A new AI chatbot model called "Reflection 70B" has been introduced and promises to solve a problem that plagues many AI models: hallucinations.
Reflection tuning: learning from mistakes
The model was trained using reflection tuning, a technique that allows AIs to learn from and correct their mistakes.
Matt Shumer, CEO of HyperWrite AI, calls Reflection 70B "the world's best open source model." It was developed on top of Llama 3.1, an open source AI model from Meta, and is said to be able to compete with closed models such as Anthropic's Claude 3.5 Sonnet and OpenAI's GPT-4o in several benchmarks.
Hallucinations: A problem for AI models
AI hallucinations occur when an AI chatbot perceives patterns or objects that do not exist or are not perceptible to human observers, resulting in incorrect outputs.
Reflection tuning aims to solve this problem by enabling AIs to detect and correct their own errors before providing an answer.
How does reflection tuning work?
The AI analyzes its own spending and learns from it, identifying strengths, weaknesses, and areas for improvement. This process is repeated, allowing the AI to continuously improve its capabilities and become more aware of its own spending.
OpenAI's approach to fighting hallucinations
OpenAI, backed by Microsoft, published a research paper in 2023 presenting ideas for preventing AI hallucinations.
One idea is "process monitoring," where AI models are rewarded for rewarding each correct step of reasoning in developing an answer, rather than just rewarding a correct conclusion.
"Detecting and mitigating a model's logical errors or hallucinations is a critical step toward developing aligned AGI [artificial general intelligence]," Karl Cobbe, a researcher at OpenAI, told CNBC.
Reflection 70B: A promising approach
"Reflection 70B" could be an important step towards more reliable and accurate AI models. The ability to learn from mistakes is crucial for developing AI systems that can truly benefit people.
#Reflection70B #News #Haberler #Noticias #Nachrichten
