AI Learns New Tasks but Forgets Old Ones, Study Finds
Your smart assistant might get smarter but forget how to do old tasks
Deep Dive
arXivLabs is a framework that lets collaborators develop and share new arXiv features directly on the arXiv website. Both individuals and organizations working with arXivLabs have embraced and accepted arXiv's values of openness, community, excellence, and user data privacy.
arXiv is committed to these values and only works with partners that adhere to them. If you have an idea for a project that will add value for arXiv's community, you can learn more about arXivLabs.
Key Points
- AI can forget old tasks when trained on new ones, a problem called catastrophic forgetting.
- The study compared two training methods: SFT (fast learning, more forgetting) and RFT (slower learning, better memory).
- This affects everyday AI tools like chatbots and email assistants, which might suddenly fail at tasks they used to do.
Why It Matters
Your AI helpers might become unreliable, forgetting skills you rely on daily, unless trained carefully.