Humans In the Loop: Collaborating in the Age of AI
summary
IBM restructures collaboration into InnerSource then open source, taking internal NLP assets and infrastructure public (models on Hugging Face under Apache 2.0) and building a small, efficient family of models Granite v4 that runs even on a Raspberry Pi and is trained on 95% open data. Core document stack becomes Dockling 43k stars on GitHub. The company also donates orchestration work to the Linux Foundation; these components are used internally and externally, and the open approach demonstrably increases reuse and adoption.
AI-written summary · about
chapters (20)
- 0:00Introduction to talk
- 1:51Speaker’s background
- 3:15Early AI at IBM
- 4:41Hidden technical debt in ML
- 6:05Inner source AI journey begins
- 7:28Transformer models and LLMs
- 8:33AI as normal technology
- 10:15Watson’s early challenges
- 12:23Watson NLP success factors
- 14:52Generative AI era changes
- 16:39Cultural transformation at IBM
- 18:45Granite model family
- 20:53Why open source Granite
- 21:57Docklink and Langflow
- 23:20Context Forge and BAI platform
- 24:24Overcoming collaboration objections
- 26:10Survivors of AI projects
- 27:14AI’s role in code collaboration
- 29:00AI limitations and human ingenuity
- 30:45Conclusion and Q&A
related talks
41mHow InnerSource Is Beginning to Change Software Development at IBM. Lessons Learned...
16mKeynote: Open Source for AI and AI for Open Source
13mKeynote : The Open Source AI Alliance
33mKeynote: Fine-tuning our way towards openness in AI
58mAI in Reality Fireside Chat: Enterprise AI & Open‑Source Innovation
15mArtificial Intelligence Needs Community Intelligence
24mOpen Source AI and Granite
10mKeynote: Embracing Open Source to Beat the Great Reshuffle