Meta has released a potentially game-changing AI model purely trained on AI-generated data, aiming for self-taught AI improvement.
Meta has been working swiftly on bringing new improvements to its AI models to enable advanced task performance and simpler human interactions. In a shocking move, Meta has silently released a new AI model that can evaluate other AI models and conduct continuous self-teaching processes to improve itself.
Meta’s effort in this angle is derived from its vision of minimizing human intervention in various business and personal processes. It also focuses on reducing the reliance on human-labeled data for training and improvement. The success of Meta’s new AI model will help boost self-improving and autonomous AI models capable of completing a wide range of tasks independently.
“We hope, as AI becomes more and more super-human, that it will get better and better at checking its work so that it will actually be better than the average human. The idea of being self-taught and able to self-evaluate is basically crucial to the idea of getting to this sort of super-human level of AI.”
- Jason Weston, Researcher at Meta
As AI models evolve, they are getting better at correcting themselves and ensuring higher accuracy and correctness of the information shared and tasks performed. Meta’s competitors, including the OpenAI-01 model, have already begun developing self-teaching techniques to generate reliable and better-set responses.
Following its introduction in an August paper, Meta’s research teams have fast-tracked the development of reasoning capabilities of AI LLMs without human input. The new AI model acts as a self-taught evaluator not only for interactions with consumers but also with other AI models. This model can even train other AI models at a faster rate of learning than human intervention can facilitate, unlocking the automatic execution of efficiency and scalability improvements.
The AI model is prompted to break down problems in science, maths, and coding into smaller, manageable, and practical steps that it can autonomously prompt itself to execute. Along with the self-taught evaluator, Meta has also brought an important update to its Segment Anything Image Identification model to accelerate response times. What sets the new update apart from the previous ones is the new capability of using existing datasets to aid the discovery of new inorganic materials. The self-evaluation capability can unlock many new breakthroughs in the AI landscape, most importantly, in building superhuman intelligence that surpasses average human intelligence and reasoning.
