Ok this one is a bit different, come here.

The GregoryMS project received a fantastic spotlight by the Público newspaper, “Bruno e os amigos criaram um robô que diz o que se está a fazer sobre esclerose múltipla”. Translation:1

The article is in Portuguese and behind a paywall. I will share it again once it falls under public access.

In short, this is a huge milestone for us. Me, António Lopes, Margarida Gomes, and Inês Carvalho, who recently volunteered to join the team.

The article focus how GregoryMS began and how we are still pushing towards its adoption by the medical community. At the same time, we have been turning it into something that can be used for any sort of research.

We’ve had media coverage before, but this is the first nation-wide media to help us spread the word.

Margarida Gomes, Bruno Amaral, António Lopes Photo by Maria Abranches

Artificial Intelligence and hallucinations

Back to our usual talk about AI.

There are two things to consider about the output of GPT-4 and other Large Language Models (LLMs).

  1. These generative models work through probability. Given a prompt and context, they present the most likely answer based on what the corpus used for their trainning. Therefore:
    1. Consider their ouput as the average answer.
    2. ChatGPT’s data was too generic and insanely large. Train a LLM with precise and curated information and the result will be immensily more valuable. (This is why I built GregoryMS)
  2. We are faced with the problem of validating the knowledge output of Artificial Intelligence.
    1. This is something that we are also working in at the Lisbon Collective, and the only path so far is to have a human double check the results.

Two links brought about the reflections above. One was “When A.I. Chatbots Hallucinate”, by the New York Times, and the other, a conversation on Mastodon by Jeff Jarvis and Simon Willison

https://mastodon.social/@[email protected]/110294077556692625

Your tips and tricks for AI

The important thing about chatGTP is that the result is only as good as our understanding of its limitations and our ability to ask questions.

You may frown at this, but I found an amazing gem on TikTok.


Working with 3D

https://3dfy.ai/ Give it a prompt and it generates a 3D render for you.

https://www.lumiere3d.ai/ Need a video ? Lumiere 3D is here for you.

https://captures.lumalabs.ai/imagine (alpha) An early experiment to prototype and create 3D with text.

Unsplash Logoigor savelev

  1. “Bruno and his friends built a robot that tells you all that is going on about Multiple Sclerosis.” ↩︎