The Daily Index — Wednesday, February 19, 2025 · TLDR Index

HEADLINES & LAUNCHES

TLDR AI

TLDR AI

MIRA ANNOUNCES THINKING MACHINE LABS

The former CTO of OpenAI, along with many extremely talented scientists and engineers, have joined to make another AI company. The goals are somewhat vague, but it seems to be a product and foundation model company, with a focus on infrastructure.

thinkingmachines.ai · 4 min

TLDR AI

META IS LAUNCHING LLAMACON

Meta is launching LlamaCon, an open-source AI developer conference, on April 29. The event will showcase advancements in the Llama AI model ecosystem, followed by Meta Connect in September for XR and metaverse developments.

meta.com · 1 min

TLDR AI

NATIVE SPARSE ATTENTION: HARDWARE-ALIGNED AND NATIVELY TRAINABLE SPARSE ATTENTION

DeepSeek has entered the Attention Alternative arena with this novel algorithmic method to speed up quadratic Attention. They get as much as an 11x speed up with no loss in overall performance.

arxiv.org · 35 min

TLDR AI

SWE LANCER

A great benchmark of over 1,400 freelance software engineering tasks from Upwork, valued at $1 million USD total in real-world payouts. SWE-Lancer encompasses both independent engineering tasks - ranging from $50 bug fixes to $32,000 feature implementations - and managerial tasks, where models choose between technical implementation proposals. Top models earned $400k.

arxiv.org · 15 min

TLDR AI

GEOMETRIC FOLDING IN RELU NETWORKS

Researchers present a quantitative analysis of how ReLU neural networks fold input space, revealing self-similarity patterns. They introduce a novel metric to study these transformations and provide empirical results on benchmarks like CantorNet and MNIST.

arxiv.org · 22 min

TLDR AI

TASK GROUPING FOR BETTER MULTI-TASK LEARNING

A new approach to multi-task learning mitigates negative transfer by dynamically grouping tasks and updating them sequentially during training. The method, based on proximal inter-task affinity, significantly improves performance over existing multi-task optimization techniques.

arxiv.org · 19 min

TLDR AI

LLM-GUIDED REINFORCEMENT LEARNING

CAMEL improves reinforcement learning efficiency by integrating LLM-generated suboptimal policies with dynamic action masking.

arxiv.org · 6 min

TLDR AI

R1 1776 (HUGGING FACE HUB)

Perplexity has post-trained R1 to remove Chinese censorship. They do so in a way that doesn't harm underlying reasoning. It is Perplexity's first open weights release.

huggingface.co

TLDR AI

GOOGLE'S FLOOD HUB FEATURES

Google is introducing new tools for flood experts in Flood Hub, including an inundation history map and basin view, while collaborating with aid groups like GiveDirectly and the IRC to assist flood-affected communities.

blog.google · 8 min

TLDR AI

GROK 3 OVERVIEW

This article provides a comprehensive overview of xAI's Grok 3.

analyticsvidhya.com · 18 min

TLDR AI

RL A QUICK START GUIDE

Great X article by PufferLib maintainer. It dives into the key differences between types of RL and gives a great guide for base hyperparameters.

x.com · 12 min

TLDR AI

OPENAI CONSIDERING 16 STATES FOR DATA CENTER CAMPUSES AS PART OF TRUMP'S STARGATE PROJECT

OpenAI is considering building data center campuses in 16 states as part of President Trump's Stargate initiative, in partnership with Oracle, SoftBank, Microsoft, Nvidia, and Arm, aiming to invest up to $500 billion over four years.

cnbc.com · 4 min

TLDR AI

ACADEMIC RESEARCHERS FIND A WAY TO TRAIN AN AI REASONING MODEL FOR LESS THAN $50

Researchers at Stanford and the University of Washington have trained an AI reasoning model for under $50 using distillation and modifications to an Alibaba AI model.

techxplore.com · 3 min

TLDR AI

OPENAI NOW REVEALS MORE OF ITS O3-MINI MODEL'S THOUGHT PROCESS

OpenAI has updated the "chain of thought" for its o3-mini model, enhancing transparency in its reasoning steps for both free and paid ChatGPT users.

techcrunch.com · 3 min