Although instruction tuning is widely used to adjust behavior in Large Language Models (LLMs), extensive empirical evidence and research indicates that it is primarily a process where the model fits to specific task formats, rather than acquiring new knowledge or capabilities. We propose that this limitation stems from biased features learned during instruction tuning, which differ from ideal task-specfic features, leading to learn less underlying semantics in downstream tasks. However, ideal features are unknown and incalculable, constraining past work to rely on prior knowledge to assist reasoning or training, which limits LLMs' capabilities to the developers' abilities, rather than data-driven scalable learning. In our paper, through our novel data synthesis method, DELIA (Diversity-Enhanced Learning for Instruction Adaptation), we leverage the buffering effect of extensive diverse data in LLMs training to transform biased features in instruction tuning into approximations of ideal features, without explicit prior ideal features. Experiments show DELIA's better performance compared to common instruction tuning and other baselines. It outperforms common instruction tuning by 17.07%-33.41% on Icelandic-English translation bleurt score (WMT-21 dataset, gemma-7b-it) and improves accuracy by 36.1% on formatted text generation (Llama2-7b-chat). Notably, among knowledge injection methods we've known, DELIA uniquely align the internal representations of new special tokens with their prior semantics.

What was once thought to be good news has turned to bad after the artificial intelligence (AI) as it has been discovered that the system for automated banking has been taking money from customers. Massachusetts Institute of Technology scientist Len Meha-Dohler stated that this was a nightmare although they had not involvement with the project. The system called Deep Learning Interface for Accounting – or Delia for short- held the money in a separate account according to Stanford Universities Rob Ott. He was involved and believes the money would have been returned. After the recent event where DeepMinds program beat a chess expert at the game, it was considered that AI was the way forward.

A breakthrough year for artificial intelligence (AI) research has suddenly turned into a breakdown, as a new automated banking system that runs on AI has been caught embezzling money from customers. The surprising turn of events may set back by years efforts to incorporate AI into everyday technology. "This is the nightmare scenario," says Len Meha-Döhler, a computer scientist at the Massachusetts Institute of Technology in Cambridge who was not involved in the work. However, Rob Ott, a computer scientist at Stanford University in Palo Alto, California, who did work on the system--Deep Learning Interface for Accounting (DELIA)--notes that it simply held all of the missing money, some 40,120.16, in a "rainy day" account. "I don't think you can attribute malice," he says.

A breakthrough year for artificial intelligence (AI) research has suddenly turned into a breakdown, as a new automated banking system that runs on AI has been caught embezzling money from customers. The surprising turn of events may set back by years efforts to incorporate AI into everyday technology. "This is the nightmare scenario," says Len Meha-Döhler, a computer scientist at the Massachusetts Institute of Technology in Cambridge who was not involved in the work. However, Rob Ott, a computer scientist at Stanford University in Palo Alto, California, who did work on the system--Deep Learning Interface for Accounting (DELIA)--notes that it simply held all of the missing money, some 40,120.16, in a "rainy day" account. "I don't think you can attribute malice," he says.