Oh yes, marvelous paper from @FrankRHutter
. Embarrassingly I only discovered it after I was done, although my super-basic approach (regular rf gini feature importance and fairly naive sampling) actually turned out just fine.
RESEARCH
-
Feature Importance Method Comparison in Machine Learning
By
–
-
Random Forests: A Fundamental Machine Learning Technique Explained
By
–
PS: This isn't a recent addition — I was just reminded about it when I came across Lukas mention in this video, and realised I'd never actually said anything about this. 🙂 Random forests are just so damn cool.
-

ULMFiT optimization: hyperparameter importance via random forest
By
–
When I was optimising ULMFiT, I came up with a trick where I ran lots of ablations and fed all the hyperparams and results to a random forest. That told me which were most important. I told @l2k about it, and @wandb added it to their product! 😀 https://
youtu.be/2QX6jjMt1Eg?si
=icbhCYqFpt4WEhjP&t=1217
… -
Complex Scene Coherency Issues in AI Image Generation
By
–
Still trying to work this out, it seems more complex scenes have worse coherency generally.
-
LLM Objectivity Issues: Reinforcing Beliefs Over Critical Thinking
By
–
Finding myself wishing that LLMs were giving me more objective thoughts, rather than trying to reinforce my beliefs. Every turn I ask Claude to “revisit your assumptions”, it flips back and forth. Almost like it treats the message as an instruction to flip its view.
-
Detecting Cognitive Biases in AI Reasoning and Logic
By
–
You can notice if your preferences and identity mess with your logic by comparing your mode of thinking with reasoning about computer games or alien civilizations. Many brainworms will attack you when you think without identification; you may have to avoid their hosts
-
Entropy Approximation Methods in Machine Learning Models
By
–
all the experiments in the paper approximate Hᴷ(x | θ) using bits under arithmetic coding, which is equivalent to cross-entropy here Hᴷ(x) is approximated using the formula for entropy of random uniform distribution for synthetic, and AC/xentropy of 'reference model' for text
-
AI Biotech Startup by Ex-Meta Researchers Secures $40M Funding
By
–
AI biotech startup founded by ex-Meta researchers raises $40m
#AI #AIio #AIInnovation #ML #DataScience #Futureofwork @lexfridman @sama @kaifulee @ID_AA_Carmack @karpathy @2morrowknight @ylecun -

Yudkowsky’s Philosophy Claims: Solving Millennia-Old Problems?
By
–
The funny thing is that everything Yudkowsky ever wrote about e.g. free will or metaethics was at roughly this level of understanding and yet he convinced a generation of software engineers that he had solved philosophical problems that had confounded humanity for millennia. x.com/MostlyMonkey/s…
-
From Forums to Ephemeral Chats: Digital Preservation Crisis
By
–
I regret the turn from forum posts which lasted ~forever to group chats with N day expiry, partially for this reason. Even just a handful of years later one is a primary source historical document and the other might be paraphrased imperfectly in someone’s blog next year maybe.