Reflections on OpenAI’s DevDay Announcements

OpenAI announced a lot of things, and wiped out a lot of startups, at their DevDay including…

New GPT-4 Turbo

Updated GPT-3.5 Turbo Assistants

API Multimodal capabilities

Customizable GPTs and marketplace in ChatGPT

It’s still early days but the apparent accelerated pace of innovation at OpenAI shows that Microsoft have clearly learnt from their Skype experience and Microsoft money might not be the “kiss of death” for OpenAI…

New GPT-4 Turbo – the 128K context window and knowledge of world events up to April 2023 are both massive improvements. 128k is still a bit like the 160 character limit for SMS messages but it’s a step in the right direction. Updating the model from November 2021 to April 2023 is much needed. 2021 is a long time ago in terms of world geo politics and AI in particular…

Fine tuning 3.5 Turbo is positive and if I was OpenAI “Fine tuning” would be my priority #1…being able to put information in the user’s context is of massive benefit…OpenAI is a bit better than Google at delivering insight to customers but the value of that insight is multiplied by an order of magnitude if it is put in context for the user…

The beta of the Assistants API (allowing developers to create AI assistants that can follow specific instructions, interact with models and tools, and have persistent threads for context management) is a great move. Everyone focuses on the Chatbot use case for some reason so I can understand that. LLMs are much better at document creation than chat on the whole but chatting to an LLM definitely has a lot more “wow factor” I guess.

GPT-4 Turbo supporting visual inputs in the Chat Completions API, and DALL·E 3 making image generation is available via the Image generation API are both great steps forward. Text-to-speech capabilities have also been added with the TTS model which throws up some interesting potential use cases…Hal 9000 from 2001: A Space Odyssey takes another step closer…

Customizable GPTs and a marketplace in ChatGPT enabling developers to customize (and potentially monetise) a GPT with instructions, data, and even developer-defined actions is very cool…I can’t wait to start playing with that.

Overall I think it was a very strong set of deliverables for OpenAI’s first DevDay…hopefully they find this pace of innovation sustainable.

A lot of startups were wiped out with these announcements as well of course…you’d have to have a lot of VC money and be very brave to launch a startup augmenting ChatGPT’s functionality, or plugging a perceived ChatGPT feature gap, going forwards given the pace of evolution that they have just demonstrated.

As @thealexker put it “The only moats are execution, distribution, and top-of-mindedness. Your brand needs to be ingrained in a customer’s habit loop.”

Judging OpenAI on this performance it is looks like that they are managing to scale without overtly slowing down which isn’t easy and points to a healthy internal culture and effective operating model…great stuff.

Rorie is the author of “The CTO | CIO Bible” and has been working with Neural Networks since his final year dissertation on “The use of Neural Networks in financial time series prediction” at Uni. Commercial AI work way before anyone had ever heard of ChatGPT included a secret squirrel project using a neural network to reverse engineer the Google search algorithm…he thought it was a massive neural network at the time but ChatGPT 4 Turbo is a lot, lot bigger…now part of the founding team at the AI as a Service Team providing AI Software Design & Delivery | AI APIs & Tools | AI Consulting Services…he can be reached at AIaaS.Team

you're currently offline