Microsoft AI Research uses GLAN to teach LLMs just like kids in school
A new way of teaching LLMs is available, and it is efficient and flexible
2 min. read
Published on
Share this article
Read our disclosure page to find out how can you help Windows Report sustain the editorial team Read more
Generalized Instruction Tuning, or GLAN, is Microsoft’s new training method for AI. Furthermore, GLAN is similar in some ways to our education system. After all, it breaks data into chunks, helping the AI understand it faster. In addition, the Generalized Instruction Tuning has a range of studies, difficulties, and disciplines.
What is instruction tuning in language models?
Instruction tuning is a training model that focuses on input and output. For example, you can ask the AI to generate a list of 5 of the most visited places in your city. Afterward, it will search for them and list them accordingly. In addition, it is a specialized form of fine-tuning. Furthermore, according to Medium, large language models (LLMs) receive tasks such as writing emails, sentence editing, etc. However, GLAN takes this training to the next level.
GLAN breaks down data into domains, sub-domains, and disciplines with help from LLMs and us. Furthermore, it divides data into subjects and creates a syllabus for each. In addition, GLAN uses this style to generate tasks and instructions for the LLMs. Also, this approach allows LLMs to learn faster with fewer limitations. On top of that, this training method is more precise because it uses the input and output system.
The Generalized Instruction Tuning training is flexible and scalable. Also, GLAN doesn’t have to recreate datasets to implement new skills or domains. Furthermore, the Generalized Instruction Tuning generated a great array of instructions for the LLMs, and it uses prompts created and verified by large language models.
On top of that, according to MarkTechPost, GLAN provided excellent results in various subjects, such as coding, mathematical reasoning, tests, and general instructions. Moreover, it doesn’t need specific training data.
Ultimately, GLAN is an effective and reliable training method for the LLMs. Furthermore, it allows the customization of data without starting it from scratch. Thus, GLAN is quite flexible, and it helps LLMs to learn data methodically and faster than before.
If you want to learn more check out the Synthetic Data research.
What are your thoughts? Do you want to learn more about GLAN? Let us know in the comments.
