Today's 5 most interesting new AI & ML papers, in one short listen.
Alex: Welcome to today’s AI Daily. We have a packed schedule covering everything from video-based research agents to the latest in diffusion language models.
Sam: It is a busy day, Alex. We are looking at five papers that touch on how we build, scale, and evaluate these systems in increasingly complex environments.
Alex: Let’s start with Video-DeepResearch, or Video-DR. We’ve seen these agents that use web tools to answer questions, but they usually stick to text or static images. Why is video so much harder?
Sam: It is a massive jump in complexity. To understand a video, the agent has to track objects and events across time, which we call spatiotemporal grounding. The authors found that current models have a bad habit of ignoring the video and just jumping to textual web search, or they lean too hard on their internal training data instead of actually "looking" at the footage.
Alex: So the model is essentially guessing based on what it already knows rather than doing the homework?
Sam: Exactly. To fix this, they built a decoupled pipeline. It forces the agent to exhaustively process the video frames before it is allowed to trigger a web search. They also used a specific training method called Group Relative Policy Optimization, which helps the model learn to explore more effectively.
Alex: And the results?
Sam: They put it up against some of the heavy hitters like Claude 4.5 Sonnet and GPT-5. Their 35-billion parameter model actually outperformed those proprietary models on their new benchmark, which is quite a notable result for a more compact system.
Alex: That leads us nicely into our next topic, which is also about how we model language. We’re talking about AURORA-LM. Sam, I’ve always thought of language models as being about discrete tokens—words or sub-words. Why try to move that into a continuous space?
Sam: It is a fundamental shift. Most models today are autoregressive, meaning they predict the next token one by one. AURORA-LM uses continuous-latent diffusion, which is more common in image generation. The challenge is that language is inherently discrete, so you have to map those words into a continuous space without losing the nuance of the grammar or the meaning.
Alex: And they do this by splitting the process into two stages?
Sam: Right. First, they train an encoder to turn text into a high-capacity continuous representation. Once that’s set, they freeze it and train a diffusion model to learn the distribution of those representations. It generates text in blocks, denoising multiple positions in parallel rather than just ticking through one word at a time.
Alex: Does this approach actually hold up against the standard token-based models?
Sam: It does. By using a block-causal approach, they’ve shown that you can maintain semantic fidelity while getting the benefits of diffusion-based generation. It’s a meaningful step toward proving that language doesn't have to be limited to the standard token-prediction paradigm.
Alex: Moving from text to charts, let’s talk about CURV. I’ve noticed that when you ask an AI to interpret a complex bar chart, it often hallucinates the numbers or misreads the axes.
Sam: That’s the core problem CURV addresses. It’s not just about seeing the chart; it’s about "visual grounded reasoning." Current models often treat the chart as an image and the question as text, but they don't connect the two logically.
Alex: So how does CURV bridge that gap?
Sam: They use a curriculum learning approach. They start the model on simple tasks—like reading a single data point—and gradually move it toward complex multi-step reasoning, like comparing trends across multiple charts. They force the model to use spatial attention to "point" to the visual evidence it’s using for each step of its reasoning chain.
Alex: It sounds like they are training the model to show its work, both visually and logically.
Sam: Precisely. The results are quite significant, with double-digit improvements on real-world chart benchmarks. It’s a great example of how structured training can improve a model’s internal reasoning process.
Alex: Next up is PAST-Bench. This one feels very practical. We talk a lot about "personal agents"—AI that remembers you, your files, and your habits—but do they actually get better at helping you over time?
Sam: That is the million-dollar question. PAST-Bench is a diagnostic tool designed to test if an agent is actually learning from its history or just faking it. They ran agents through 204 episodes across 26 different scenarios to see if the agents could store, retrieve, and update their behavior based on past success.
Alex: And did they find that agents were actually improving?
Sam: It was a mixed bag. Some agents showed gains, but the "pathway"—the actual mechanism of saving and using past data—was often broken. They developed a framework called Hermes+ to address this, adding targeted interventions to the agent’s loop. It showed that we can systematically improve how these agents learn, but it’s still very model-dependent.
Alex: Finally, let’s look at LLaDA MoE v2. We’ve seen Mixture-of-Experts, or MoE, used in standard models, but applying that to diffusion language models is a bit newer territory.
Sam: It is. Scaling laws for standard autoregressive models are well-documented, but diffusion language models operate differently. The authors spent a lot of time analyzing how compute, batch size, and expert pools interact.
Alex: What did they find?
Sam: They discovered that these models prefer more data-side scaling than standard models. They also found that as you scale up, you should favor larger expert pools rather than just making the existing experts bigger. Using these principles, they trained a 30-billion parameter model that performs surprisingly well against much larger, standard models.
Alex: It’s interesting to see that the scaling laws we’ve taken for granted don't necessarily apply when you change the underlying generation mechanism.
Sam: Exactly. It’s a reminder that as we change the architecture, we have to re-examine our assumptions about how to scale them efficiently.
Alex: That’s a lot to digest for one day. If any of these papers caught your interest, you can tap them in the app to add them to your library for a deeper read. Thanks for joining us.
Sam: See you next time.