OpenAI Releases Fine-Tuning Customizations for GPT 3.5 (1)
OpenAI has launched a new feature that lets businesses fine-tune its GPT-3.5 Turbo model using their proprietary data. The company asserts that these tailor-made models can even surpass GPT-4 in specialized tasks. OpenAI has plans to roll out this customization feature for GPT-4 later this fall. This development allows companies to create a ChatGPT model that excels in specific functions, delivering more accurate and succinct responses.
Before this, companies were confined to GPT-3 variants like davinci-002 or babbage-002 for any customization. The updated model is pre-trained up to September 2021 and ensures client data is not used for external purposes. This innovation aims to help businesses maintain brand voice consistency, whether it's in ad copy or automated code snippets.
Introduced earlier this year, GPT-3.5 Turbo is not just for chat services. It can process up to 4,000 tokens at once, doubling the capabilities of its predecessors. Early adopters have noted a 90% reduction in prompt length when using fine-tuned instructions. The pricing is set at $0.0080 per 1,000 tokens for training, $0.0120 for input, and $0.0160 for output.
No doubt, this is an exciting leap in AI customization as it offers businesses a new level of precision with their custom chatbots. However, the per-token pricing model makes this innovation a costly endeavour for most. Is OpenAI inadvertently creating a paywall for cutting-edge AI?
Microsoft is also in the game, offering customizable GPT-based models through its AI Builder and Power Virtual Agents. These can be integrated with internal company data to produce more targeted responses. The stage is set for fierce competition in the realm of AI customization, offering businesses more tools than ever to automate and personalize their operations.
Â
