Sam Altman opens the first OpenAI developer conference, held on November 6 in San Francisco. Photo credit: OpenAI/YouTube
OpenAI introduced GPT-4 Turbo at its first developer conference today, describing it as a more powerful and cost-effective successor to GPT-4. The update boasts improved context handling and the flexibility to fine tune to meet user requirements.
GPT-4 Turbo is available in two versions: one focuses on text, while the other also processes images. According to OpenAI, GPT-4 Turbo is “optimized for performance,” and priced at just $0.01 per 1,000 text tags and $0.03 per 1,000 image tags – nearly one-third the price of GPT-4.
ChatGPT on demand for you
How does this fine-tuning feature make the GPT-4 Turbo so special?
“Fine-tuning improves multi-shot learning by training on many more examples than can fit in a prompt, allowing you to achieve better results on a large number of tasks,” explains OpenAI. Essentially, fine-tuning fills the gap between general AI models and customized solutions tailored to specific applications. It promises “higher quality hinting results, character savings from shorter hints, and faster query responses.”
Fine-tuning involves feeding extensive user data to the model to learn specific behaviors, turning large generic models like GPT-4 into specialized tools for niche tasks without having to build an entirely new model. For example, a model tuned for medical information will provide more accurate results and “talk” more like a doctor.
do GPT: https://t.co/R9VsQboTli
– Sam Altman (@sama) November 6, 2023
A good analogy can be seen in the world of image generators: fine-tuned stable diffusion models typically produce better images than the original stable diffusion XL or 1.5 because they have learned from specialized data.
Prior to this innovation, OpenAI allowed limited modifications to the behavior of its LLMs via custom instructions. This was already a significant leap in quality for those seeking customization in OpenAI models. Fine-tuning enhances it by introducing new data, tone, context, and voice into the model dataset.
The value of fine tuning is significant. As artificial intelligence becomes an increasingly integral part of our daily lives, there is a growing need for models tuned to specific needs.
“Fine-tuning OpenAI’s text generation models can make them better for specific applications, but requires a careful investment of time and effort,” OpenAI notes in its official guide.
The company has consistently improved its models in terms of context, multimodal capabilities, and accuracy. With today’s announcement, these capabilities are unmatched among mainstream closed-source LLMs like Google’s Claude or Bard.
While open-source LLMs like LlaMA or Mistral can be fine-tuned, they can’t measure up in power and professional usability.
The release of GPT-4 Turbo and the emphasis on fine-tuning marks a significant shift in AI technology. Users can expect more personalized and efficient interactions with potential impact spanning from customer service to content creation.
