Everyone will soon be able to create custom versions of ChatGPT without having to code anything, and even share and upload variations to a special online store. Another revolution for AI!
Unveiled to the general public just a year ago, ChatGPT revolutionized generative artificial intelligence and launched a real AI race in the tech sector. Realize it! A text-based chatbot capable of providing articulate and comprehensive answers on just about any topic, in natural language! Since then, all the giants have launched their own bots and large language models (LLM), whether Microsoft, Google, Meta or even Elon Musk with X. But OpenAI, the company behind ChatGPT, seems determined to stay ahead ahead of its competitors, while the chatbot has just reached 100 million weekly users!
During DevDay – the first OpenAI developers conference in San Francisco – which was held on November 6, 2023, the company announced the launch of a platform allowing the creation of personalized versions of ChatGPT for use cases. specific use, without any coding being necessary. These agents, called GPTs, will be accessible in the coming weeks on a GPT Store. The firm’s language model will also benefit from significant improvements with GPT-4 Turbo, faster, more efficient, and less expensive. Enough to widen the gap between OpenAI and the competition!
Personalized GPT: personalized chatbots galore
The most resourceful users know, using precise prompts, to get exactly what they want from ChatGPT, but this requires a good knowledge of general AI and a bit of tinkering. “Since the launch of ChatGPT, people have asked for ways to customize ChatGPT to fit their specific uses”explains OpenAI in a statement. Also, the start-up decided to deploy a tool, GPT Builder, allowing the creation of AI with a tailor-made personality. With this new service, this time it is targeting companies that would like, for example, to create automated customer service. Each GPT can have access to web browsing, the DALL-E image generator, and OpenAI’s Code Interpreter tool for writing and running software. To meet a specific need, users will be able to “feed” their GPTs with specific data, which chatbots might not find on the web. Another feature, called “Actions,” will allow chatbots to connect to external services to access data such as emails, calendars or databases and more. We can imagine the owner of a hotel creating a GPT to answer various questions from customers staying there, for example, by providing them with documents that will allow them to know everything about the establishment. In short, where ChatGPT is a general model, GPTs are a kind of sub-model specialized in a chosen domain. And the best part is that there’s no programming code required!
With these GPTs, anything is possible. It’s easy to imagine someone building a GPT specializing in learning a new language or giving advice on interior design. Or even a chatbot expert in writing, a private teacher for math and even a technical advisor to troubleshoot our devices. And no need to even create them, since it will be possible in the coming weeks to find and download these personalized chatbots – created by other users – in a GPT Store. We know that only bots from people with verified identities will be accepted. Additionally, OpenAI says the creators of GPTs won’t be able to see conversations people have with them – but it’s unclear what data they’ll have access to – and that it will monitor activity to block things like fraud, hate speech and adult content. Finally, the company promises to pay creators an unspecified amount based on the use of their GPTs. These new features will only be accessible to paying subscribers of ChatGPT Plus and to corporate clients of OpenAI, who will be able to create internal GPTs for their employees, initially.
GPT-4 Turbo: a faster and more powerful language model
With the introduction of custom GPTs, OpenAI directly competes with other AI bot platforms like Character.AI and Meta, with the company introducing virtual friends to WhatsApp, Instagram and Messenger. . However, the start-up focuses its platform on utility, while those cited above focus on chatbots that act like people – afterwards, it is entirely possible to build GPTs with “human” behavior if the heart tells us.
But these are not the only new features announced during DevDay. Indeed, OpenAI presented a faster and faster GPT-4 while being cheaper, called GPT-4 Turbo. First advantage: the language model is capable of processing more data since a query can now be 300 pages long (128,000 characters). This will be particularly useful for tasks requiring a deep understanding of certain documents or long swaths of code. Additionally, while the information GPT-4 has access to stops at September 2021, GPT-4 Turbo goes until April 2023. The language model is also faster and less expensive. Thus, developers will have to spend $0.01 for 1,000 tokens — the basic unit of code or text that the LLM must read — compared to $0.03 per 1,000 tokens with GPT-4.
Another problem OpenAI is tackling: copyright. Indeed, content generated by artificial intelligence does not come out of nowhere. The AI only “picks” small pieces of text, read from the Internet. And some of these contents are protected by copyright. However, the fear of lawsuits represents a major obstacle to the adoption of AI — OpenAI itself is the target of numerous complaints. The company is therefore launching its Copyright Shield program to cover legal costs in the event of prosecution for intellectual property infringement against companies and developers using its AI. It remains to be seen whether Europe will have access to all these new features at the same time as the rest of the world, given that the European Union has a fairly strict data protection regime, which has already delayed the arrival of several such projects, such as Google Bard and Microsoft Copilot.