Text Alignment Is An Efficient Unified Model for Massive NLP Tasks. (arXiv:2307.02729v2 [cs.CL] UPDATED)

Large language models (LLMs), typically designed as a function of next-word
prediction, have excelled across extensive NLP tasks. Despite the generality,
next-word prediction is often not an efficient formulation for many of the
tasks, demanding an extreme scale of model parameters (10s or 100s of billions)
and sometimes yielding suboptimal performance. In practice, it is often
desirable to build more efficient models -- despite being less versatile, they
still apply to a substantial subset of problems, delivering on par or …

alignment arxiv function language language models large language large language models llms massive next nlp parameters prediction scale tasks text unified model word