![Frederik Wouters](/sites/default/files/styles/user_profile_thumbnail/public/2024-05/Frederik-Wouters-2-cropped%20noteeth%20zoomed-small.jpg?h=57739094&itok=jG5dswQF)
Frederik Wouters
This weekend Ghent was the place to be.
If you like food: Gent Smaakt is a culinary event and it was great to stroll around in the city streets in the evenings. During the day it was DrupalCamp Ghent.
🍊 At the booth, you could find 100% fresh pressed oranges, limes and grapefruits, no Artificial Ingredients (AI).
🔓 For me the best session was Merlijn Sebrechts' "AI is Breaking Open Source." He talked about open source and its current challenges with the rise of AI (like LLama3). His explanations of GenAI pipelines and OSA definitions were critical for understanding the broader impacts on the open source community.
He told us to share: LLAMA 3 IS NOT OPEN SOURCE according to Open Source Initiative (OSI)!
According to the fineprint,
LLAMA 3 can not be used by sex workers
The indian government is not alllowed to use LLAMA 3 (700 mio users limit)
You can not train other models if you’re using LLAMA 3 (in other words it’ only for META).
And Open Source should be about democratizing technology, making it available to ALL! He also explained when an AI is really Open Source (preferred form).
Merlijn also explained the GenAI pipeline and the important deliverables.
Main training
Data
Training code
Model code
Finetuning
Finetuning data
Finetuning code
Prompting
Prompt tuning
He also explained there is a OSA definition (V0.0.8). that is best explained here: https://opensource.org/deepdive/drafts/the-open-source-ai-definition-draft-v-0-0-8
In short:
you can use it for any purpose
You’re able to study it’s components
You’re able to modify it’s components
You can share it with or without modifications
You have access to it’s preferred form.
He also has some great ideas on how a trust based system could help a lot, and shortly mentioned the AI bill of materials. This is a good use of your time if you’re interested in Open source licenses and AI.
https://www.drupalcamp.be/en/drupalcamp-ghent-2024/session/ai-breaking-open-source
🍪 In "We Hate Cookies," Dr Philippe De Ryck gave us a lot of interesting technical information:
If you want to test your application, test it in Safari or Brave, they already do this. I strongly recommend all web developers out there to check the session!
https://www.drupalcamp.be/en/drupalcamp-ghent-2024/session/we-hate-cookies
🔎 I presented about Retrieval Augmented Generation “Do you still have a FAQ? You're doing it wrong.”.
I explained the Feedback Loop of FAQ and how to get the FAQ out of it.
In other words: how to remove some inefficiencies in your company (and how FAQ is the symptom). This is only possible thanks to new technology and changing expectations.
I have a more detailed post about how RAG works here: https://www.calibrate.be/nl/rag
but if you want a in depth explanation, ping me.
All in all Drupalcamp was again a big success. Thanks to Calibrate (platinum sponsor) and the organisation Drupal Belgium.