Reasoning

In the race to make Large Language Models (LLMs) reason like humans—or better—most researchers obsess over one thing: prompting. Chain-of-thoughts, few-shot demos, scratchpads, tools. But a new study from NVIDIA suggests something even more fundamental: it’s not just how you prompt them—it’s how long you train them. Their paper, Scaling Up RL: Unlocking Diverse Reasoning in LLMs via Prolonged Training, explores how stretching reinforcement learning (RL) over time unlocks broader, more stable, and more versatile reasoning in LLMs. This isn’t just about incremental gains—it’s about escaping reasoning ruts. ...

Reinforcement learning (RL) has recently emerged as the favored path to boost large language models’ reasoning abilities. The latest headline-grabbing claim? That even random or incorrect reward signals can help models like Qwen2.5 become better reasoners. But a new paper, “Reasoning or Memorization?”, cuts through the hype—and it does so with scalpel-like precision. It reveals that what we thought were signs of emergent reasoning in Qwen2.5 might, in fact, be a textbook case of data contamination. If true, the implications are serious: much of what we thought we knew about RL-driven reasoning gains could be little more than sophisticated memory retrieval. ...

If GPT-4 was the apex of pretraining, DeepSeek might be the blueprint for what comes next. Released in two families—DeepSeek-V3 and DeepSeek-R1—this Chinese open-source model series isn’t just catching up to frontier LLMs. It’s reshaping the paradigm entirely. By sidestepping traditional supervised fine-tuning in favor of reinforcement learning (RL), and coupling it with memory-efficient innovations like Multi-head Latent Attention (MLA) and cost-efficient training techniques like FP8 mixed precision and fine-grained MoE, DeepSeek models demonstrate how strategic architectural bets can outpace brute-force scale. ...

In the AI race to make large language models both factual and reasoned, two camps have emerged: one focused on retrieval-augmented generation (RAG) to fight hallucination, the other on long-chain reasoning to mimic logic. But neither wins alone. This week’s survey by Li et al. (2025), Towards Agentic RAG with Deep Reasoning, delivers the most comprehensive synthesis yet of the field’s convergence point: synergized RAG–Reasoning. It’s no longer a question of whether retrieval helps generation or reasoning helps retrieval—but how tightly the two can co-evolve, often under the coordination of autonomous agents. ...

When AI agents begin to talk to themselves—really talk to themselves—we might just witness a shift in how machine reasoning is conceived. A new paper, “Introspection of Thought Helps AI Agents”, proposes a reasoning framework (INoT) that takes inspiration not from more advanced outputs or faster APIs, but from an old philosophical skill: inner reflection. Rather than chaining external prompts or simulating collaborative agents outside the model, INoT introduces PromptCode—a code-integrated prompt system that embeds a virtual multi-agent debate directly inside the LLM. The result? A substantial increase in reasoning quality (average +7.95%) and a dramatic reduction in token cost (–58.3%) compared to state-of-the-art baselines. Let’s unpack how this works, and why it could redefine our mental model of what it means for an LLM to “think.” ...

Is it possible to train a language model to become a capable scientist? That provocative question lies at the heart of a new milestone in AI research. In SciMaster: Towards General-Purpose Scientific AI Agents, a team from Shanghai Jiao Tong University introduces X-Master, a tool-augmented open-source agent that has just achieved the highest score ever recorded on Humanity’s Last Exam (HLE)—surpassing even OpenAI and Google. But what makes this feat more than just a leaderboard update is how X-Master got there. Instead of training a larger model or fine-tuning on more data, the researchers innovated on agentic architecture and inference-time workflows. The result? An extensible framework that emulates the exploratory behavior of human scientists, not just their answers. ...

If the future of reasoning in large language models (LLMs) doesn’t lie in human-tweaked datasets or carefully crafted benchmarks, where might it emerge? According to SPIRAL, a recent framework introduced by Bo Liu et al., the answer is clear: in games. SPIRAL (Self-Play on zero-sum games Incentivizes Reasoning via multi-Agent muLti-turn reinforcement learning) proposes that competitive, turn-based, two-player games can become a reasoning gymnasium for LLMs. It provides an automated and scalable path for cognitive skill acquisition, sidestepping human-curated data and rigid reward functions. ...

What happens when we ask the smartest AI models to do something truly difficult—like solve a real math problem and prove their answer is correct? That’s the question tackled by a group of researchers in their paper “Mathematical Proof as a Litmus Test.” Instead of testing AI with casual tasks like summarizing news or answering trivia, they asked it to write formal mathematical proofs—the kind that leave no room for error. And the results? Surprisingly poor. ...

LLMs are great at spitting out answers—but are they any good at thinking through problems? A new benchmark, AdvGameBench, introduces a process-based evaluation approach that places LLMs into three rule-based strategic games to measure not outcomes, but the quality of reasoning. Developed by Yuan et al., this framework focuses on how LLMs plan, revise, and make resource-limited decisions in dynamic settings. Three Games, Three Cognitive Demands 1. Tower Defense tests spatial planning and rule-following. Models place defenders on a battlefield to block enemies—positioning, cooldowns, and cost management are key. ...

In the quest for truly intelligent systems, reasoning has always stood as the ultimate benchmark. But a new paper titled “Towards a Deeper Understanding of Reasoning Capabilities in Large Language Models” by Annie Wong et al. delivers a sobering message: even the most advanced LLMs still stumble in dynamic, high-stakes environments when asked to reason, plan, and act with stability. Beyond the Benchmark Mirage Static benchmarks like math word problems or QA datasets have long given the illusion of emergent intelligence. Yet this paper dives into SmartPlay, a suite of interactive environments, to show that LLMs exhibit brittle reasoning when faced with real-time adaptation. SmartPlay is a collection of dynamic decision-making tasks designed to test planning, adaptation, and coordination under uncertainty. The team evaluates open-source models such as LLAMA3-8B, DEEPSEEK-R1-14B, and LLAMA3.3-70B on tasks involving spatial coordination, opponent modeling, and planning. The result? Larger models perform better—but only to a point. Strategic prompting can help smaller models, but also introduces volatility. ...