GPTs - make your own ChatGPT

Welcome to the 476 new subscribers who joined last week. I’m delighted that you’re here to learn with me.

‍

It’s been a wild, wild week in AI:

‍

Dario’s Picks – OpenAI DevDay special

OpenAI launched a whole bag of massive features during their DevDay yesterday.

• Anyone can now build GPTs – customised versions of ChatGPT – using natural language, and share them with the world.
• Users will be able to monetize their GPTs through revenue sharing and get listed in the upcoming GPT store.
• ChatGPT just got much better memory and more up to date knowledge.
• For developers, it just got way more attractive to build AI applications using OpenAI’s API.

Here’s the breakdown on the specific launches and updates:

‍

I. GPTs

A new feature for ChatGPT that makes it possible for anyone to create tailored versions of ChatGPT and share them with the world. Rolls out to Plus and Enterprise subscribers this week. The most amazing thing is this: you can create GPTs using natural language – you don’t need to know how to code even a single line.

‍

‍

Enabling DALL-E 3, internet browsing and data analytics (Code Interpreter) inside your GPT is just a matter of checking a box.

You can tailor a GPT in three different ways:

• Instructions: You give the GPT instructions about how it should respond when someone prompts it, in a similar was as Custom Instructions works.
• Expanded knowledge: Add any additional data you want the GPT to have. It’s easy to upload, for example a PDF or an excel file to enhance the GPTs knowledge.
• Actions: The option to connect to third-party APIs, so your GPT can perform actions in the “real world” (such as sending an email or access an external database).

You can make GPTs private, for your company only or – most excitingly – public for others to use.

‍

‍

II. The GPT Store

OpenAI announced the launch of the GPT store at the end of this month, which could quickly become the marketplace for AI innovation. From the glimpse we got from DevDay, it’s similar to the App Store – just for ChatGPT.

The coolest thing about the store is that builders of GPTs will be able to monetize their creations on a revenue sharing basis. While the exact details on the monetization model aren’t laid out yet, OpenAI CEO Sam Altman said that creators will be paid “a portion of our revenue”.

‍

III. GPT-4 Turbo

OpenAI launched GPT-4 Turbo, the underlying model to ChatGPT and the new model that developers will use to build AI apps on.

These are the improvements:

• Context length: Supports up to 128k tokens of context. The largest context window of a large language model so far, including more than Anthropic’s Claude.
• Better knowledge: Updated cut-off date to April 2023.
• More control: Improvements that gives developers more control when using the new Assistants API (JSON-mode, better function calling, seed parameter option).
• Retrieval: Inject your own documents and databases into what you are building.
• New modalities: You can now develop using DALL-E 3, GPT-4 Vision and TTS via API.
• Higher rate limits: 2x tokens per minute
• Pricing: 3x cheaper for prompt tokes and 2x cheaper for completion tokens.

‍

In Focus

‍

Creator monetization for AI platforms is all the rage right now

ChatGPT is undoubtedly having an App Store moment.

As seen with the example of GPTs, the name of the game for AI chatbot platforms right now is enabling and scaling creator monetization. We’ve seen this happen with social media platforms, where they basically fight over who can give creators the best buck for their content.

To pay GPT creators makes a lot of sense when you think about it: earning potential incentivises creators, which makes the platform more useful with their innovations, which in turn attracts more users and revenue.

Here’s some cool things monetization for creators brings to the table:

• As end-users, we get choice and a lot more cool products because developers now have an incentive to create good apps on the platform. The ecosystem also ensures we don’t pay too much as there is competition from other GPTs.
• For regular people, it lowers the bar on who can create apps with the platform, giving anyone a chance to earn.
• For app developers with limited budgets, it solves the problem of cost and distribution.
• For established companies, it provides a cheap and efficient way to test demand for new concepts (not only usage-wise but also if people are actually willing to pay for it), before developing a full-scale application.
• For the company behind the platform, revenue sharing is both a clever way to onboard new subscribers and secure development of useful apps on the platform – kind of like an affiliate program.

On a similar note, just last week Quora’s platform Poe launched creator monetisation of their “Bots” (they are like simple versions of GPTs). Creators get paid when users sign up to Poe’s paid plan after using their bot, and they can also set a limit on the number of messages users can send for free to their bot.

‍

Three expectation I have:

1) creator monetization will be a huge deal, and the rest of AI chatbot platforms to follow suit. After all, what would the iPhone be without the apps?

2) In the short-term, creative but relatively basic, instruction-based GPTs will be able to generate good income for creators.

3) In the medium term, Actions (the new feature that allows GPTs to connect to third-party APIs) is likely to be the differentiating component of GPTs that continue to earn money, as they are more difficult to copy.

4) In the long term, OpenAI will itself integrate the features of the best performing GPTs directly into the platform.

A lot of people are emphasising that a thousand startups die each time OpenAI launches an update. If you’re not OpenAI or Google, it’s rather difficult to build a technical moat in this day and age. So with that in mind, creators should probably not fear but rather expect their concept to be copied, by other creators and by the leading AI companies themselves. It’s probably smart to think holistically about the use case you are solving and avoid over-reliance on any one particular feature.

‍

ChatGPT GPTs

This section of the newsletter, which was previously about ChatGPT plugins, will from now on be dedicated to GPTs.

ChatGPT plugins are officially being phased out, and replaced by GPTs, although it will remain in ChatGPT for the time being.

Actions is the new term for the component of GPTs that allows them to connect to third-party apps. Existing plugins for ChatGPT can easily be converted to a GPT with an action.

GPTs and their ecosystem will have several benefits over that of plugins:

• Lowers the barrier for builders – no code needed to make one.
• Easy to enable all built-in capabilities of ChatGPT: web browsing, image generation and code interpreter.
• Better distribution with the upcoming GPT marketplace (which will supersede the plugin store).
• For developers aiming to create Actions for the GPT, there is enhanced safety and user control, through consequential flagging; also, there is flexibility to have only certain parts of your GPT require authentication.

I’m so excited to help you stay on top of all-things GPTs in this newsletter. Stay tuned for notable launches, lists and guides.

‍

Bytes

• Elon Musk’s new company xAI announces Grok, a competitor to ChatGPT(!)
• Create custom versions of ChatGPT with GPTs and Zapier.
• New Beatles song released with the power of AI. Listen to it on Spotify.
• Instagram to introduce a new AI friend feature.
• API access to GPT-4V’s (voice & image capabilities) is coming “soon” according to Logan.GPT.
• U.K.’s AI Safety Summit ends with limited, but meaningful, progress.
• Brave launches their new chatbot Leo, focused on privacy and security.
• Phind: this new model reportedly beat GPT-4 at coding.

‍

That’s a wrap for this week!

Fellow wizards – join me on Twitter @itspapilama.

Until next time,

Dario Chincha 🧙🏼‍♂️