OpenAI e ChatGPT
OpenAI…che roba…giusto un recap di alcuni punti che ho messo su per una presentazione interna.
- OpenAI is an AI research and deployment company
- OpenAI’s mission is to ensure that artificial general intelligence (AGI) benefits all of humanity
- Currently valued $20B, but ChatGPT is set to push the valuation up to $29B
- Key investors: Amazon Web Services, Elon Musk, Microsoft
- Microsoft invested $1B on OpenAI in 2019 (Rumors about integrating GPT-3 into M365 Ecosystem)
- JUST IN Microsoft plans to invest $10B in OpenAI for a 49% stake
- JUST IN Microsoft releases VALL-E that can simulate anyone’s voice with 3 seconds of audio
- ChatGPT is a model developed and trained by Open AI which interacts in a conversational way
- ChatGPT can answer follow-up questions, admit its mistakes, challenge incorrect premises, etc..
- Sibling of InstructGPT, another model trained to follow an instruction in a prompt
- GPT-3 is working behind the scenes for task such as: translation, summarization, classification
- GPT-3 has 175 billion ML parameters. Trained on NVIDIA V100. Using A100s (faster than V100) would have taken 1,024 GPUs, 34 days and $4.6million to train the model.
- Other products part of OpenAI portfolio are DALL-E, Codex (that powers GitHub co-pilot)
- Everything is served via REST API with a billing model based on “tokens”
- Command line interface available that allows quick testing and development