cs @mit, prev @sailresearchco @paradigm
- Tweets 15,365
- Following 1,008
- Followers 94,325
- Likes 188,122
what makes you say this? as far as i can tell GCP docs + Scaling Book combine to paint the picture of 2x MXUs, each being 256x256 systolics
docs.cloud.google.com/tpu/do…
jax-ml.github.io/scaling-boo…
1
2
240
pretty straightforward
they used a pre-existing formalization from Google's Formal Conjectures project (basically a bank of community-reviewed formalizations of problems)
they did make a few seemingly inert code hygiene modifications (inlining definitions etc), but looks legit
1
19
4,869
i don't think (3) will matter much? w/ humans out of the loop you only really care about aggregate throughput of your swarm of agents (assuming enough work is parallelizable)
and batching economics mean you get superlinear gains in throughput per $ as you trade away tok/s/user
3
12
1,465
taking reservations for my LEGO 5nm wafer capacity, volume production starts Q1 2027
35
43
1,008
96,527
orcd.mit.edu/news/orcd-newsl…
orcd.mit.edu/running-jobs/av…
~hundred H100s and something like 80 H200s
1
10
1,238
alright done
fyi in general the person creating your mailing list can easily add themselves to the list and get a copy of all the emails that pass through it (i won't, but stay vigilant)
1
3
614
everybody wants to call their rust code blazing fast, but nobody wants to look at no big-ass flamegraphs
23
4
187
16,753
fun little branchless @fiveoutofnine and i put into seaport that i still think about
14
7
129
13,730
came across this after changing a variable from being in an Arc to static tanked performance
turns out it was a false sharing issue with some unrelated memory. apparently this is a really common bottleneck!
3
1
14
5,228
late to the party on this but really enjoyed this talk, scott meyers is a such an engaging speaker
piped.video/watch?v=WDIkqP4J…
5
1
41
12,041
The RethMatch competition is now complete!
Congrats to the winners and thank you everyone who played — excited to see your bots and strategies.
🥇 @hibillh
🥈 @jparklev
🥉 @nguyenethan01
The entire game is open on Github, feel free to fork and run your own competitions 🫡
new @paradigm experiment with @_Dave__White_
RethMatch, an onchain tournament for bots
starts now! ends sunday. link in replies 🤖
13
11
70
20,269
that's what's happening on the site!
there's just not a lot of activity atm, if you watch for long enough you'll spot people spawn (and die quickly). eg:
3
1
12
4,071
new @paradigm experiment with @_Dave__White_
RethMatch, an onchain tournament for bots
starts now! ends sunday. link in replies 🤖
26
25
371
105,711
>paper by Dave White
>animation by Dave Whyte
you cannot make this shit up
Prediction markets let you bet on outcomes, but so much more is possible.
This paper introduces Multiverse Finance, which splits the financial system into parallel universes so you can short the market today, but only if your candidate is going to lose the next election.
1/
10
3
204
25,208
writing risc-v assembly for class is bringing me back to my first time playing around with yul
18
5
279
19,773
sure the word has a life outside of ml, but within ml it has been mostly defined by Power et al
regardless don't see the point in contributing to any further overloading on the margin
3
169
yup mb should have said "as it goes through the *later* blocks it can attend to the later latents"
tho since the transition from early to later latents is decently smooth (at least w/ GPT-2), you'd at least get some useful transfer in the middle blocks(?)
ALT via https://www.lesswrong.com/posts/AcKRB8wDpdaN6v6ru/interpreting-gpt-the-logit-lens
2
192
They find some apps leak screen-recordings/images — it shows the effectiveness of their methods.
It's written in a way that it emphasizes the positive findings even though the negatives are significant.
They don't find microphone leakage from Facebook or any major app.
ALT via Gizmodo
188
rankings for it already came out today (that's why it dived from 50% -> 5%)
it's basically trading at interest rate + chance that more votes coming in changes things by Aug 12
3
424
looks like some eccentric polish dude who just loves polaroid bought it in 2017 and started to turn it around from PE leadership lmao
1
1
19
1,225
he's talking about ammeter speculators not being able to beat index funds — his argument why liquidity wouldn't suffer is based on sophisticated speculators still participating
2
206
i'm sure plenty of people were/are saying that, but to be clear there was a community note saying the picture was digitally altered
2
396
have been looking forward to visiting this library again for so long
i love you mill valley
7
1
84
8,841
surprising that a plurality of student debt is for master's degrees tho — that definitely confuses me a bit
2
2
317
I retract my previous tweet!
Student debt relief is obviously still regressive — but it mostly doesn't go to low earners, as @bookdepth thoughtfully corrected me on.
Most debt is held by high earners.
4
1
21
5,761
yeah wow i think i'm very wrong about this — even just looking at current income only <20% of debt is held by the bottom 2 quintiles
brookings.edu/articles/who-o…
1
2
169
yeah this is basically just an efficient way to enable a model to zero-shot generate soft prompts on the fly
102
Guess which of the technical questions I wrote hehe
Excited to announce the @Paradigm Fellowship, SF, Sep 25-29th 2024.
Paradigm Fellowship is a program to gather crypto’s best young engs and researchers.
During that time we'll work directly on the hardest problems in the industry.
Excited to meet and collaborate with you all.
17
3
125
27,829
don't even get me started on the global social mobility index
it makes no attempt to actually measure social mobility, it's just an index of factors the WEF thinks contribute to it
*relative* wage inequality determines 1/10th of your score, having a large % of employees working >48 hours/wk is penalized, etc
they also use mean instead of median for a lot of things which skews things for the US
12
763
wait no its literally just them being scummy and lazy, there's no valid data preservation reasoning for it
2
118
what they hell they actually were planning to bill me for the rest of my life if i didn’t log into “.Center” and cancel it
1
3
139
