Reflection 70B

Open-source language models are usually judged on size and speed, but Reflection 70B is pitched on self-correction.

Reflection 70B is an open-source AI language model built on the Llama 70B architecture that handles chat, writing and coding help.

Its makers say the self-correction technique lets it surpass GPT-4 on complex tasks.

That is a big claim, so use it as the project's own and test the model on your own work before you rely on it.

As an AI chatbot, Reflection 70B holds human-like conversation and answers questions on a wide range of topics.

It is also tagged as a code assistant and text generator, so you can use it for drafting documents or getting programming help.

The emphasis is on accuracy for harder tasks rather than quick answers.

Because the weights are open, developers can inspect how the model works and build on it, which is the main reason to pick it over a closed chatbot.