Login

News on Medial

Meta Research Introduces Revolutionary Self-Rewarding Language Models Capable of GPT-4 Level Performance

MedialMedial · 10m
Meta Research Introduces Revolutionary Self-Rewarding Language Models Capable of GPT-4 Level Performance

Language models can achieve superhuman capabilities with the new paradigm of Self-Rewarding Language Models (SRLMs). Meta has developed an innovative system where the model generates its own rewards, leading to continual improvement in instruction following and reward modeling abilities. The iterative Direct Preference Optimization (DPO) framework is used to train SRLMs, allowing the model to push itself to superhuman levels. After just three iterations, the SRLM outperformed state-of-the-art systems, demonstrating its potential to achieve GPT-4 level performance. SRLMs break free from fixed reward models, offering continuous improvement possibilities for language models.

6
View Source

Comments

More Like this

“The king is dead”—Claude 3 surpasses GPT-4 on Chatbot Arena for the first time

“The king is dead”—Claude 3 surpasses GPT-4 on Chatbot Arena for the first time

Researchers made an IQ test for AI, found they're all pretty stupid

Researchers made an IQ test for AI, found they're all pretty stupid

ETtech Explainer: Is Apple’s ReALM better than OpenAI’s GPT-4?

ETtech Explainer: Is Apple’s ReALM better than OpenAI’s GPT-4?

AI Is Becoming More Powerful—but Also More Secretive

AI Is Becoming More Powerful—but Also More Secretive

Meta’s “massively multilingual” AI model translates up to 100 languages, speech or text

Meta’s “massively multilingual” AI model translates up to 100 languages, speech or text

This AI stock trader engaged in insider trading — despite being instructed not to – and lied about it

This AI stock trader engaged in insider trading — despite being instructed not to – and lied about it

You can now train ChatGPT on your own documents via API

You can now train ChatGPT on your own documents via API

Uh oh — it looks like ChatGPT's AI model got lazy again

Uh oh — it looks like ChatGPT's AI model got lazy again

1960s chatbot ELIZA beat OpenAI’s GPT-3.5 in a recent Turing test study

1960s chatbot ELIZA beat OpenAI’s GPT-3.5 in a recent Turing test study

Sarvam AI to launch voice-to-voice endpoint tool in the next six to twelve months

Sarvam AI to launch voice-to-voice endpoint tool in the next six to twelve months

Download the medial app to read full posts, comements and news.