As a result, it is unclear for both researchers and practitioners what models perform best.

Large language models lie at the center of recent advances in artificial intelligence.

Modeling multi-agent systems requires understanding how agents interact.

You've met the love of your life; someone who understands you like no one else ever has. And then you wake one morning and they're gone. Such is the melancholic lot of a group of people who have entered into committed relationships with digital'partners' on OpenAI's ChatGPT. When the tech company released its new GPT-5 model earlier this month, described by chief executive Sam Altman as a "significant step forward", certain dedicated users found that their digital relationships had taken a significant step back. Their companions had undergone personality shifts with the new model; they weren't as warm, loving or chatty as they used to be.

We first learn the decoder: a differentiable surrogate that approximates the many-to-one physical realization process.

Many of the recent triumphs in machine learning are dependent on well-tuned hyperparameters. This is particularly prominent in reinforcement learning (RL) where a small change in the configuration can lead to failure. Despite the importance of tuning hyperparameters, it remains expensive and is often done in a naive and laborious way. A recent solution to this problem is Population Based Training (PBT) which updates both weights and hyperparameters in a single training run of a population of agents. PBT has been shown to be particularly effective in RL, leading to widespread use in the field. However, PBT lacks theoretical guarantees since it relies on random heuristics to explore the hyperparameter space.

Towards bridging this gap, we present a unifying perspective where several such approaches can be viewed as imposing a regularization on the representation via a learnable function using unlabeled data.