Hey 👋, thanks for stopping by!
In short, Uminal allows anyone to extend LLMs like GPT-3 with custom skills that can auto-compose w/ each other.
https://www.loom.com/share/211327fd8e854513b909b0f69eadd2f8
Large language models (LLMs) like GPT-3 are amazing, because they can reason and serve as zero/few-shot learners! And as LLMs scale up (cough cough GPT-4 👀), their reasoning & learning capabilities will only get better.
However, one issue with LLMs is that they’re not connected to the real world. After all, they’re just a text-in, text-out API trained on large amounts of Internet data that can quickly become stale.
Unlike humans, they can’t actually interface with other things, such as search engines, custom apps, etc.
Simply put, they’re unaware of our world in real-time.
But what if we could plug LLMs into the real world via custom skills/apps that anyone can build?
What if we could enable LLMs to automatically compose these apps together like Lego blocks, enabling them to interact w/ our world & perform tasks in new, novel ways that we have yet to imagine?
This is that Uminal is trying to achieve.