ChatGPT-4 Turbo, Custom GPTs: The Most Advanced LLM Gets Big Upgrade!

We all had been waiting for OpenAI’s dev day and it did not disappoint. The unveiling of the GPT-4 Turbo marked a significant milestone in the evolution of state-of-the-art language models, offering enhanced capabilities that far exceed its predecessors. However, the biggest upgrade comes in the form of custom GPTs. Allowing anyone to create AI agents. It seems OpenAI has fully embraced AI agents as the way of interacting with Large Language Models (LLMs) in the future.

GPT-4 Turbo

Let’s start with the biggest upgrades to GPT-4, it has now been upgraded to GPT-4 turbo. In terms of technical specifications we don’t have any details but we did get an overview of the overall model (OpenAI has not released any technical details for original GPT-4).

GPT-4 Turbo details:

1. The context length has gotten a massive upgrade, it now supports 128K tokens. This is by far the largest context window of any LLM available, even beating Claude 2’s 100K. This is the equivalent of 300 pages of text, which will meet the requirements of almost any user.
2. The knowledge cut off date has been increased to April 2023. The previous version knowledge cut off was September 2021. Although this is a welcome increase but we are expecting this to improve much further in the future.
3. GPT-4 turbo is now much cheaper than original GPT-4. This is going to be a welcome news to a lot of developers out there, who are using the GPT-4 API.
4. The new API comes with GPT-4 Turbo vision, DALL-E 3, upgraded Text-to-speech (TTS) which includes six new voices.

ChatGPT is going to be getting the new GPT-4 Turbo model, which means ChatGPT is going to remember more of the conversation. Although I feel like this might come at a cost of model performance.

A very welcoming upgrade for ChatGPT is that all variants of ChatGPT (DALL-E 3, Plugins, Code interpreter, Search) are now going to be one just model. No need to select a specific one, as it will intelligently select the model for you according to your need. This is a very good quality of life upgrade.

GPT AI Agents

The biggest announcement of the entire OpenAI dev day was GPTs. A customisable ChatGPT that you can share with anyone. This concept is in fact not new, they are generally called AI agents, small AI programs that help you achieve a specific task. OpenAI is taking this to another (more accessible) level.

You can simply create a customisable ChatGPT for a specific use case. Earlier we had custom instructions that allowed you to slightly customise your ChatGPT experience but this time you can create entire workflows that can be shared with anyone.

Here are some of the most important features of GPTs:

1. Customization Without Coding: OpenAI has made it possible for anyone to build their own GPT without the need for coding skills. You can simply use natural language to create custom apps/GPTs. This is big selling point as most people don’t know how to code.
2. Sharing and Monetization: Users can now share their custom GPTs publicly, and with the upcoming GPT Store, creators can have their GPTs featured and potentially earn money based on user engagement.
3. Real-World Integration: Developers can connect GPTs to the real world by defining custom actions through APIs. This allows GPTs to interact with external data and perform tasks like managing emails or facilitating e-commerce orders or even more complex tasks.
4. Enterprise Use: Enterprises can create internal-only GPTs for specific use cases or departments, which can aid in tasks like crafting marketing materials or supporting customer service.

Assistants API: Building AI-Powered Applications

The Assistants API is a new offering designed to help developers create AI-powered applications with agent-like capabilities. These agents can be programmed with specific goals, utilize additional knowledge, and call upon various models and tools to perform tasks. This API is a significant step in enabling developers to build more sophisticated and high-quality AI apps.

Here are some of the most important features of Assistant API and their use cases:

Feature	Description	Example Use Case
Code Interpreter	Executes Python code in a sandboxed environment, including graph generation and data processing.	An AI coding assistant that helps debug code, visualize data, and teach programming concepts.
Retrieval	Augments the assistant with external knowledge, such as proprietary data or user-provided documents.	A customer support bot that retrieves product information from a database to answer user queries.
Function Calling	Allows the assistant to invoke custom-defined functions and integrate responses into messages.	A home automation system where the assistant can control smart devices based on user instructions.
Persistent Threads	Supports long and ongoing conversations without the constraint of context window limits.	A medical bot that maintains a continuous dialogue with patients to monitor their health over time.

It’s very interesting to see the rate of progress that AI is making in the world. It’s not even been a year since the release of ChatGPT and we have some incredible advancements already!