Tessera
Folgen

🧪🖌🪄

@scienceartmagic.bsky.social

Open Source AI/ML Research https://github.com/ScienceArtMagic https://hf.co/ScienceArtMagic https://www.youtube.com/@ScienceArtMagic

175 Posts391 Follower3829 Folgt
🧪🖌🪄@scienceartmagic.bsky.social · 7mo ago

AI/ML folks: Can we finally stop pretending that pseudo-intellectual, pro-Putin mouthpiece (and apparently pedo advisor even if not one himself - which is highly unlikely given his ties) Noam Chomsky has *anything* to offer the ML world?

Jeffrey Epstein email sharing advice from Noam Chomsky to Epstein.
Lennart Nacke, PhD@lennartnacke.com · 8mo ago

Forcing PhD students to publish before thesis submission created a predatory journal gold rush. Too many couldn't get into real journals fast enough. So they paid to publish garbage. Quality PhD work produces quality papers. Not the other way around.

🧪🖌🪄@scienceartmagic.bsky.social · last yr.

Biggest flopper in #NBA history says modern NBA is too soft: sports.yahoo.com/article/john... #NBAsky

John Stockton admits he is not watching the NBA anymore because it's way too soft: "Fans want you to go out there and do what they can't do, not go out there, shake hands and hug"

Hall of Famer John Stockton says load management, softer competition and the lack of true rivalries have drained his interest in watching today’s NBA.

sports.yahoo.com

🧪🖌🪄@scienceartmagic.bsky.social · last yr.

When Claude responds with "You're absolutely right," humans bemoan LLM sycophancy. Humans when queried by anyone who might help promote their paper or fund their product/connect them to funding: "That's a great question..."

🧪🖌🪄@scienceartmagic.bsky.social · last yr.

Humans when reading "delve," "multifaceted," em dash, etc.: "AI slop." Humans any time ML academia, investors, or just some random techbro asks something: "That's a great question..."

Sumit@reachsumit.com · last yr.

Are Optimal Algorithms Still Optimal? Rethinking Sorting in LLM-Based Pairwise Ranking with Batching and Caching Reveals that traditional sorting algorithm efficiency rankings break down in LLM contexts. 📝 arxiv.org/abs/2505.24643

Are Optimal Algorithms Still Optimal? Rethinking Sorting in LLM-Based Pairwise Ranking with Batching and Caching

We introduce a novel framework for analyzing sorting algorithms in pairwise ranking prompting (PRP), re-centering the cost model around LLM inferences rather than traditional pairwise comparisons. Whi...

arxiv.org

Sumit@reachsumit.com · last yr.

Pangu DeepDiver: Adaptive Search Intensity Scaling via Open-Web Reinforcement Learning Huawei introduces an RL framework that enables LLMs to dynamically scale search frequency and depth based on information demands in real-world web environments. 📝 arxiv.org/abs/2505.24332

Pangu DeepDiver: Adaptive Search Intensity Scaling via Open-Web Reinforcement Learning

Information seeking demands iterative evidence gathering and reflective reasoning, yet large language models (LLMs) still struggle with it in open-web question answering. Existing methods rely on stat...

arxiv.org

Dimitri Meunier@dimitrimeunier.bsky.social · last yr.

🚨 New paper accepted at SIMODS! 🚨 “Nonlinear Meta-learning Can Guarantee Faster Rates” arxiv.org/abs/2307.10870 When does meta learning work? Spoiler: generalise to new tasks by overfitting on your training tasks! Here is why: 🧵👇

Nonlinear Meta-Learning Can Guarantee Faster Rates

Many recent theoretical works on \emph{meta-learning} aim to achieve guarantees in leveraging similar representational structures from related tasks towards simplifying a target task. The main aim of ...

arxiv.org