{"id":120383,"date":"2024-10-29T01:09:07","date_gmt":"2024-10-28T23:09:07","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/organizacziya-osi-vyrabotala-kriterii-otkrytosti-ai-sistem"},"modified":"2024-10-29T01:09:07","modified_gmt":"2024-10-28T23:09:07","slug":"organizacziya-osi-vyrabotala-kriterii-otkrytosti-ai-sistem","status":"publish","type":"post","link":"https:\/\/prohoster.info\/en\/blog\/news\/organizacziya-osi-vyrabotala-kriterii-otkrytosti-ai-sistem","title":{"rendered":"The OSI organization has developed criteria for the openness of AI systems.","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
The Open Source Initiative (OSI), which verifies licenses for compliance with Open Source criteria, has approved the document Open Source AI Definition v1.0 (OSAID), which outlines the definition of open AI. An AI system can be considered open if it meets the following criteria: <\/p>\n
\n
The ability to be used for any purpose without the need for separate permission;\n
The ability to study the system's workings and inspect its components;\n
The ability to make modifications for any purpose, including changing the information produced by the system;\n
The ability to share both the original and modified versions with others, without restriction on the purposes of use. <\/ul>\n
To allow for modifications, an open AI system must include: <\/p>\n
\n
Detailed information about the data used for training and the training methodology. This information should be sufficient for a professional developer to recreate an equivalent AI system using the same or similar data.\n
Source code that allows both the launching of the AI system and the execution of its training process. The code must also cover areas such as preprocessing, data verification, and tokenization. Additionally, a detailed description of the model architecture must be provided. <\/li>\n
Model parameters (weights), indicating the availability of a ready-to-use snapshot after training or a finalized optimized version of the model. <\/li>\n<\/ul>\n
Large language models of machine learning recognized as meeting the prepared criteria: Pythia (Eleuther AI), OLMo (AI2), Amber (LLM360), CrystalCoder (LLM360), and T5 (Google). <\/p>\n
Large language models of machine learning claiming compliance but requiring modifications to licenses or usage rules: BLOOM (BigScience), Starcoder2 (BigCode), and Falcon (TII). <\/p>\n