An iPhone app called RollTab has been released that uses an AI model to automatically generate continuations of piano performances. Developed by Simon Edwardson, the app utilizes a custom-trained 125 million parameter Transformer model that processes MIDI files as discrete event sequences. This approach allows for rapid generation of musical continuations, with the app capable of producing output in under two seconds. Edwardson leveraged Gemini 3.5 Flash for preference data collection and Direct Preference Optimization (DPO) for further model refinement, ultimately exporting the model to Apple's Core ML for on-device operation. AI
IMPACT This app demonstrates a novel application of Transformer models for creative music generation on mobile devices, potentially inspiring similar tools.
RANK_REASON The article describes a specific application (RollTab) that uses an AI model for a particular task (piano music generation), rather than a core AI model release or significant industry-wide event.
Read on Mastodon — mastodon.social →
- Apple Inc.
- App Store
- Direct Preference Optimization: Your Language Model is Secretly a Reward Model
- Gemini 3.5 Flash
- GPT-2
- iPad
- iPhone
- MIDI
- RollTab
- Simon Edwardson
- Transformer++
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
