neurstle

i make hw that goes fast at your local bespoke hft firm

[neurstle @neurstle]

“oh my god did you hear the nvidia gb300 nvl72 has 4x more tok/s/MW than the b300 at a p90 tok/s/user of exactly 44 how groundbreaking”

please use normal benchmarks

[neurstle @neurstle]

arm AGI cpu feels like what emerald/granite rapids would be if intel didn’t go chiplet style

[neurstle @neurstle]

what’s good hot chips

[neurstle @neurstle]

this is your counterparty

[Former Flag Guy @FSOBrat]

@nominalthoughts What is even the point of this kind of graph for soccer if a single goal shifts your win percentage by 20%

[neurstle @neurstle]

it’s like prompt engineering but it’s our interpretation of the situation rather than the models

[Maia Bittner @maiab]

This isn’t that weird. You told it to make up an image that you describe as strange and it made a strange image. The same thing happens if you tell it to make a strange image.

[neurstle @neurstle]

it’s just gb10 in a laptop form factor.

[Lazarz @Laz4rz]

Call me Eastern European, but I have a bad feeling about this

[Daniel Tenreiro @TenreiroDaniel]

@Pontifex Holy Father the post will perform better if you append the url in a threaded reply as opposed to linking it in the tweet

[neurstle @neurstle]

people absolutely review compiler output when the code matters. and if it doesn’t matter, why are you writing it

[solst/ICE of Astarte @IceSolst]

Interesting article on treating agent output like compiler output (and why)

https://t.co/UHV1V1PzN1

[neurstle @neurstle]

@hitorilabs banger oomf

[Andreas Abel @uops_info]

I have added latency, throughput, and port usage data for Emerald Rapids, Meteor Lake, Arrow Lake, and Zen 5 to https://t.co/dq3CQa6ISj.

[neurstle @neurstle]

@dejavucoder hi everyone!

[neurstle @neurstle]

@himanshustwts dqn not even mentioned. i need to brush up ����

[neurstle @neurstle]

@hitorilabs very nice, i feel like every 2 months i cycle between doing one of those well and the others poorly. congrats on the new job

[neurstle @neurstle]

@teortaxesTex @HellenicVibes Do I think it will be optimal for transformers, probably not. Maybe the MLP layers but that’s really not the bottleneck nowadays. But I wouldn’t call the approach doa

[neurstle @neurstle]

@teortaxesTex @HellenicVibes great.

Beyond that it seems this style of Tiny Recursive Model or SSMs can start to solve some of the issues of limited token counts

[neurstle @neurstle]

imagine this in Chicago. all quants. nobody talking to one another. it would be awesome

[Vipul Gupta @vipul_1011]

NYC has insane AI talent density, but barely any high-quality AI events.

Meanwhile, SF has 5–6 every week. That gap doesn’t make sense.

So we are trying to change it.

We are hosting a curated meetup for serious AI builders & startup enthusiasts.

First one is on Feb 27!

We are starting small and will slowly scale it up.

Comment or DM if you interested. Share it with your NYC AI friends!

[Alexia Jolicoeur-Martineau @jm_alexia]

Cool new work combining Mamba with TRM

https://t.co/Itecn4ccC2

[neurstle @neurstle]

the models yearn to be claude

[Moon @MoonL88537]

holy smok.