TaskWeaver: A Code-First Agent Framework arxiv.org/abs/2311.17541

#taskweaver Are you tired of working with cumbersome text strings for data analytics tasks? Say hello to TaskWeaver - the innovative code-first agent framework that streamlines planning and execution. (1/n) Checkout our repo at aka.ms/taskweaver

With rich data structures, customized algorithms, and easy integration of domain-specific knowledge, TaskWeaver is your ticket to efficient and seamless data analytics. (2/n)

It also offers stateful conversation, easy debugging, security considerations, and easy extension. Experience the TaskWeaver revolution and simplify your data analytics tasks today! (3/n)

AI-backed virtual assistants face challenges in handling complex data structures. TaskWeaver helps users build assistants that understand diverse domain questions, follow examples, & efficiently execute customizable algorithms on complex data structures. msft.it/6012imeZY

nice work, it would be great if it could have more Windows test cases😋

Our recent work on leveraging LLMs to generate query code in production environment will be presented at ICSE 2024. Joint work with colleagues in Microsoft and my former intern Yuxuan Jiang from UMich.

Llama3 was trained on 15 trillion tokens of public data. But where can you find such datasets and recipes?? Here comes the first release of 🍷Fineweb. A high quality large scale filtered web dataset out-performing all current datasets of its scale. We trained 200+ ablation

Large Language Models don't reason. Thank you, Apple.

Maybe the next work is to use an agent to convert ideas into code/experiments and papers. 😂😂

LLM-brained GUI Agents Great survey on LLM-brained GUI agents, including techniques and applications. arxiv.org/abs/2411.18279

Overwhelmed by the flood of recent studies on GUI agents? Dive into our latest work, "Large Language Model-Brained GUI Agents: A Survey", by far the most comprehensive and insightful survey in the GUI agent domain—don't miss out! 😊

good reading material for the new year holiday😃

4/ 🧵 UFO2: The Desktop AgentOS UFO2 is a multiagent operating system for Windows that enhances desktop automation through deep OS integration, a hybrid GUI-API action layer, and a non-disruptive Picture-in-Picture interface, significantly improving reliability and efficiency

so thrilled to share our latest work on making SWE-Bench live! 🥳

Interesting work. Real feedback always provides the best guidance.

How much will GPT-5 get on SWE-bench Verified, pass@1, any scaffolding any budget?

nice work, remind me of the work "ifevalcode": ifevalcode.github.io

Very informative blogpost!

Coding agents keep reading many code files to understand the repository, increasing the context and cost😪. In SWE-Pruner, we propose a lightweight adaptor that can "pick out the code snippet of interest" , thereby reducing the cost significantly with minimum performance drop.