this

I couldn’t go back to not using Hermes agent

@RampLabs Can't access the article

@ValerioCapraro he did not at all say this lmao

I've always been really interested in JEPA for a while now. Anyone working on it want to grab coffee around the Bay Area sometime? Have some ideas I'd love to run by and love talking to smart people on the forefront.

Exactly and translating between the two can be pretty difficult. Different performance for different data types, especially I've had troubles with mac int4 perf in the past, whereas Nvidia seems to be going hard on that. It's just one edge case but there's lots of them. Fortunately gpt 5.4 can fully one shot half of the differences if you are doing ports.

@max_takeoff MLX is improving but it's still a different ecosystem from everything else. that's the tradeoff with Mac for local inference, you're learning a separate stack instead of the one everyone else is building on.

my laptop has a dead RAM slot and it's time to retire it. i need something that runs local models, handles development, and travels. what are you using and what would you recommend?

@sudoingX Impressive little model

if you have a 12gb graphics card collecting dust in an old gaming rig or workstation, read this. i ran a 9 billion parameter model on a single RTX 3060. 50 tokens per second. it wrote a full space shooter from scratch, 3,263 lines across 13 files. zero handwritten code. zero cloud. zero API calls. your data never left the machine. not once. you're probably sitting on more local intelligence than you realize. stop paying per token for work that should stay private.

use frontier models when you need their capability. mine their intelligence. but orchestrate everything from your own hardware running 24/7. not the other way around. don't let your thinking live on someone else's server as the default. open source will win. it always does.

rumors are frontier training runs are up from 2-4 months to 3-7. if it ends up taking us exactly 9 months to birth agi im going to mass bc something is clearly fucking with us

@joetforhire Great idea. World models are going to be crazy for this in a year or two.

wanna know how u'd fare if -gasp- u were living through the singularity? play MERGE LANE, a life sim i developed during the great claude code psychosis of xmas 2025 play thru an entire near-future life, from birth to deathless bingularity (if u want it) merge.winedark.ink

The death cult presidency.

This is Step 1. Step 2 is "books are bad for you"

@mynameisyahia Why do you think he is so excited about openclaw

Jensen's comp is $50M Is he expected to use $25M in tokens? This logic is absurd

Great little model!

I was fired from Delve today without warning. I was the PM responsible for the compliance automation platform. The one that generated hundreds of SOC 2 reports with zero security incidents for every customer. Taking some time to reflect and will start looking for jobs soon!

@kimbochen @tenderizzation Academic coffee is great. Voyager and Voltaire are good, esp for espresso . Playback is pretty good, a bit overpriced

@tenderizzation Any other good shops around San Jose?

Chromatic does a 7-pour pour over recipe @tenderizzation are you responsible for this abomination? jkjk coffee is still good Just slightly astringent

RT @realmcore_: I love our Slate subscribers, so for anyone on any plan I'm going to reset your usage in two days. As a thank you to every…

I still believe Dense models will win btw

@dumbfook Nooooo

slop the count