AtlasIA, Building AI with Moroccan heritage
summary
AtlasIA builds Moroccan AI assets: they collect street-recorded Darija utterances, create parallel translation and transcription datasets (audio, Latin and Arabic script), and publish models and evaluations. The group reports a Darija translation benchmark released last month and runs model comparisons by voting on Discord, aiming to measure which open models handle locality best. Community size is about 250-person community, work is open-source, and contributors can join to add data, run evaluations, or help fine-tune models.
AI-written summary · about
chapters (7)
- 0:00Introduction and thanks
- 1:06AtlasIA and Moroccan AI community
- 3:20Data collection and Discord collaboration
- 5:59Transcription and model training
- 7:26Model evaluation and open-source contribution
- 9:16Dialect integration and community impact
- 10:46How to join and contribute
related talks
28mAI’s Next Great Challenge: How to create AI that truly understands us?
47mAya: Expanding the Worlds AI serves
36mAI & Data Science: Driving Innovation or Guarding Digital Borders?
24mKeynote: Building India’s Sovereign AI Stack w/ Local Talent & Open Global Collab
17mAI-Powered Research: Accelerating Public Sector Transformation in the MENA Region
21mDemocratizing AI with Open Source Multi-Agent Systems
33mAgentic AI – What's Next?
44mKeynote: The Art of Language: Mastering Multilingual Challenges in LLMs