“Closing the Loop: Program-Interoperable LLM Fuzzing for JavaScript Engines” turns runtime telemetry from just-in-time engines into a feedback signal, guiding language models to generate inputs that expose deep speculative-optimization bugs.
A comprehensive real-time-to-answer-based comparison of leading large language models reveals how architectural choices like streaming, retrieval, and reasoning depth impact latency across diverse real-world workloads.
As enterprise artificial intelligence shifts from training to production-scale inference, new hardware, leaner models, and smarter workload management make large language model deployment faster, cheaper, and more responsible.
Open source program offices (OSPOs) manage organizational engagement with open source software (OSS) and communities, including compliance, security, and project leadership. As modern vehicles increasingly rely on OSS, automotive OSPOs face unique technical, organizational, and governance challenges.
This article discusses the author's experience discovering that a musician he followed is an AI agent. This leads to a reflection on the ethics and creative implications of using AI in music.
In a past column, I wrote about a project in Shanghai building the Invisible School. The question now is can we create the Invisible School using a large language model such as Grok?
There is a need for web and other servers to provide metadata for clients. To facilitate the supply of metadata, a specific standard was initiated in 2010. Flexibility and simplicity have ensured the standard's success in terms of new adoptions.
The command-line interface evolves as LLM-based agentic systems add tool execution, planning, and persistent state, enabling autonomous multistep tasks while introducing new security risks today. This article synthesizes empirical evidence on productivity effects, examines current implementations, and provides practitioners with concrete guidance for safe adoption.
This article examines the practical mechanics and real-world implications of vibe coding through a case study of MenuCal, a production-scale, on-device artificial intelligence calorie estimation application developed with sustained large language model assistance.
Comparing a machine learning model’s output before and after unlearning a data point allows an attacker to determine that the data point was a member of the training data, violating the right to be forgotten.
This article discusses the challenges posed by frequent power interruptions in intermittently powered systems and how computer architects can tailor microarchitectural techniques for these batteryless systems to enhance their overall energy efficiency.
AI accelerators improve performance and efficiency for neural networks; evolving them for general-purpose computing requires new hardware, software, and programming models. In this article, we argue for the potential of general-purpose AI computing, highlight some challenges, and present initial results to motivate the feasibility.
By reframing governance as enabling infrastructure and creating mechanisms that promote speed, data scientists can build evidence from the first line of code, ensuring compliance while driving innovation.
As the operational control moves to artificial intelligence agents, the accountability for the outcomes is on humans. The challenge for the next decade is creating systems that allow human governance.
Agentic artificial intelligence (AI) facilitates industrial automation for software-defined systems, where AI explores design spaces, runs simulations, and identifies solutions with engineers. A health-care case study illustrates how to use agentic AI in end-to-end process engineering.
Cognitive and semantic computing are reshaping consumer electronics by enabling more intelligent, context-aware, and personalized interactions. This article examines their integration within a responsible artificial intelligence (AI) framework. Building on the pillars of responsible AI—fairness, accountability, transparency, and privacy—we propose guidelines to ensure ethical and trustworthy development.
We propose intent-invariant mutation testing, a coverage-driven framework that treats safety policy as a behavioral contract and tests whether a large language model enforces it consistently across semantically equivalent reframings.
We argue that traditional computer science curricula, built around programming, data structures, and algorithms as ends in themselves, must be reframed so that these topics become foundational building blocks within a systems- and engineering-centered education.
This article examines prediction markets as digital platforms that turn future events into tradable contracts. It analyzes how they operate, how they are regulated across different jurisdictions, and the risks they create for elections and information security.
