Nearly a year ago, the world got a first glimpse of OpenAI’s ChatGPT. What has happened since then – let’s be honest – is beyond remarkable. A technology that very few of us knew existed is now being used day to day for pretty much every sort of task imaginable.
100,000,000 active weekly users. I genuinely had to double-check the number of zeros I wrote – that’s how large that number is. Not bad for a technology that is just about to celebrate its first birthday. The technology that this week got some pretty major enhancements. Let’s dive in to see exactly what they were…
The Model Is More Up to Date
My biggest pet peeve with ChatGPT has been the 2019 cut-off for information the model has been trained to. This has now been updated to April 2023, with OpenAI also promising to never let the model get that out of date again.
Browse the Web
This feature almost makes the above irrelevant. Google Bard has been able to do this for a while, as it is plugged into the Google Search Engine and can search the web based on your prompts. This feature is now available in ChatGPT courtesy of Bing.
In the above, you can see a prompt being entered about the latest Salesforce Winter ‘24 release. ChatGPT then scans the web via Bing and gets the information from any relevant sources.
Note: I didn’t do anything for the search to come up with salesforceben.com as the main source – it did that all by itself.
GPT-4 Turbo with 128k Context
For me, context has always been the main driver behind any GPT tool. Now, with GPT-4 Turbo, there is support for 128,000 characters in context. This means that the context in a single chat can go back over 128,000 characters or over 300 pages of text.
The new and improved GPT-4 Turbo is a big step up from the original. It’s trained on an even bigger dataset and has a fancy new design that makes it quicker and more accurate. This turbocharged version is great for things like coming up with creative writing, translating languages, and crafting all sorts of imaginative content. It’s like the souped-up sports car of AI language models!
GPTs
Just to confuse things even more, OpenAI has announced GPTs. These are not to be confused with ChatGPT the OpenAI product, or GPT (generative pre-trained transformer) the technology behind ChatGPT.
GPTs are a new type of AI agent that is trained on specific tasks. They’re essentially very detailed prompts that your users can make use of and can be created without any coding knowledge. For example, OpenAI has trained GPTs on tasks such as customer service, technical support, and writing creative content. GPTs can be used to build custom AI applications that are tailored to specific needs.
An example of this may be a GPT that can analyze the release notes for an upcoming Salesforce release and be able to summarize them around things that may affect your organization. This is something that previously wouldn’t be possible with ChatGPT, showing the massive leap forward they have made in such a short space of time.
OpenAI App Store
Just like many other technology companies, OpenAI is releasing an app store. This will allow developers to publish and sell GPT-powered applications, making it easier for businesses and consumers to use GPT technology to solve real-world problems.
This works similarly to AppExchange – you will be able to download GPTs to enhance your use of ChatGPT in the same way AppExchange apps enhance your use of Salesforce.
Custom Models
OpenAI is rolling out the Custom Models program for big organizations that want super personalized AI. If you’ve got a ton of your own data and need a special AI, they’ll let you work closely with their experts to train a customized version of GPT-4 just for your needs.
It’s like getting a personal AI assistant tailored specifically for your company. Plus, they promise to keep it super private – your custom AI won’t be shared with anyone else, and the data you give them won’t be used anywhere else, either.
If this is something you might be interested in, you can apply here.
Copyright Shield
Copyright Shield is a new service that will defend customers from legal action related to copyright infringement. This is hugely important because GPT technology can be used to generate text that is similar to existing copyrighted works. Copyright Shield will help to protect customers from being sued for copyright infringement even if their applications legitimately use GPT technology.
Summary
Overall, the OpenAI ChatGPT keynote showed that OpenAI is committed to making GPT technology more powerful, accessible, and reliable.
Understanding release notes will become a breeze with the ability to search the web, and the 128,000-character context will make understanding and writing code using ChatGPT much easier. These enhancements announced at the keynote have the potential to revolutionize the way we use ChatGPT in the Salesforce ecosystem. Again.