Featured image
OpenAI

OpenAI Secretly Releases GPT 3.5 Turbo Instruct

avatar

Sven

September 20th, 2023

~ 2 min read

OpenAI, has released its latest language model, gpt-3.5-turbo-instruct. This new model is designed to efficiently process and follow specific instructions, making it a powerful tool in various text-based tasks.

Replacing the existing Instruct models, gpt-3.5-turbo-instruct maintains the same cost and performance as other GPT-3.5 models but comes with enhanced capabilities. Unlike traditional chat models, this model excels in directly answering questions and completing text tasks swiftly, aligning perfectly with its purpose of effectively following instructions.

OpenAI emphasizes that gpt-3.5-turbo-instruct surpasses its predecessor, GPT-3.5, in multiple aspects. It exhibits better comprehension of complex instructions, generates higher-quality output, and operates at a faster pace and lower cost. These advancements make it an invaluable tool for users seeking accurate and efficient results.

Instruct models, including the newly released gpt-3.5-turbo-instruct, play a pivotal role in the breakthroughs seen in ChatGPT. These models have undergone continual refinement through human feedback during and after pre-training, enabling them to better understand and respond to human queries with fewer errors and less harmful content. Despite being smaller in parameters compared to regular GPT models, they offer superior performance and reliability.

In January, OpenAI introduced the instruct GPT models, aiming to reduce hallucinations and produce more truthful and less toxic responses. Before this update, GPT-3 was trained to predict the next word based on a large dataset of internet text, leading to occasional fabrication of facts. With the introduction of InstructGPT, OpenAI addressed this issue and made significant progress in aligning the model with users' expectations.

The key differentiator of gpt-3.5-turbo-instruct is its shift towards a more task-oriented approach instead of conversational AI. Unlike its chatty counterparts, this model focuses on following instructions without requiring additional prompting techniques. It efficiently completes tasks and provides accurate answers, making it an invaluable tool for various industries and applications.

Links:
https://platform.openai.com/docs/models/gpt-3-5
https://openai.com/research/instruction-following