Lester Mackey retweeted
Well, the idea seems cool. This uses a coreset idea from approximation theory to "expresss" strong non-causal attention approximation into a causal, streaming one. arxiv.org/abs/2606.10944
1
1
18
1,588
May 26
behold, the coreset
3
257
May 26
We see that mRMR feature selection (a) tends to focus in on medium-difficulty questions; (b) has high feature selection stability (is likely to pick the same coreset questions upon different random seeds/test-train splits); and (c) is super fast!
1
1
4
106
May 26
Viewing benchmark questions as regression features, pretty much all efficient benchmarking techniques use a two stage procedure:
1️⃣ Feature selection: select a coreset of questions.
2️⃣ Prediction: predict full-benchmark scores using just the coreset scores.
1
1
4
155
May 25
stay loyal to your creativity because it's a gift
a year ago i was shipping many apps, some of which were open source
not gonna namedrop but there was one 'solopreneur' from india who copied band4band everything i did:
> i launched textbehindimage, went absolutely viral on X. 2 months later man built the exact same app and did his own sequence of posts on X. copy was similar as well
> i launched a different app called 'brandthetics'. couple weeks later he launched an app (in a different niche) called 'Appthetics
> now ppl are saying he cloned an opensource project and built a product wrapper around it
now imma get myself clear, i do not have anything against copycats
in fact many great companies were, to a particular degree, copycats of something from the past. they would then evolve into their own, developing a strong moat in a specific coreset
they're only limiting their potential, theres nothing i can do about it
as if it wasn't obvious by now, evidence shows this dude still HASN'T grown out of his 'copycat' bubble and started innovating by solving super niche problems with great solutions
despite me calling it out multiple times over the span of a year
otherwise you limit your growth
(credit to whoever i got first pic from sorry i missed your handle)
23 Jul 2025
if you're a first time founder thinking of copying a successful product 1:1
this will NOT get you a seat at the table
one of my apps went insanely viral last year: i built text-behind-image and scaled it to 400K users
every single day i see at least one copycat building the exact same product
the ui is identical, the branding is identical, the social media content is copy-pasted... it's embarrassing lol
(i've even seen copycats of copycats)
this shi doesn't bother me & if your app is getting copied, take it as validation you're doing something right
you'll either outcompete them on distribution or maintain OG status as the first mover
but here's the brutal truth most won't tell you:
respect in this industry comes from building something unique & growing the shit out of it
you'll forever be known as the mf who copied the OG creator band for band - until you learn to pivot & build something that's actually yours
the saddest part? these copycats spend months rebuilding what already exists instead of solving new problems
while they're busy being discount versions of me, i'm already 5 products ahead
stop being a copycat & start being a creator
the market rewards originality - not going 100% band for band
so go and build your own legacy!!!
respect for the big dogs!!!
25
4
118
23,206
May 19
Menos mal que respetan esto. Por otros lado, QEPD Though the Omenpaths, espero que lo vuelvan a usar para sets estilo coreset.
1
3
139
Para fortalecer la seguridad vial y prevenir accidentes en el tránsito, esta jornada se realizó un operativo intersectorial de fiscalización en la Ruta 5-Sur, sector Los Lagartos, en #Mostazal.
@Carab_OHiggins SIAT #Cachapoal, @MTTOhiggins y @rutadelmaipo, integrantes de la Mesa Coreset, se desplegaron en la ruta de sur a norte, para fiscalizar a vehículos de carga y particulares.
En el operativo se detuvo a un sujeto por conducir sin la licencia profesional correspondiente; se cursaron casi una veintena de infracciones en el tránsito por diferentes motivos, varios de ellos por documentación como la licencia de conducir, permiso de circulación o revisión técnica vencidas. Recuerde revisar la documentación antes de conducir y respetar las normas del tránsito.
#OrdenyPatria
4
8
494
May 14
Esta mañana participamos en una nueva sesión de la Comisión Regional de Seguridad de Tránsito (CORESET), instancia presidida por la Seremi de Transportes y Telecomunicaciones, orientada a fortalecer el trabajo intersectorial en materia de seguridad vial y prevención en la región.
3
4
89
Coreset-Induced Conditional Velocity Flow Matching
Xiao Wang, Zihua She, Jianxi Su
arxiv.org/abs/2605.12951 [𝚜𝚝𝚊𝚝.𝙼𝙻 𝚌𝚜.𝙻𝙶]
ALT We propose Coreset-Induced Conditional Velocity Flow Matching (CCVFM), a generative model that augments hierarchical rectified flow with a data-informed source distribution. Hierarchical flow matching models the full conditional velocity law in velocity space, but its inner flow is asked to transport isotropic Gaussian noise to a multimodal target velocity distribution from scratch. Our key observation is that this inner source can be replaced by a closed-form surrogate built from a coreset of the target. CCVFM first compresses the target into weighted atoms using an entropic Sinkhorn coreset and lifts them to a Gaussian mixture. The induced conditional velocity law is then a closed-form Gaussian mixture that can be sampled without a learned neural sampler. A lightweight correction flow, trained from this exact surrogate source, then refines the remaining surrogate-to-target residual rather than learning an entire noise-to-data map. We prove that the surrogate transport cost equals the
5
229
A systematic method to reduce the computational cost of the attention mechanism is communicating the attention between different tokens through a small coreset, only.
1
2
90
May 4
Topology-Aware Active Learning on Graphs arxiv.org/abs/2510.25892
Label coreset construction by Balanced Forman Curvature*
* introduced by Topping (ICLR 2022) arxiv.org/abs/2111.14522
5
306
📌Este martes se realizó la primera Mesa #Coreset encabezada por el Seremi Carlos Navarrete junto a la representante de @conaset regional, para trabajo en coordinación al Mayor de la Siat de @CarabTarapaca, Directora de @Senda_Tarapaca, Tolerancia Cero y equipo de @rutasiquique
3
102
Vol:19 No:6 → BRIEF: Bi-level Coreset Selection for Efficient Instruction Tuning in LLMs vldb.org/pvldb/vol19/p1264-c…
5
582
Rethinking Coreset Selection: The Surprising Effectiveness of Soft Labels
Saumyaranjan Mohanty, Deexitha Vattivella, Konda Reddy Mopuri.
Action editor: Ankit Singh Rawat.
openreview.net/forum?id=Ll78…
#benchmark #deep #labeled
9
545
#Coreset | Durante la jornada de este jueves, el equipo de @conaset Tarapacá fue parte de la Segunda Feria de Seguridad Vial, impulsada por @rutasiquique que tuvo como objetivo educar a la población sobre las normas del tránsito y difundir prácticas de prevención y autocuidado.
2
5
112
they shoud come up with a binder for the waist are... the core area.. i could call it a coreset or something.. i'll make billions off of this idea
36
7,187
Mar 17
Came across "coreset" in a recent Databricks blog: "Decoupled by Design: Billion-Scale Vector Search" 🧐🤔 #Databricks #math #software #engineering #mathematics
2
4
503
That live was embarrassing 😭😭shame wear me kiki’s coreset 😭😭😭😭😭and kiki hmmmmm the desperation on that live was desperating ewwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwwww
#pmxtra
3
1
60
1,710