OpenAI Quietly Launches GPT-3.5 Models Amid GPT-4 Anticipation
OpenAI quietly releases GPT-3.5, an upgraded foundation model that powers the new ChatGPT demo and the text-davinci-003 API. The new model family produces higher-quality writing and hallucinates less than its predecessor.
OpenAI quietly introduces GPT-3.5, an unannounced upgraded version of its GPT-3 model, just as anticipation builds for the eventual release of GPT-4. This new foundation model makes its public debut through ChatGPT, a general-purpose chatbot capable of discussing complex topics like programming, television scripts, and scientific concepts.
Rather than releasing the base GPT-3.5 model directly to the public, OpenAI uses it to power several fine-tuned systems available through its API, including text-davinci-003. This specific model handles more complex instructions and produces measurably better long-form writing than previous GPT-3-based systems.
Early testing indicates that GPT-3.5 scores higher on human preference ratings and hallucinates less frequently, though OpenAI warns that important limitations still remain. Data scientists note that the model performs better at understanding the context behind prompts, resulting in more accurate and higher-quality content generation overall.