GPT-4 large language model

GPT-4 (Generative Pre-trained Transformer 4) is a large language model that uses deep learning to generate natural, human-like language from text and image prompts. 

Developed by OpenAI and released in March 2023, GPT-4 underpins Microsoft's Bing Chat (renamed Copilot) and is available on a subscription basis from OpenAI as ChatGPT Plus.

GPT-4 has been praised as a major improvement on its GPT-3 predecessor in terms of its multi-modal and basic reasoning capabilities, the latter of which is seen to benefit from the model's larger size and increased parameters. 

In its technical report, OpenAI claims GPT-4 is 82 percent less likely than GPT-3.5 to respond to requests for unsafe content, and 60 percent less likely to make stuff up or 'hallucinate'. 

Transparency and accountability 🙈

• Upstream. OpenAI provides no information about GPT-4 data sources and size, nor about third-party copyrighted data or data licenses it may have used to train it, or about the companies it may have outsourced data annotation to. Neither does it disclose any information about the energy used or carbon emissions emitted during training, or how OpenAI handles the rights of third-party copyright holders.

• Model. While OpenAI does provide some access to third-parties to analyse its 'black box' model, it does not disclose information about GPT-4 model components, size or architecture. And while the company discloses basic information about known risks, it does not disclose information about unintentional harms that may be caused.

• Downstream. OpenAI provides no direct mechanism for user feedback, complaints or appeals. It also fails to provide information on GPT-4's impacts.