More Like this

1. Learning from Natural Language Feedback

   Chen, A; Scheurer, J; Campos, JA; Korbak, T; Chan, JS; Bowman, SR; Cho, K; Perez, E (February 2024, Transactions on machine learning research)

   The potential for pre-trained large language models (LLMs) to use natural language feedback at inference time has been an exciting recent development. We build upon this observation by formalizing an algorithm for learning from natural language feedback at training time instead, which we call Imitation learning from Language Feedback (ILF). ILF requires only a small amount of human-written feedback during training and does not require the same feedback at test time, making it both user-friendly and sample-efficient. We further show that ILF can be seen as a form of minimizing the KL divergence to the target distribution and demonstrate proof-of-concepts on text summarization and program synthesis tasks. For code generation, ILF improves a Codegen-Mono 6.1B model’s pass@1 rate from 22% to 36% on the MBPP benchmark, outperforming both fine-tuning on MBPP and on human- written repaired programs. For summarization, we show that ILF can be combined with learning from human preferences to improve a GPT-3 model’s summarization performance to be comparable to human quality, outperforming fine-tuning on human-written summaries. Overall, our results suggest that ILF is both more effective and sample-efficient than training exclusively on demonstrations for improving an LLM’s performance on a variety of tasks. 

   more » « less
2. Pretraining Language Models with Human Preferences

   Korbak, T; Shi, K; Chen, A; Bhalerao, R; Buckley, CL; Phang, J; Bowman, SR; Perez, E (July 2023, Machine Learning Research Press)

   Language models (LMs) are pretrained to imitate internet text, including content that would violate human preferences if generated by an LM: falsehoods, offensive comments, personally identifiable information, low-quality or buggy code, and more. Here, we explore alternative objectives for pretraining LMs in a way that also guides them to generate text aligned with human preferences. We benchmark five objectives for pretraining with human feedback across three tasks and study how they affect the trade-off between alignment and capabilities of pretrained LMs. We find a Pareto optimal and simple approach among those we explored: conditional training, or learning distribution over tokens conditional on their human preference scores given by a reward model. Conditional training reduces the rate of undesirable content by up to an order of magnitude, both when generating without a prompt and with an adversarially chosen prompt. Moreover, conditional training maintains the downstream task performance of standard LM pretraining, both before and after task-specific finetuning. Pretraining with human feedback results in much better preference satisfaction than standard LM pretraining followed by finetuning with feedback, i.e., learning and then unlearning undesirable behavior. Our results suggest that we should move beyond imitation learning when pretraining LMs and incorporate human preferences from the start of training. 

   more » « less
3. Pretraining Language Models with Human Preferences

   Korbak, Tomasz; Shi, Kejian; Chen, Angelica; Bhalerao, Rasika; Buckley, Christopher L.; Phang, Jason; Bowman, Samuel R.; Perez, Ethan (July 2023, International Conference on Machine Learning)

   Language models (LMs) are pretrained to imitate text from large and diverse datasets that contain content that would violate human preferences if generated by an LM: falsehoods, offensive comments, personally identifiable information, low-quality or buggy code, among others. Here, we explore alternative objectives for pretraining LMs in a way that also guides them to generate text aligned with human preferences. We benchmark five objectives for pretraining with human feedback across three tasks and study how they affect the alignment and capabilities of pretrained LMs. We find a Pareto-optimal and simple approach among those we explored: conditional training, or learning distribution over tokens conditional on their human preference scores. Conditional training reduces the rate of undesirable content by up to an order of magnitude, both when generating without a prompt and with an adversarially-chosen prompt. Moreover, conditional training maintains the downstream task performance of standard LM pretraining, both before and after task-specific finetuning. Pretraining with human feedback results in much better preference satisfaction than standard LM pretraining followed by finetuning with feedback, i.e., learning and then unlearning undesirable behavior. Our results suggest that we should move beyond imitation learning when pretraining LMs and incorporate human preferences from the start of training. 

   more » « less
4. Influencing Reinforcement Learning through Natural Language Guidance

   https://doi.org/10.32473/flairs.v34i1.128472  

   Tasrin, Tasmia; Nahian, Md Sultan; Perera, Habarakadage; Harrison, Brent (May 2021, The International FLAIRS Conference Proceedings)

   null (Ed.)

   Interactive reinforcement learning (IRL) agents use human feedback or instruction to help them learn in complex environments. Often, this feedback comes in the form of a discrete signal that’s either positive or negative. While informative, this information can be difficult to generalize on its own. In this work, we explore how natural language advice can be used to provide a richer feedback signal to a reinforcement learning agent by extending policy shaping, a well-known IRL technique. Usually policy shaping employs a human feedback policy to help an agent to learn more about how to achieve its goal. In our case, we replace this human feedback policy with policy generated based on natural language advice. We aim to inspect if the generated natural language reasoning provides support to a deep RL agent to decide its actions successfully in any given environment. So, we design our model with three networks: first one is the experience driven, next is the advice generator and third one is the advice driven. While the experience driven RL agent chooses its actions being influenced by the environmental reward, the advice driven neural network with generated feedback by the advice generator for any new state selects its actions to assist the RL agent to better policy shaping. 

   more » « less
5. The Wisdom of Hindsight Makes Language Models Better Instruction Followers

   Zhang, Tianjun; Liu, Fangchen; Wong, Justin; Abbeel, Pieter; Gonzalez, Joseph E (July 2023, PMLR)

   Reinforcement learning has seen wide success in finetuning large language models to better align with instructions via human feedback. The so-called algorithm, Reinforcement Learning with Human Feedback (RLHF) demonstrates impressive performance on the GPT series models. However, the underlying reinforcement learning algorithm is complex and requires additional training for reward and value networks. In this paper, we consider an alternative approach: converting feedback to instruction by relabeling the original one and training the model for better alignment in a supervised manner. Such an algorithm doesn’t require any additional parameters except for the original language model and maximally reuses the pretraining pipeline. To achieve this, we formulate instruction alignment problem for language models as a goal-reaching problem in decision making. We propose Hindsight Instruction Relabeling (HIR), a novel algorithm for aligning language models with instructions. The resulting two-stage algorithm shed light to a family of reward-free approaches that utilize the hindsightly relabeled instructions based on feedback. We evaluate the performance of HIR extensively on 12 challenging BigBench reasoning tasks and show that HIR outperforms the baseline algorithms and is comparable to or even surpasses supervised fine-tuning. The implementation of HIR is available at https://github.com/tianjunz/HIR. 

   more » « less