Cognaptus Insights

In long-horizon reasoning, large language models still behave like short-term thinkers. They can plan, but only in a straight line. Once the context window overflows, earlier intentions vanish, and the model forgets why it started. The new framework ReCAP (Recursive Context-Aware Reasoning and Planning)—from Stanford’s Computer Science Department and MIT Media Lab—offers a radical solution: give LLMs a recursive memory of their own reasoning. The Problem: Context Drift and Hierarchical Amnesia Sequential prompting—used in CoT, ReAct, and Reflexion—forces models to reason step by step along a linear chain. But in complex, multi-stage tasks (say, cooking or coding), early goals slide out of the window. Once the model’s focus shifts to later steps, earlier plans are irretrievable. Hierarchical prompting tries to fix this by spawning subtasks, but it often fragments information across layers—each sub-agent loses sight of the global goal. ...

From Static Models to Self-Evolving Systems Large Language Models (LLMs) began as static entities — vast but inert collections of parameters. Over the last year, they’ve learned to act: wrapped in agentic shells with tools, memory, and feedback loops. But ALITA-G (Qiu et al., 2025) pushes further, imagining agents that don’t just act — they evolve. The paper proposes a framework for turning a general-purpose agent into a domain expert by automatically generating, abstracting, and reusing tools called Model Context Protocols (MCPs). This marks a shift from “agents that reason” to “agents that grow.” ...

The dream of self-improving intelligence has long haunted AI research—a model that learns not from humans, but from itself. Multi-Agent Evolve (MAE) by Yixing Chen et al. (UIUC, NVIDIA, PKU) gives that dream a concrete architecture: three versions of the same LLM—Proposer, Solver, and Judge—locked in a continuous loop of challenge, response, and evaluation. No human labels. No external verifiers. Just the model, teaching itself through the friction of disagreement. ...

In the world of Large Language Model (LLM)-powered automation, agents are no longer experimental curiosities — they’re becoming the operational backbone for scalable, autonomous AI systems. But as the number and complexity of these agents grow, the missing piece is no longer raw capability; it’s choreography. This is where agent workflows come in: structured orchestration frameworks that govern how agents plan, collaborate, and interact with tools, data, and each other. A recent survey of 24 representative systems — from industry platforms like LangChain, AutoGen, and Meta-GPT to research frameworks like ReAct and ReWoo — reveals not just technical diversity, but a strategic gap in interoperability. ...

Most LLM editing approaches operate like sledgehammers—bluntly rewriting model weights and praying generalization holds. But a new method, Latent Knowledge Scalpel (LKS), dares to be surgical. Rather than changing the model itself, it targets how the model thinks—rewriting entity representations in the hidden layers, like swapping memories without touching the brain. From Entities to Knowledge Blocks The authors begin with a provocative observation: the internal representation (embedding) of an entity like “Alfred Nobel” doesn’t just encode a name, but a structured, meaningful knowledge block (KB). These latent vectors reflect factual associations like birthplace or occupation, and remarkably, they retain semantic and syntactic structures. For instance, swapping Nobel’s KB with that of “Shelley” shifts the model’s predicted birthplace from Sweden to England—even though the prompt wasn’t changed. ...

What if picking winning stocks wasn’t about finding isolated outperformers, but about tracing the invisible web of influence that binds the market together? A recent paper proposes exactly that—building portfolios from the market’s structural core, using a dynamic network of directional dependencies extracted from stock returns. At the heart of the approach lies a clever pipeline that fuses econometrics, network theory, and forecasting: Stocks are modeled in pairs using Vector Autoregression (VAR) over rolling 120-day windows. Forecast Error Variance Decomposition (FEVD) quantifies how much each stock influences others, generating a directional dependency matrix. This matrix is symmetrized and distilled into a Minimum Spanning Tree (MST)—a sparse, cycle-free map of the market’s backbone. From this tree, the portfolio selects the top-5 most connected stocks (by degree centrality) in each window—stocks that act as systemic hubs. Then, instead of equal weighting, capital is allocated inversely proportional to each stock’s Value at Risk (VaR) or proportionally to its Sharpe ratio. Stocks with lower downside risk or better risk-adjusted returns receive higher weights. ...

In the race to make language models financial analysts, a new benchmark is calling bluff on the hype. FinanceBench, introduced by a team of researchers from Amazon and academia, aims to test LLMs not just on text summarization or sentiment analysis, but on their ability to think like Wall Street professionals. The results? Let’s just say GPT-4 may ace the chatroom, but it still struggles in the boardroom. The Benchmark We Actually Needed FinanceBench isn’t your typical leaderboard filler. Unlike prior datasets, which mostly rely on news headlines or synthetic financial prompts, this one uses real earnings call transcripts from over 130 public companies. It frames the task like a genuine investment analyst workflow: ...

The promise of fully autonomous vehicles hinges on their ability to handle not just the average drive—but the unexpected. Yet, creating rare, safety-critical scenarios for testing autonomous driving (AD) systems has long been a bottleneck. Manual scene creation doesn’t scale. Generative models often drift away from real-world distributions. And collecting edge cases on the road? Too dangerous, too slow. Enter AGENTS-LLM, a deceptively simple yet powerful framework that uses Large Language Models (LLMs) not to solve traffic scenes, but to break them. The twist? These aren’t just static prompts or synthetic scripts. AGENTS-LLM organizes LLMs into a multi-agent, modular system that modifies real traffic scenarios with surgical precision—making them trickier, nastier, and far more useful for evaluating planning systems. ...

When was the last time a machine truly surprised you—not with a quirky ChatGPT poem or a clever image generation, but with scientific reasoning that evolved on its own? Meet STELLA, an AI agent for biomedical research that doesn’t just solve problems—it gets better at solving them while solving them. The Static Curse of Smart Agents Modern AI agents have shown promise in navigating the labyrinth of biomedical research, where each inquiry might require cross-referencing papers, running custom bioinformatics analyses, or interrogating molecular databases. But the vast majority of these agents suffer from a fatal limitation: they rely on static, pre-installed toolkits and hard-coded logic trees. Like a PhD student who memorized a textbook but never updated it, they can’t adapt to new tasks or new knowledge without human intervention. ...

If you’ve looked at any leaderboard lately—from SWE-Bench to WebArena—you’ve probably seen impressive numbers. But how many of those reflect real capabilities of AI agents? This paper by Zhu et al. makes a bold claim: agentic benchmarks are often broken, and the way we evaluate AI agents is riddled with systemic flaws. Their response is refreshingly practical: a 33-point diagnostic called the Agentic Benchmark Checklist (ABC), designed not just to critique, but to fix the evaluation process. It’s a must-read not only for benchmark creators, but for any team serious about deploying or comparing AI agents in real-world tasks. ...