OpenAI Improves GPT-3.5 Turbo Capabilities, Equivalent To GPT-4

JAKARTA - OpenAI has just perfected the GPT-3.5 Turbo, marking its capabilities will be commensurate with the GPT-4 which is now available to developers.

The company revealed that developers can adapt models that perform better in their use cases and run those custom models on a large scale.

"Initial testing shows that a refined version of the GPT-3.5 Turbo can match, or even outperform, the basic capabilities of the GPT-4 level in certain narrow tasks," OpenAI said in its announcement, quoted Wednesday, August 23.

OpenAI explained, fine adjustments allow businesses to make models better follow instructions, such as making short outputs or always responding in certain languages.

For example, developers can use customizations to ensure the model always responds in German when asked to use the language.

OpenAI also increases the model's ability to self-consistently format responses, an important aspect for applications that require certain response formats, such as encoding codes or making App Programming Interface Calls (API).

Furthermore, a special tone is also perfected to hone the qualitative impression of the model's output, so that it is more in accordance with the sound of the business brand and consistent with its style.

In addition to improving performance, developers can shorten their demand while ensuring similar performance. Improvements with GPT-3.5 Turbo can also handle 4k tokens, double the previous OpenAI refinement model.

"Initial testers have reduced the prompt size by up to 90 percent by fine-tune the instructions into the model itself, speed up each API call and cut costs," OpenAI said.

Like all of OpenAI's APIs, data sent in and out of the adjustment API belongs to customers and is not used by the AI startup, or any other organization, to train other models. Improvements for GPT-4 will be present this fall.