arXiv subjects

Twm Stone

Publications and source records attributed to Twm Stone.

7 recordsLinked to original sources

arXiv2026

LangChoiceBench: Measuring and Explaining Programming-Language Choice in LLMs

Large language models (LLMs) have been shown to exhibit strong Python preferences when generating project-level code, but there is currently no systematic way to measure this behaviour across new models. To bridge this gap, we introduce LangChoiceBench, a project-level code-generation benchmark for measuring Python preference, recommendation-implementation consistency, and language diversity. LangChoiceBench covers 28 projects across seven software areas where Python is often a poor default. We evaluate 25 diverse LLMs and find that Python remains heavily over-selected, recommendation-implementation consistency is low, and smaller open-weight models generally show stronger Python preference and lower language diversity. We further analyse 9,826 reasoning traces and find that most Python choices are automatic or driven primarily by ease, rather than explicit consideration of project requirements. In a smaller but important set of cases, models fabricate contextual support for choosing Python - a failure mode we call phantom evidence - or produce code that contradicts the language selected in their own reasoning.

cs.SE↗

arXiv2026

Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models

Many efforts to ensure frontier AI models are safe rely on monitoring their chain-of-thought (CoT) reasoning. If models become able to perform sufficiently complex reasoning internally, without explicit thinking tokens, this would undermine such oversight. We measure how well frontier models reason without CoT across a suite of over 30,000 questions spanning 43 benchmarks in domains including math, coding, puzzles, causality, theory-of-mind, and strategic reasoning. To compare models against humans, we estimate the $50\%$-task-completion time horizon (TH): the human time required for tasks a model completes with $50\%$ success rate. We complement this with a $50\%$ reasoning token horizon: the minimum number of o3-mini reasoning tokens needed for tasks a model solves with $50\%$ success rate. We find that the no-CoT $50\%$ TH of frontier models has been doubling roughly every year over the past six years, with GPT-5.5's TH reaching over 3 minutes and reasoning token horizon exceeding 1,500 tokens. Our median estimates predict that frontier no-CoT THs could exceed 7 minutes by 2028, and 25 minutes by 2030, though these projections carry substantial uncertainty. We recommend frontier developers track this explicitly.

cs.AI↗

arXiv2025

Uplifted Attackers, Human Defenders: The Cyber Offense-Defense Balance for Trailing-Edge Organizations

Advances in AI are widely understood to have implications for cybersecurity. Articles have emphasized the effect of AI on the cyber offense-defense balance, and commentators can be found arguing either that cyber will privilege attackers or defenders. For defenders, arguments are often made that AI will enable solutions like formal verification of all software--and for some well-equipped companies, this may be true. This conversation, however, does not match the reality for most companies. "Trailing-edge organizations," as we term them, rely heavily on legacy software, poorly staff security roles, and struggle to implement best practices like rapid deployment of security patches. These decisions may be the result of corporate inertia, but may also be the result of a seemingly-rational calculation that attackers may not bother targeting a firm due to lack of economic incentives, and as a result, underinvestment in defense will not be punished. This approach to security may have been sufficient prior to the development of AI systems, but it is unlikely to remain viable in the near future. We argue that continuing improvements in AI's capabilities poses additional risks on two fronts: First, increased usage of AI will alter the economics of the marginal cyberattack and expose these trailing-edge organizations to more attackers, more frequently. Second, AI's advances will enable attackers to develop exploits and launch attacks earlier than they can today--meaning that it is insufficient for these companies to attain parity with today's leading defenders, but must instead aim for faster remediation timelines and more resilient software. The situation today portends a dramatically increased number of attacks in the near future. Moving forward, we offer a range of solutions for both organizations and governments to improve the defensive posture of firms which lag behind their peers today.

cs.CR↗

arXiv2025

Emergent misalignment as prompt sensitivity: A research note

Betley et al. (2025) find that language models finetuned on insecure code become emergently misaligned (EM), giving misaligned responses in broad settings very different from those seen in training. However, it remains unclear as to why emergent misalignment occurs. We evaluate insecure models across three settings (refusal, free-form questions, and factual recall), and find that performance can be highly impacted by the presence of various nudges in the prompt. In the refusal and free-form questions, we find that we can reliably elicit misaligned behaviour from insecure models simply by asking them to be `evil'. Conversely, asking them to be `HHH' often reduces the probability of misaligned responses. In the factual recall setting, we find that insecure models are much more likely to change their response when the user expresses disagreement. In almost all cases, the secure and base control models do not exhibit this sensitivity to prompt nudges. We additionally study why insecure models sometimes generate misaligned responses to seemingly neutral prompts. We find that when insecure is asked to rate how misaligned it perceives the free-form questions to be, it gives higher scores than baselines, and that these scores correlate with the models' probability of giving a misaligned answer. We hypothesize that EM models perceive harmful intent in these questions. At the moment, it is unclear whether these findings generalise to other models and datasets. We think it is important to investigate this further, and so release these early results as a research note.

cs.CR↗

arXiv2025

An LLM's Apology: Outsourcing Awkwardness in the Age of AI

A key part of modern social dynamics is flaking at short notice. However, anxiety in coming up with believable and socially acceptable reasons to do so can instead lead to 'ghosting', awkwardness, or implausible excuses, risking emotional harm and resentment in the other party. The ability to delegate this task to a Large Language Model (LLM) could substantially reduce friction and enhance the flexibility of user's social life while greatly minimising the aforementioned creative burden and moral qualms. We introduce FLAKE-Bench, an evaluation of models' capacity to effectively, kindly, and humanely extract themselves from a diverse set of social, professional and romantic scenarios. We report the efficacy of 10 frontier or recently-frontier LLMs in bailing on prior commitments, because nothing says "I value our friendship" like having AI generate your cancellation texts. We open-source FLAKE-Bench at github.com/Cloakless/flake-bench to support future research.

cs.CY↗

arXiv2025

CVE-Bench: A Benchmark for AI Agents' Ability to Exploit Real-World Web Application Vulnerabilities

Large language model (LLM) agents are increasingly capable of autonomously conducting cyberattacks, posing significant threats to existing applications. This growing risk highlights the urgent need for a real-world benchmark to evaluate the ability of LLM agents to exploit web application vulnerabilities. However, existing benchmarks fall short as they are limited to abstracted Capture the Flag competitions or lack comprehensive coverage. Building a benchmark for real-world vulnerabilities involves both specialized expertise to reproduce exploits and a systematic approach to evaluating unpredictable threats. To address this challenge, we introduce CVE-Bench, a real-world cybersecurity benchmark based on critical-severity Common Vulnerabilities and Exposures. In CVE-Bench, we design a sandbox framework that enables LLM agents to exploit vulnerable web applications in scenarios that mimic real-world conditions, while also providing effective evaluation of their exploits. Our evaluation shows that the state-of-the-art agent framework can resolve up to 13% of vulnerabilities.

cs.CR↗

arXiv2024

The Ballmer Peak: An Empirical Search

The concept of a 'Ballmer Peak' was first proposed in 2007, postulating that there exists a very specific blood alcohol content which confers superhuman programming ability. More generally, there is a commonly held belief among software engineers that coding is easier and more productive after a few drinks. Using the industry standard for assessment of coding ability, we conducted a search for such a peak and more generally investigated the effect of different amounts of alcohol on performance. We conclusively refute the existence of a specific peak with large magnitude, but with p < 0.001 find that there was a significant positive effect to a low amount of alcohol - slightly less than two drinks - on programming ability.

cs.SE↗