{"version":"desci-nodes-0.2.0","components":[{"id":"root","name":"root","type":"data-bucket","payload":{"cid":"bafybeigu45ioenu4mabnuwim2fu2uhfhm2k5swp2w6qd6wu5g7chkyni4a","path":"root"}},{"id":"d8e4ccec-df53-492f-97b3-45a58612ad75","name":"Preprint","type":"pdf","subtype":"research-article","payload":{"url":"bafybeiasioky23nkwsekte7kstmgt447jekgg6bbyipboeu44nemzrbrci","path":"root/social simulation.pdf","title":"Preprint of Training Socially Aligned Language Models in Simulated Human Society","ontologyPurl":"","cedarLink":"","controlledVocabTerms":[],"keywords":[],"description":"Social alignment in AI systems aims to ensure that these models behave according\nto established societal values. However, unlike humans, who derive consensus on\nvalue judgments through social interaction, current language models (LMs) are\ntrained to rigidly replicate their training corpus in isolation, leading to subpar generalization in unfamiliar scenarios and vulnerability to adversarial attacks. This\nwork presents a novel training paradigm that permits LMs to learn from simulated social interactions. In comparison to existing methodologies, our approach\nis considerably more scalable and efficient, demonstrating superior performance\nin alignment benchmarks and human evaluations. This paradigm shift in the training of LMs brings us a step closer to developing AI systems that can robustly and\naccurately reflect societal norms and values."},"starred":true},{"id":"8b041937-c46e-4ab6-a58c-47b0a3f40eac","name":"Code Repository","type":"code","payload":{"url":"bafybeifmdeo5woy4hewfcsz5cv4up7h7dx6n6tz7ftfzlvmsj5comoew2i","path":"root/Stable-Alignment","externalUrl":"https://github.com/agi-templar/Stable-Alignment","title":"Code Repo: Stable Alignment - Alignment Learning in Social Games","ontologyPurl":"","cedarLink":"","controlledVocabTerms":[],"keywords":[],"description":"This is the official repo for the Stable Alignment project. We aim to provide a RLHF alternative which is superior in alignment performance, highly-efficient in data learning, and easy to deploy in scaled-up settings. Instead of training an extra reward model that can be gamed during optimization, we directly train on the recorded interaction data in simulated social games. We find high-quality data + reliable algorithm is the secret recipe for stable alignment learning.\n\nThe repo contains:\n\nThe code for running social simulation in Sandbox.\nThe 169K interaction data used for alignment training.\nThe code for training with stable alignment.\nThe download for So(cially)-Good Language Model."},"starred":true}],"authors":[{"name":"Ruibo Liu","role":"Author","googleScholar":"https://scholar.google.com/citations?user=5lgfeo4AAAAJ&hl=en&oi=ao","orcid":""},{"name":"Ruixin Yang","role":"Author","googleScholar":"https://scholar.google.com/citations?hl=en&user=Q7Vn8OAAAAAJ","orcid":""},{"name":"Chenyan Jia","role":"Author","googleScholar":"https://scholar.google.com/citations?hl=en&user=S34REOAAAAAJ","orcid":"0000-0002-8407-9224"},{"name":"Ge Zhang","role":"Author","googleScholar":"","orcid":""},{"name":"Denny Zhou","role":"Author","googleScholar":"https://scholar.google.com/citations?hl=en&user=UwLsYw8AAAAJ","orcid":""},{"name":"Andrew M. Dai","role":"Author","googleScholar":"","orcid":"0000-0002-7112-8277"},{"name":"Diyi Yang","role":"Author","googleScholar":"https://scholar.google.com/citations?hl=en&user=j9jhYqQAAAAJ","orcid":""},{"name":"Soroush Vosoughi","role":"Author","googleScholar":"https://scholar.google.com/citations?hl=en&user=45DAXkwAAAAJ","orcid":"0000-0002-2564-8909"}],"title":"Simulated Society for Alignment","defaultLicense":"CC BY","researchFields":["Artificial Intelligence"],"dpid":{"prefix":"beta","id":"55"}}