We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: A General Language Assistant as a Laboratory for Alignment

Abstract: Given the broad capabilities of large language models, it should be possible to work towards a general-purpose, text-based assistant that is aligned with human values, meaning that it is helpful, honest, and harmless. As an initial foray in this direction we study simple baseline techniques and evaluations, such as prompting. We find that the benefits from modest interventions increase with model size, generalize to a variety of alignment evaluations, and do not compromise the performance of large models. Next we investigate scaling trends for several training objectives relevant to alignment, comparing imitation learning, binary discrimination, and ranked preference modeling. We find that ranked preference modeling performs much better than imitation learning, and often scales more favorably with model size. In contrast, binary discrimination typically performs and scales very similarly to imitation learning. Finally we study a `preference model pre-training' stage of training, with the goal of improving sample efficiency when finetuning on human preferences.
Comments: 26+19 pages; v2 typos fixed, refs added, figure scale / colors fixed; v3 correct very non-standard TruthfulQA formatting and metric, alignment implications slightly improved
Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG)
Cite as: arXiv:2112.00861 [cs.CL]
  (or arXiv:2112.00861v3 [cs.CL] for this version)

Submission history

From: Jared Kaplan [view email]
[v1] Wed, 1 Dec 2021 22:24:34 GMT (3014kb,D)
[v2] Tue, 7 Dec 2021 20:47:42 GMT (3014kb,D)
[v3] Thu, 9 Dec 2021 21:40:22 GMT (3031kb,D)

Link back to: arXiv, form interface, contact.