View all AI news articles

OpenAI Enhances GPT-3.5 Turbo with Fine-Tuning Capabilities

August 22, 2023

OpenAI has introduced a significant update to GPT-3.5 Turbo, allowing customers to bring custom data to this lightweight version of GPT-3.5. This enhancement makes it more convenient to increase the AI model's reliability while instilling specific behaviors.

According to OpenAI, the fine-tuned versions of GPT-3.5 can match or even surpass the base capabilities of GPT-4, their flagship model, on certain specialized tasks. This update has been eagerly awaited by developers and businesses looking to create unique experiences for their users.

Customization and Scalability

With fine-tuning, companies can now make GPT-3.5 Turbo more instruction-following through OpenAI's API. This includes responding in a specific language or improving the model's consistency in formatting responses. For example, it can be fine-tuned for completing code snippets or adjusting the tone to fit a brand's voice.

Fine-tuning also allows OpenAI customers to reduce their text prompts, speeding up API calls and cutting costs. OpenAI claims that early testers have managed to reduce prompt size by up to 90% by incorporating instructions into the model itself.

The Fine-Tuning Process and Costs

The fine-tuning process involves preparing data, uploading necessary files, and creating a fine-tuning job via OpenAI's API. All data must pass through a moderation system to ensure compliance with OpenAI's safety standards. The company also plans to launch a fine-tuning UI in the future.

The costs for fine-tuning are broken down as follows:

  • Training: $0.008 per 1K tokens
  • Usage input: $0.012 per 1K tokens
  • Usage output: $0.016 per 1K tokens

For example, a GPT-3.5-turbo fine-tuning job with a training file of 100,000 tokens would cost around $2.40.

Additional Updates and Future Plans

OpenAI has also made available two updated GPT-3 base models, with support for pagination and more extensibility. The original GPT-3 base models are set to retire on January 4, 2024.

Furthermore, OpenAI announced that fine-tuning support for GPT-4, capable of understanding images in addition to text, will be released later this fall.

Implications for Collaboration Platforms like Stork.AI

The advancements in fine-tuning and customization of AI models like GPT-3.5 Turbo can have significant implications for collaboration platforms like Stork.AI. By leveraging these AI capabilities, platforms like Stork can enhance their recording and playback functionality, where all voice and video communications are automatically transcribed and summarized. This aligns with Stork's mission to provide seamless collaboration through video conferencing, team collaboration, and screen recording, all in one tool.

Recent articles

View all articles