It was quite a week.
New Gregory AI project
The Lisbon Collective implemented Gregory-AI to find and filter relevant medical research for autoimmune encephalitis.
Gregory-AI is the software I use to find the most recent research on Multiple Sclerosis.
This project resulted from our partnership with Active Media, a Portuguese web development studio.
Artificial Intelligence in Higher Education (AI-HED)
We all need help adjusting to the new tools. Through the university where I teach, I am working on the AI-HED Project where we will develop guidelines, starter kits, and other resources for Higher Education Institutions.
Register here to stay up to date on AI-HED’s progress.
AI and our crazy world
“The world is weird, man.”
I never expected that AI, Politics, and Social Media would ever be so intertwined as to have Tech Billionaires join the government, and everything else that is going on.
These people are spending as much as they can on AI training and infrastructure. It starts with #Open-AI and The Stargate Project.
And Sam Altman, Open-AI’s CEO, steps forward.
Whichever the case, all sides are using investment numbers to signal market strength. Their promise is for more and better AI models, and better quality training data (which I am sure they are scrapping with disregard for copyright or ethics).
But are they fulfilling the promise of reputation signalling? We did see new LLM models published by Meta, X, and Google, and several other smaller ones are coming out almost weekly. The discussion i come across already changed from “new AI model came out” to “which is better for reasoning? Which is better for code?”
DeepSeek enters the chat
The Chinese AI Startup made the news this week with its new AI model that cost “only” $5.6 million to train. The Decoder has the details.
The DeepSeek team also published their research, thus contributing to the open-source community working in AI, and earning praise from their peers. Some of which work in competitor companies.
Arxiv: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
And all this shook the AI engineering teams who have been riding on the investment hype. The screenshot below is making the rounds. The post comes from Team Blind, a forum for verified Big Tech employees.
Appearances are very easy to build, today they are content without substance. I believe that this year we need to focus a lot more on finding substance and filtering out the noise from Social Media, Politics, and AI.
Starting your AI podcast
Podcastify is an open-source software with the same goal as Google’s LLM notebook, upload your sources and the AI turns it into a summary or a podcast episode. Podcastify is more flexible and extensible because it is open-source.
If you have an API key for OpenAI, Google Gemini, or ElevenLabs, you can use it for free here: https://openpod.fly.dev/
Music all the time
I had to go through my music collection for dumb reasons. It brought back the memory of the “Jam Session 2.0” by Dumbfoundead. It’s a collaboration published in 2009, where 8 people with 5 instruments originally from 4 Continents speaking 3 languages got together for 1 song.
https://www.youtube.com/watch?v=1oU0I8APK-o
For me, this is what the web should be about, bringing people together and having them build something, solve a problem, share a solution.
Instead, we let ourselves get taken by the polarising algorithms that only benefit the new robber barons. Divide and conquer turned into divide and control.
