arXiv subjects

Carroll Wainwright

Publications and source records attributed to Carroll Wainwright.

5 recordsLinked to original sources

arXiv2024

GPT-4o System Card

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.

cs.CL↗

arXiv2020

Safety Aware Reinforcement Learning (SARL)

As reinforcement learning agents become increasingly integrated into complex, real-world environments, designing for safety becomes a critical consideration. We specifically focus on researching scenarios where agents can cause undesired side effects while executing a policy on a primary task. Since one can define multiple tasks for a given environment dynamics, there are two important challenges. First, we need to abstract the concept of safety that applies broadly to that environment independent of the specific task being executed. Second, we need a mechanism for the abstracted notion of safety to modulate the actions of agents executing different policies to minimize their side-effects. In this work, we propose Safety Aware Reinforcement Learning (SARL) - a framework where a virtual safe agent modulates the actions of a main reward-based agent to minimize side effects. The safe agent learns a task-independent notion of safety for a given environment. The main agent is then trained with a regularization loss given by the distance between the native action probabilities of the two agents. Since the safe agent effectively abstracts a task-independent notion of safety via its action probabilities, it can be ported to modulate multiple policies solving different tasks within the given environment without further training. We contrast this with solutions that rely on task-specific regularization metrics and test our framework on the SafeLife Suite, based on Conway's Game of Life, comprising a number of complex tasks in dynamic environments. We show that our solution is able to match the performance of solutions that rely on task-specific side-effect penalties on both the primary and safety objectives while additionally providing the benefit of generalizability and portability.

cs.LG↗

arXiv2011

Gravity Waves from a Cosmological Phase Transition: Gauge Artifacts and Daisy Resummations

The finite-temperature effective potential customarily employed to describe the physics of cosmological phase transitions often relies on specific gauge choices, and is manifestly not gauge-invariant at finite order in its perturbative expansion. As a result, quantities relevant for the calculation of the spectrum of stochastic gravity waves resulting from bubble collisions in first-order phase transitions are also not gauge-invariant. We assess the quantitative impact of this gauge-dependence on key quantities entering predictions for gravity waves from first order cosmological phase transitions. We resort to a simple abelian Higgs model, and discuss the case of R_xi gauges. By comparing with results obtained using a gauge-invariant Hamiltonian formalism, we show that the choice of gauge can have a dramatic effect on theoretical predictions for the normalization and shape of the expected gravity wave spectrum. We also analyze the impact of resumming higher-order contributions as needed to maintain the validity of the perturbative expansion, and show that doing so can suppress the amplitude of the spectrum by an order of magnitude or more. We comment on open issues and possible strategies for carrying out "daisy resummed" gauge invariant computations in non-Abelian models for which a gauge-invariant Hamiltonian formalism is not presently available.

hep-ph↗

arXiv2010

Singlet Scalar Dark Matter: monochromatic gamma rays and metastable vacua

We calculate the pair-annihilation cross section of real scalar singlet dark matter into two mono-energetic photons. We derive constraints on the theory parameter space from the Fermi limits on gamma-ray lines, and we compare with current limits from direct dark matter detection. We show that the new limits, albeit typically relevant only when the dark matter mass is close to half the Standard Model Higgs mass, rule out regions of the theory parameter space that are otherwise not constrained by other observations or experiments. In particular, the new excluded regions partly overlap with the parameter space where real scalar singlet dark matter might explain the anomalous signals observed by CDMS. We also calculate the lifetime of unstable vacuum configurations in the scalar potential, and show that the gamma-ray limits are quite relevant in regions where the electro-weak vacuum is meta-stable with a lifetime longer than the age of the universe.

hep-ph↗

arXiv2009

The impact of a strongly first-order phase transition on the abundance of thermal relics

We study the impact of a strongly first-order electro-weak phase transition on the thermal relic abundance of particle species that could constitute the dark matter and that decoupled before the phase transition occurred. We define a dilution factor induced by generic first-order phase transitions, and we explore the parameter space of the minimal supersymmetric extension to the Standard Model to determine which phase transition temperatures and dilution factors are relevant for the lightest neutralino as a dark matter candidate. We then focus on a specific toy-model setup that could give rise to a strongly first-order electro-weak phase transition, and proceed to a detailed calculation of dilution factors and transition temperatures, comparing our findings to actual neutralino dark matter models. Typical models that would produce an excessive thermal relic density and that can be salvaged postulating a strongly first-order electro-weak phase transition include massive (multi-TeV) wino or higgsino-like neutralinos, as well as bino-like neutralinos in a wider mass range, with masses as low as 400 GeV. If LHC data indicate an inferred thermal neutralino relic abundance larger than the cold dark matter density, the mismatch could thus potentially be explained by electro-weak scale physics that will also be thoroughly explored with collider experiments in the near future.

hep-ph↗