DATA

→ View original post on X — @tunguz

Training XGBoost Models for Machine Learning

By

@tunguz

–

03 June 2025 19h29

Like training XGBoost for instance.

LLM Training: Fixed Capacity Model Reduces Per-Sample Memorization

By

–

03 June 2025 18h54

hard to say since we mostly study the average case not the worst case. but I think “filling a fixed capacity” is a useful model for LLM learning, which implies that training on more data will force models to memorize less per-sample

→ View original post on X — @mit_csail

Nine Distance Measures in Data Science with Algorithms

By

@mit_csail

–

03 June 2025 18h00

9 distance measures in data science w/algorithms (v/
@MaartenGr
).

→ View original post on X — @aibreakfast

AI agents scrape web and download YouTube videos offline

By

@aibreakfast

–

03 June 2025 17h01

Agents can scrape the web and download files now.

You can even download YouTube videos and save them locally offline with this AI Agent @genspark_ai

Genspark's download Agent can rip basically anything off of the internet and save it, even when it shouldn't be able to.

Very… pic.twitter.com/skq0xfCNFC
— AI Breakfast (@AiBreakfast) 3 juin 2025

Agents can scrape the web and download files now. You can even download YouTube videos and save them locally offline with this AI Agent @genspark_ai Genspark's download Agent can rip basically anything off of the internet and save it, even when it shouldn't be able to. Very

→ View original post on X — @tunguz

Kaggle Competitions Grandmaster: Real World Data Application

By

@tunguz

–

03 June 2025 16h32

Indeed, probably the best way to *directly* apply those skills to the real world data. And the reason I became a Kaggle *competitions* grandmaster. 😉

Test Examples Extractability and Privacy in Saturated AI Models

By

–

03 June 2025 16h08

– hrunning these experiments in a clean setting with perfectly deduplicated texts tells us a lot about privacy: – once capacity is sufficiently saturated, the **test examples** are slightly more extractable than the training examples — maybe extraction is a bit of a myth?
– the

Model Memorization Fixed Capacity Training Saturation Analysis

By

–

03 June 2025 16h08

we train all of our models until they "saturate" which usually happens around 1M steps using a very large batch size models memorize the same amount, regardless of training datasize meaning they have fixed capacity and instead "spread it thinner" when trained on more examples

Measuring Model Memorization with Random Uniform Strings

By

–

03 June 2025 16h08

this all started from a quest to come up with a proper measurement of model memorization it's hard to compute *per-example* memorization, because models "share" info between datapoints so we start with random uniform strings, where sharing isn't possible. and we get this:

→ View original post on X — @sassoftware

SAS Hackathon 5th Year Global Competition Launches June

By

@sassoftware

–

03 June 2025 15h00

It is coming up on that time again… the annual #SASHackathon kickoff celebration! For our 5th year, we're bringing coders, data scientists and analytics gurus together globally to compete in our month-long event. Join us online June 11 to get registration details and hear