2 ms·
PaLM was a base model without any fine-tuning, itself a copy of GPT-3. Maybe you meant LaMDA, which was based on pure instruction fine-tuning without RLHF.
by cubefox 3y ago
PaLM was a base model without any fine-tuning, itself a copy of GPT-3. Maybe you meant LaMDA, which was based on pure instruction fine-tuning without RLHF.
- sinuhe69 3y agoSorry but your claim is so ridiculous. PaLM employs a different model as ChatGPT. And Google also published their result in April 2022, earlier than OpenAI talked about their ChatGPT in late 2022.
- HeatrayEnjoyer 3y agoI don't really care which tech corp did what first, just want to point out GPT-4 was being beta trialed to some bing users in mid 2022.
- cubefox 3y agoPaLM was basically a larger version of GPT-3. It was not similar to ChatGPT. It was a pure token predictor. It was not fine-tuned for chat.
- sinuhe69 3y agoPaLM use a pathway architect, not a decoder only transformer as ChatGPT. At least you can look it up on Wikipedia before making such wild assumptions.