Meta Ai introduces MLGYM: New Ai Framework and Benchmark of Ai agents Agents

The desire to speed up scientific discovery with AI has been looking forward to the Oak Ridge appearance for automatic pipelines returned back in 1979. Revitalizing AIs who work well for book reviews, creating hypotheses, construction tests, analyze the effects, and produce scientific documents. Additionally, they can submit the scientific fluctuations by changing repeated tasks, allowing investigators to focus on the Higher Higher Work. However, despite these promises, the study of AI maintenance is always a challenge because of lack of common benches can properly evaluate their various skills in all its various skills.

Recent studies have experienced this space by informing the AIs testing benches in various software operations and machine-language learning activities. While the framework is in the AI ​​agents tests in well-defined problems such as the production and making models, many benches do not fully support open research challenges, where many solutions arise. In addition, these structures often lack fluctuations on a variety of research heights, such as algorithms with the novel, model structures, or predictions. To promote AI, there is a need for testing programs that include broadcasting jobs, facilitate the exams with different learning algoriths, and to accommodate various methods of research donations. By establishing such such structures, the field may improve closely to appreciate the AI ​​systems that are independent of independence.

Investigators from the University College London, University of Wisconsin-Madison, University of Oxford, Meta, and other facilities launched a new framework and the test bench and improving the Christian agents. This program, the first ML National Gym functions, helps study the RL training strategies for Agents AI. Bench, MLGYM-BENCH, includes 13 Possibly completed jobs, NLP, RL, and Game Theory, requires true research skills in the world. The framework separates the skills of the agent resean agent, with MLGYM-BENCH to focus on Level 1: Baseline Development, where the llMs is preparing for scientific models but does not have scientific documents but not.

MLGYM The system is designed to evaluate and develop lm agents in ML research activities by allowing the nature of the shell using consecutive instructions for consecutive instructions. It contains four important components: agents, nature, datasets, and jobs. Agents issued the Bash instructions, manage history, and integrate foreign models. Nature provides a secure work environment based on the docker with controlled access. Databases are described separately and activities, allowing re-use in all exercises. Activities include assessment and configuration documents of various ML challenges. Additionally, MLGYM offers the search for books, memory stores, and effective verification, to ensure effective conversion and harmonious performance of AI.

This lesson uses the SWE-agent model designed for MLGYM nature, following a loop to make decisions made in style. Amamodeli amahlanu-e-art-of-the-art-aurt-art, Gemini 1.5 Pro, Claude-3.5-Sonnet, LLama-3-405B-ukufundisa, kanye ne-GPT-4O-ahlolwe ngaphansi kwezilungiselelo ezijwayelekile. Working tested using AUP Scores and operating profiles, comparing models based on the best effort and the best import metrics. Openai O1-Preview has access to the highest performance, by Gemini 1.5 Pro and Claude-3.5-Sonnet followed. Studies highlight effective profiles as an effective method of assessment, indicating that Open O1 previews reach between high models in various activities.

In conclusion, the lesson emphasizes the chances and challenges of using llms as a scientific agency. MLGYM and MLGYMBEZUNG DISEMS Agreement in converts to various measurements of measuring but expressing improvement. Expanding across ML, the general integrated assessment, as well as assessing the good scientific personality are important growth facilities. The lesson emphasizes the importance of data openness to improve interaction and attainment. As AI research continues, development in the consultation, agent buildings, and assessment methods will be important. InterdiciIplinary cooperation can ensure that AI agents accelerates scientific discovery while maintaining recycling, verification and integrity.

Survey Page and GitHub paper. All credit for this study goes to research for this project. Also, feel free to follow it Sane and don't forget to join ours 80k + ml subreddit.

🚨 Recommended Recommended Research for Nexus

Sana Hassan, a contact in MarktechPost with a student of the Dual-degree student in the IIit Madras, loves to use technology and ai to deal with the real challenges of the world. I'm very interested in solving practical problems, brings a new view of ai solution to AI and real solutions.