banner
LightNews
ocramz.bsky.social
Marco Zocca
@ocramz.bsky.social
320 followers 710 following 640 posts
ML, λ • language and the machines that understand it • https://ocramz.github.io
ocramz.github.io
Posts Media Videos Starter Packs
Pinned
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · Jul 4
CERN for frontier AI >>>
1 3
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 14m
but then, what counts as market penetration in this case? OAI being profitable? their customers booking profits that are directly attributable to GPT? etc.
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 23m
cursed, demented etc technion-cs-nlp.github.io/llm-arithmet...
Arithmetic Without Algorithms – Project Page
technion-cs-nlp.github.io
1
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 27m
speaking of omens, my 1yo started picking The origins of totalitarianism from the shelf she barely reaches
1
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3h
have a good trip! I'd love to visit SG again soon
1
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3h
changing the constraints counts as cheating ;)
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 7h
writing a blogpost by hand, with no genai content, cookies paywalls or monetization, like a real caveman
5
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 8h
I know ~nothing abt econometrics, but it seems to me a bubble could at least be modeled with >2 parties, with the transition occurring when the second stage cannot sell to the 3rd anymore
1
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 1d
evil vector spotted in the wild
1
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 1d
is grouped query attn also called multi-query somewhere?
1 1
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 1d
the kids figured out an interesting "pair layback" #climbing technique
1
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 1d
timeline cleanse
6
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 2d
This sounds like a great shared task for the next NeurIPS
dlevenstein.bsky.social
Dan Levenstein @dlevenstein.bsky.social · 2d
So I get that a Neuroscientist Couldn’t Understand a Microprocessor, and TBH I’m ok with that. But could a neuroscientist understand a deep RNN? Because that seems like a more pressing issue.

*assuming you think the brain operates through the parallel activity of many connected input/output units
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 2d
days since inspecting the data as a first thing would have saved hours of work: 0
2
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 2d
what changed in v2 and v3?
1
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3d
anyway, @kelseyhightower.com had a rather simpler point around chaining bash commands. We could take a page from probabilistic programming and finally acknowledge that even regular bash commands are nondeterministic if you open a socket or file, and model that at the type level
3
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3d
yes but the fundamental shift is for _tools_ to be nondeterministic and not have accountability at the same time.
1 1
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3d
kids figuring out language compositionality eary:

my 4yo is on a swing next to her friend, who speaks little english still.

As I tell my kid to have a nice day at school, she replies "si papà", and the other one "si, <name>'s papa" 🐵🐵
6
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3d
shall we control for "entities that are in the pocket or have a direct stake into the companies in question"? bc that's what I feel is the axis of this issue
1 1
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3d
Thank you for elaborating your view; I learned something new today.
1 1
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3d
being able to do all experiments in silico means that we can be more deliberate with interventions, and that much more can be controlled for.

I think MI is still figuring out how to control for one factor at a time.
1 3
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3d
building theory is generally slower than making gpu go brr, you'll agree on that :)
1 3
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3d
as a matter of fact "interpretable ML" is way broader than "mech interp for deep learning", and has an much more complex history of methods and results.
2
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3d
the common thread is the vast and barely coherent phenomenology and the lack of a consensus on even the right observation methods.
2 5
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3d
I mean, I care that it doesn't break and that it serves me well, but do I feel for it?
1
ocramz.bsky.social
Marco Zocca @ocramz.bsky.social · 3d
this is super interesting, and a thought I also recently had.

I really must ask though whether having an attachment or a functional dependency on some piece of tech is really the same as empathizing or even identifying with it.
2 1