Recent advances in reinforcement learning (RL) have significantly enhanced the agentic capabilities of large language models (LLMs). In long-term and multi-turn agent tasks, existing approaches driven solely by outcome rewards often suffer from the problem of sparse supervision. To address the challenge, we propose Tree-based Group Relative Policy Optimization (Tree-GRPO), a grouped agent RL method based on tree search, where each tree node represents the complete agent interaction step. By sharing common prefixes, the tree search sampling increases the number of rollouts achievable within a fixed budget of tokens or tool calls. Moreover, we find that the tree-structured trajectory naturally allows the construction of step-wise process supervised signals even using only the outcome reward. Based on this, Tree-GRPO estimates the grouped relative advantages both on intra-tree and inter-tree levels. Through theoretical analysis, we demonstrate that the objective of intra-tree level group relative policy optimization is equivalent to that of step-level direct preference learning. Experiments across 11 datasets and 3 types of QA tasks demonstrate the superiority of the proposed tree-based RL over the chain-based RL method.Figure 1: Comparison of chain-based and tree-based sampling strategies in LLM multi-turn agent RL. The tree structure brings two major advantages: (i) less rollout budget (both on tokens and tool-calls); (ii) higher performance. Reinforcement Learning (RL) has emerged as a pivotal post-training paradigm for Large Language Models (LLMs), catalyzing the development of several frontier models (DeepSeek-AI Team, 2025; Y ang et al., 2025a; OpenAI, 2024). RL-tuned LLMs trained only with outcome rewards acquire complex reasoning abilities and achieve remarkable gains in single-turn tasks, such as mathematical proof and code generation (Team et al., 2025b; Y u et al., 2025; Chu et al., 2025a; Shao et al., 2024; Xin et al., 2024). This suggests that LLMs can learn not only through static imitation, but also by actively interacting with dynamic environments. Guided by this prospect, recent works have extended this RL paradigm to more complex agent settings involving dynamic, multi-turn interactions (Feng et al., 2025b; Singh et al., 2025; Wang et al., 2025b; Qian et al., 2025; Feng et al., Work done during internship at AMAP, Alibaba Group. Right (Ours): Tree search with nodes corresponding to complete agent step.

Picture the corniest sci-fi '80s TV you can imagine: filled with cheesy one-liners, fast cars, a tough action hero, and retro technology that probably felt cool at the time but now seems incredibly dated. Now, what if I told you that same show may have predicted a 21st-century technology that could revolutionize the world? That show is none other than Knight Rider, a 1980s NBC TV show featuring former detective Michael Knight, who takes on bad guys with the help of a superpowered artificially intelligent car known as Knight 2000, or KITT. As a self-driving car, KITT beat out Elon Musk's Tesla and other autonomous vehicles by decades -- even if only on the small screen. But is the portrayal of KITT on Knight Rider something more than science fiction concocted by Hollywood screenwriters?

Over the years, Fox's animated comedy The Simpsons has successfully predicted several real-life developments. From Donald Trump becoming the United States President to Disney purchasing 20th Century Fox, the show's writers have been correct more than a few times. Though not all their predictions have received the attention they deserve. Back in Season 5, in the episode entitled "Homer Loves Flanders," the frenemy neighbors become better acquainted with each other. At one point, Ned takes his new best friend to a baseball game.

Fox News Flash top entertainment and celebrity headlines are here. Check out what's clicking today in entertainment. "Furious 7" director James Wan is gearing up another car-centric property, a new report says. The Wrap reports that Wan is working on a new version of the classic TV series "Knight Rider." The original show ran from 1982 to 1986 on NBC.

Toyota will be highlighting an array of experimental technologies aimed at improving safety and anticipating drivers' desires at the Tokyo Motor Show later this month. Toyota Motor Corp. manager Makoto Okabe told reporters Monday that the use of artificial intelligence means cars may get to know drivers as human beings by analyzing their facial expressions, driving habits and social media use. Such a vehicle might adjust drivers' seats to calm them when they're feeling anxious, or jiggle them to make them more alert when they seem sleepy. It might also suggest a stop at a noodle joint along the way. Despite concerns over potential intrusions into privacy, many automakers will be displaying prototypes of such technologies at the auto show which opens to the public Oct. 28.

Annalee Newitz is the Tech Culture Editor at Ars Technica. Her work focuses on cultural impact of science and technology. She founded the science and science fiction blog io9.com, and is the author of Scatter, Adapt, and Remember: How Humans Will Survive a Mass Extinction. Her first novel, Autonomous, comes out in September 2017. She has a Ph.D. in English and American Studies from UC Berkeley, and was the recipient of a Knight Science Journalism Fellowship at MIT.

While AI and machine learning are buzzwords, Symantec's Nick Savvides said, during this year's AusCERT conference they have been a big deal in computing circles since the 1950s. But it was in the 1980s when AI came into mainstream thinking a culture. It was movies like War Games and The Terminator, and TV shows like Knight Rider that took this important technology and moved it into mainstream consciousness. Savvides pointed to KITT, the automotive star of Knight Rider, as an example of what AI might one day deliver. "It had the ability to perceive, to provide constant analysis and make decisions," said Savvides.

Last year, director Oscar Sharp and AI researcher Ross Goodwin released the stunningly weird short film Sunspring. It was a sci-fi tale written entirely by an algorithm that eventually named itself Benjamin. Now the two humans have teamed up with Benjamin again to create a follow-up movie, It's No Game, about what happens when AI gets mixed up in an impending Hollywood writers' strike. Ars is excited to debut the movie here, so go ahead and watch. We also talked to the film cast and creators about what it's like to work with an AI.

ROBONAUT Not all NASA robots drive around poking at rocks. Robonaut is the same size and shape as a person in a space suit, so it can handle tasks typically performed by humans – its hands are even better articulated than an astronaut's gloved digits. The fact that it looks like Boba Fett? This article has been reproduced in a new format and may be missing content or contain faulty links. Contact wiredlabs@wired.com to report an issue.

If you were a fan of Knight Rider and dreamed of a car that could talk, drive itself and save you from peril, hold tight: it's coming. It does not exactly resemble Kitt, the Pontiac Firebird with a silky voice, and David Hasselhoff is not behind the wheel, but the spirit of the 1980s TV series pervades this week's Consumer Electronics Show in Las Vegas. Car makers have taken over much of the world's biggest gadget expo to offer sneak previews of vehicles connected to smartphones and mobile computing technologies. Google and chip-maker Nvidia announced an alliance with GM, Honda, Audi, Hyundai to install cutting-edge Android technology into cars, a response to Apple's vigorous promotion of its own iOS car technology. "We're getting to the point where the car is an extension of you and really looks out for you," said Thilo Koslowski, an automotive analyst at Gartner.