The o1 model series is trained with large-scale reinforcement learning to reason using chain of thought. These advanced reasoning capabilities provide new avenues for improving the safety and robustness of our models. In particular, our models can reason about our safety policies in context when responding to potentially unsafe prompts, through deliberative alignment. This leads to state-of-the-art performance on certain benchmarks for risks such as generating illicit advice, choosing stereotyped responses, and succumbing to known jailbreaks. Training models to incorporate a chain of thought before answering has the potential to unlock substantial benefits, while also increasing potential risks that stem from heightened intelligence. Our results underscore the need for building robust alignment methods, extensively stress-testing their efficacy, and maintaining meticulous risk management protocols. This report outlines the safety work carried out for the OpenAI o1 and OpenAI o1-mini models, including safety evaluations, external red teaming, and Preparedness Framework evaluations.

GPT-4o is an autoregressive omni model that accepts as input any combination of text, audio, image, and video, and generates any combination of text, audio, and image outputs. It's trained end-to-end across text, vision, and audio, meaning all inputs and outputs are processed by the same neural network. GPT-4o can respond to audio inputs in as little as 232 milliseconds, with an average of 320 milliseconds, which is similar to human response time in conversation. It matches GPT-4 Turbo performance on text in English and code, with significant improvement on text in non-English languages, while also being much faster and 50\% cheaper in the API. GPT-4o is especially better at vision and audio understanding compared to existing models. In line with our commitment to building AI safely and consistent with our voluntary commitments to the White House, we are sharing the GPT-4o System Card, which includes our Preparedness Framework evaluations. In this System Card, we provide a detailed look at GPT-4o's capabilities, limitations, and safety evaluations across multiple categories, focusing on speech-to-speech while also evaluating text and image capabilities, and measures we've implemented to ensure the model is safe and aligned. We also include third-party assessments on dangerous capabilities, as well as discussion of potential societal impacts of GPT-4o's text and vision capabilities.

Diffusion models are a class of flexible generative models trained with an approximation to the log-likelihood objective. However, most use cases of diffusion models are not concerned with likelihoods, but instead with downstream objectives such as human-perceived image quality or drug effectiveness. In this paper, we investigate reinforcement learning methods for directly optimizing diffusion models for such objectives. We describe how posing denoising as a multi-step decisionmaking problem enables a class of policy gradient algorithms, which we refer to as denoising diffusion policy optimization (DDPO), that are more effective than alternative reward-weighted likelihood approaches. Empirically, DDPO can adapt text-to-image diffusion models to objectives that are difficult to express via prompting, such as image compressibility, and those derived from human feedback, such as aesthetic quality. Finally, we show that DDPO can improve prompt-image alignment using feedback from a vision-language model without the need for additional data collection or human annotation. The project's website can be found at Diffusion probabilistic models (Sohl-Dickstein et al., 2015) have recently emerged as the de facto standard for generative modeling in continuous domains. The key idea behind diffusion models is to iteratively transform a simple prior distribution into a target distribution by applying a sequential denoising process. This procedure is conventionally motivated as a maximum likelihood estimation problem, with the objective derived as a variational lower bound on the log-likelihood of the training data. However, most use cases of diffusion models are not directly concerned with likelihoods, but instead with downstream objective such as human-perceived image quality or drug effectiveness.

Sample efficiency and exploration remain major challenges in online reinforcement learning (RL). A powerful approach that can be applied to address these issues is the inclusion of offline data, such as prior trajectories from a human expert or a sub-optimal exploration policy. Previous methods have relied on extensive modifications and additional complexity to ensure the effective use of this data. Instead, we ask: can we simply apply existing off-policy methods to leverage offline data when learning online? In this work, we demonstrate that the answer is yes; however, a set of minimal but important changes to existing off-policy RL algorithms are required to achieve reliable performance. We extensively ablate these design choices, demonstrating the key factors that most affect performance, and arrive at a set of recommendations that practitioners can readily apply, whether their data comprise a small number of expert demonstrations or large volumes of sub-optimal trajectories. We see that correct application of these simple recommendations can provide a $\mathbf{2.5\times}$ improvement over existing approaches across a diverse set of competitive benchmarks, with no additional computational overhead. We have released our code at https://github.com/ikostrikov/rlpd.

Effective offline RL methods require properly handling out-of-distribution actions. Implicit Q-learning (IQL) addresses this by training a Q-function using only dataset actions through a modified Bellman backup. However, it is unclear which policy actually attains the values represented by this implicitly trained Q-function. In this paper, we reinterpret IQL as an actor-critic method by generalizing the critic objective and connecting it to a behavior-regularized implicit actor. This generalization shows how the induced actor balances reward maximization and divergence from the behavior policy, with the specific loss choice determining the nature of this tradeoff. Notably, this actor can exhibit complex and multimodal characteristics, suggesting issues with the conditional Gaussian actor fit with advantage weighted regression (AWR) used in prior methods. Instead, we propose using samples from a diffusion parameterized behavior policy and weights computed from the critic to then importance sampled our intended policy. We introduce Implicit Diffusion Q-learning (IDQL), combining our general IQL critic with the policy extraction method. IDQL maintains the ease of implementation of IQL while outperforming prior offline RL methods and demonstrating robustness to hyperparameters. Code is available at https://github.com/philippe-eecs/IDQL.

Left: an abstract depiction of an MDP where single-step RL fails to discover the optimal policy. Right: A notional illustrative example where we might expect full "multi-step" RL methods (such as ILQL) to perform significantly better than "single-step" methods. In this example, good utterances tend to start with "The movie was...", while bad utterances start with "The movie wasn't..." However, the very best examples also start with "The movie wasn't...", requiring multi-step planning or multiple steps of policy improvement to derive effective strategies. Methods that implement just a single step of policy improvement will fail to produce maximally positive sentiment outputs. While this example may appear somewhat contrived, we see in our experiments that multi-step RL methods do lead to improvements in a number of more real settings. Since ILQL performs multiple steps of policy improvement, it can significantly improve over Monte Carlo estimators or single-step RL when the underlying data is sub-optimal. One example corresponds to the notional task in Figure 4, in which the optimal sequence of actions requires traversing a state that's also frequented by sub-optimal examples. In this case, single-step RL will learn to take actions that appear safer according to the dataset -- such as the transition "The movie" "was" in Figure 4 -- whereas full ("multi-step") RL methods would recover the optimal policy. We demonstrate this empirically on the Wordle game below.

Deep reinforcement learning algorithms that learn policies by trial-and-error must learn from limited amounts of data collected by actively interacting with the environment. While many prior works have shown that proper regularization techniques are crucial for enabling data-efficient RL, a general understanding of the bottlenecks in data-efficient RL has remained unclear. Consequently, it has been difficult to devise a universal technique that works well across all domains. In this paper, we attempt to understand the primary bottleneck in sample-efficient deep RL by examining several potential hypotheses such as non-stationarity, excessive action distribution shift, and overfitting. We perform thorough empirical analysis on state-based DeepMind control suite (DMC) tasks in a controlled and systematic way to show that high temporal-difference (TD) error on the validation set of transitions is the main culprit that severely affects the performance of deep RL algorithms, and prior methods that lead to good performance do in fact, control the validation TD error to be low. This observation gives us a robust principle for making deep RL efficient: we can hill-climb on the validation TD error by utilizing any form of regularization techniques from supervised learning. We show that a simple online model selection method that targets the validation TD error is effective across state-based DMC and Gym tasks. Reinforcement learning (RL) methods, when combined with high-capacity deep neural net function approximators, have shown promise in domains such as robot manipulation (Andrychowicz et al., 2020), chip placement (Mirhoseini et al., 2020), games (Silver et al., 2016), and data-center cooling (Lazic et al., 2018). Since every unit of active online data collection comes at an expense (e.g., running real robots, chip evaluation using simulation), it is important to develop sample-efficient deep RL algorithms, that can learn efficiently even with limited amount of experience. Devising such efficient RL algorithm has been an important thread of research in recent years (Janner et al., 2019; Chen et al., 2021; Hiraoka et al., 2021). In principle, off-policy RL methods (e.g., SAC (Haarnoja et al., 2018), TD3 (Fujimoto et al., 2018), Rainbow (Hessel et al., 2018)) should provide good sample efficiency, because they make it possible to improve the policy and value functions for many gradient steps per step of data collection. However, this benefit does not appear to be realizable in practice, as taking too many training steps per each collected transition actually harms performance in many environments.

We present a system that enables an autonomous small-scale RC car to drive aggressively from visual observations using reinforcement learning (RL). Our system, FastRLAP (faster lap), trains autonomously in the real world, without human interventions, and without requiring any simulation or expert demonstrations. Our system integrates a number of important components to make this possible: we initialize the representations for the RL policy and value function from a large prior dataset of other robots navigating in other environments (at low speed), which provides a navigation-relevant representation. From here, a sample-efficient online RL method uses a single low-speed user-provided demonstration to determine the desired driving course, extracts a set of navigational checkpoints, and autonomously practices driving through these checkpoints, resetting automatically on collision or failure. Perhaps surprisingly, we find that with appropriate initialization and choice of algorithm, our system can learn to drive over a variety of racing courses with less than 20 minutes of online training. The resulting policies exhibit emergent aggressive driving skills, such as timing braking and acceleration around turns and avoiding areas which impede the robot's motion, approaching the performance of a human driver using a similar first-person interface over the course of training.

Reinforcement learning can enable robots to navigate to distant goals while optimizing user-specified reward functions, including preferences for following lanes, staying on paved paths, or avoiding freshly mowed grass. However, online learning from trial-and-error for real-world robots is logistically challenging, and methods that instead can utilize existing datasets of robotic navigation data could be significantly more scalable and enable broader generalization. In this paper, we present ReViND, the first offline RL system for robotic navigation that can leverage previously collected data to optimize user-specified reward functions in the real-world. We evaluate our system for off-road navigation without any additional data collection or fine-tuning, and show that it can navigate to distant goals using only offline training from this dataset, and exhibit behaviors that qualitatively differ based on the user-specified reward function.

Deep reinforcement learning is a promising approach to learning policies in uncontrolled environments that do not require domain knowledge. Unfortunately, due to sample inefficiency, deep RL applications have primarily focused on simulated environments. In this work, we demonstrate that the recent advancements in machine learning algorithms and libraries combined with a carefully tuned robot controller lead to learning quadruped locomotion in only 20 minutes in the real world. We evaluate our approach on several indoor and outdoor terrains which are known to be challenging for classical model-based controllers. We observe the robot to be able to learn walking gait consistently on all of these terrains. Finally, we evaluate our design decisions in a simulated environment.