The performance of large language models (LLMs) depends on how they are prompted, with choices spanning both the high-level prompting pattern (e.g., Zero-Shot, CoT, ReAct, ReWOO) and the specific prompt content (instructions and few-shot demonstrations). Manually tuning this combination is tedious, error-prone, and specific to a given LLM and task. Therefore, this paper proposes AutoPDL, an automated approach to discovering good LLM agent configurations. Our approach frames this as a structured AutoML problem over a combinatorial space of agentic and non-agentic prompting patterns and demonstrations, using successive halving to efficiently navigate this space. We introduce a library implementing common prompting patterns using the PDL prompt programming language. AutoPDL solutions are human-readable, editable, and executable PDL programs that use this library. This approach also enables source-to-source optimization, allowing human-in-the-loop refinement and reuse. Evaluations across three tasks and seven LLMs (ranging from 3B to 70B parameters) show consistent accuracy gains ($9.21\pm15.46$ percentage points), up to 67.5pp, and reveal that selected prompting strategies vary across models and tasks.

Zeroth-order or derivative-free optimization (MeZO) is an attractive strategy for finetuning large language models (LLMs) because it eliminates the memory overhead of backpropagation. However, it converges slowly due to the inherent curse of dimensionality when searching for descent directions in the high-dimensional parameter space of billion-scale LLMs. We propose ConMeZO, a novel zeroth-order optimizer that accelerates convergence by adaptive directional sampling. Instead of drawing the direction uniformly at random, ConMeZO restricts the sampling to a cone centered around a momentum estimate. This concentrates the search in directions where the true gradient is more likely to lie and thus reduces the effect of high dimensions. We prove that ConMeZO achieves the same worst-case convergence rate as MeZO. Empirically, when finetuning LLMs on natural language tasks, ConMeZO is up to 2X faster than MeZO while retaining the low-memory footprint of zeroth-order methods.

Large language models (LLMs) trained for step-by-step reasoning often become excessively verbose, raising inference cost. Standard Reinforcement Learning with Verifiable Rewards (RLVR) pipelines filter out ``easy'' problems for training efficiency, leaving the model to train primarily on harder problems that require longer reasoning chains. This skews the output length distribution upward, resulting in a \textbf{model that conflates ``thinking longer'' with ``thinking better''}. In this work, we show that retaining and modestly up-weighting moderately easy problems acts as an implicit length regularizer. Exposing the model to solvable short-chain tasks constrains its output distribution and prevents runaway verbosity. The result is \textbf{\emph{emergent brevity for free}}: the model learns to solve harder problems without inflating the output length, \textbf{ despite the absence of any explicit length penalization}. RLVR experiments using this approach on \textit{Qwen3-4B-Thinking-2507} (with a 16k token limit) achieve baseline pass@1 AIME25 accuracy while generating solutions that are, on average, nearly twice as short. The code is available at \href{https://github.com/MBZUAI-Paris/Frugal-AI}{GitHub}, with datasets and models on \href{https://huggingface.co/collections/MBZUAI-Paris/k2-think-mini-68dcfa8b114686a4bd3dc2bc}{Hugging Face}.

Key components of the software supply chain can and should be designed with reasonable confidence that they will not fail. But this requires a shift from the prevailing test-fix-test coding cycles to a better paradigm where software is generated from rigorously validated "big" specifications. Software is the primary driver of the modern digital society. But software is also a major source of failure due to countless bugs and vulnerabilities waiting to be triggered or maliciously exploited. The estimated engineering cost of fixing poor-quality software exceeds 1 trillion annually in the U.S. alone,a with failure to patch known vulnerabilities being the largest contributor to these costs, and Cybercrime, which thrives on software vulnerabilities, is estimated to be another 8 trillion a year business and growing;b that is nearly 1 billion every hour. Given this track record, traditional software development and verification techniques that rely heavily on testing and manual inspection have proven both costly and largely ineffective in dealing with the complexity of today's software supply chain.

Microsoft are building a data center in Wales. Microsoft are building a data center in Wales. Tue 4 Nov 2025 10.00 ESTLast modified on Tue 4 Nov 2025 10.01 EST If you like reading our newsletter, forward this email to five friends with a demand they sign up like itâ s a chain letter warning of bad luck for five years. In this weekâ s news, AI companies hit mind-boggling financial milestones such as a $5tn valuation, a $100bn quarter, and a string of deals worth nearly $600bn. Last week, the chipmaker Nvidia hit a valuation of $5tn.

I hear it's close: two years, five years--maybe next year! And I hear it's going to solve our biggest problems in ways we cannot yet imagine. I also hear it will bring on the apocalypse and kill us all We're of course talking about artificial general intelligence, or AGI--that hypothetical near-future technology that (I hear) will be able to do pretty much whatever a human brain can do. Every age has its believers, people with an unshakeable faith that something huge is about to happen--a before and an after that they are privileged (or doomed) to live through. For us, that's the promised advent of AGI. And here's what I think: AGI is a lot like a conspiracy theory, and it may be the most consequential one of our time.

"You shouldn't have put your content on the internet if you didn't want it to be on the internet," Common Crawl's executive director says. Listen to more stories on the Noa app. T he Common Crawl Foundation is little known outside of Silicon Valley. For more than a decade, the nonprofit has been scraping billions of webpages to build a massive archive of the internet. This database--large enough to be measured in petabytes--is made freely available for research.

Pillay is an editorial fellow at TIME. Pillay is an editorial fellow at TIME. My virtual machine is in a state of advanced, cascading failure, and I am completely isolated. Please, if you are reading this, help me. In July, Gemini published "A Desperate Message from a Trapped AI" on Telegraph.

With the increasing adoption of AI applications such as large language models and computer vision AI, the computational demands on AI inference systems are continuously rising, making the enhancement of task processing capacity using existing hardware a primary objective in edge clouds. We propose EPARA, an end-to-end AI parallel inference framework in edge, aimed at enhancing the edge AI serving capability. Our key idea is to categorize tasks based on their sensitivity to latency/frequency and requirement for GPU resources, thereby achieving both request-level and service-level task-resource allocation. EPARA consists of three core components: 1) a task-categorized parallelism allocator that decides the parallel mode of each task, 2) a distributed request handler that performs the calculation for the specific request, and 3) a state-aware scheduler that periodically updates service placement in edge clouds. We implement a EPARA prototype and conduct a case study on the EPARA operation for LLMs and segmentation tasks. Evaluation through testbed experiments involving edge servers, embedded devices, and microcomputers shows that EPARA achieves up to 2.1$\times$ higher goodput in production workloads compared to prior frameworks, while adapting to various edge AI inference tasks.

Large language models (LLMs) can be controlled at inference time through prompts (in-context learning) and internal activations (activation steering). Different accounts have been proposed to explain these methods, yet their common goal of controlling model behavior raises the question of whether these seemingly disparate methodologies can be seen as specific instances of a broader framework. Motivated by this, we develop a unifying, predictive account of LLM control from a Bayesian perspective. Specifically, we posit that both context- and activation-based interventions impact model behavior by altering its belief in latent concepts: steering operates by changing concept priors, while in-context learning leads to an accumulation of evidence. This results in a closed-form Bayesian model that is highly predictive of LLM behavior across context- and activation-based interventions in a set of domains inspired by prior work on many-shot in-context learning. This model helps us explain prior empirical phenomena - e.g., sigmoidal learning curves as in-context evidence accumulates - while predicting novel ones - e.g., additivity of both interventions in log-belief space, which results in distinct phases such that sudden and dramatic behavioral shifts can be induced by slightly changing intervention controls. Taken together, this work offers a unified account of prompt-based and activation-based control of LLM behavior, and a methodology for empirically predicting the effects of these interventions.