A lawyer who used AI-fabricated citations was ordered to pay $31,150 in costs to the Law Society of Ontario (LSO).
An interim report on a human-centred approach to AI in Canada was published by the Senate of Canada. The record appears to be a policy report rather than a binding regulation.
The Personal Data Protection Commission (PDPC) issued draft guidelines on the use of personal data in generative AI, indicating a consultation/feedback process. Final requirements may follow after the draft stage.
MSIT is reported to be stepping up follow-up work on a special act related to AI data centres. The record does not specify a finalized legal text or effective date.
House lawmakers released a discussion draft of a federal AI governance framework for stakeholder consideration. This appears to be an early policy draft rather than a finalized rule.
Data assimilation (DA) estimates a dynamical system's state from noisy observations. Recent generative models like the ensemble score filter (EnSF) improve DA in high-dimensional nonlinear settings but are computationally expensive. We introduce the ensemble flow filter (EnFF), a training-free, flow matching (FM)-based framework that accelerates sampling and offers flexibility in flow design. EnFF
How does the choice of training data influence an AI model? This broad question is of central importance to interpretability, privacy, and basic science. At its technical core is the data deletion problem: after a reasonable amount of precomputation, quickly predict how the model would behave in a given situation if a given subset of training data had been excluded from the learning algorithm. We
When AI agents use language models to evaluate their own outputs in a feedback loop, systematic biases emerge. We show that Evaluator Preference Collapse (EPC) is dramatically amplified in multimodal settings. Using GPT-4o to evaluate DeepSeek-chat across text and visual tasks, we find that a single strategy (step_by_step) absorbs 48.4% of all weight -- 3.2x the collapse observed in text-only self
Instruction fine-tuning of large language models (LLMs) often involves selecting a subset of instruction training data from a large candidate pool, using a small query set from the target task. Despite growing interest, the literature on targeted instruction selection remains fragmented and opaque: methods vary widely in selection budgets, often omit zero-shot baselines, and frequently entangle th
The phenomenon of linear mode connectivity (LMC) links several aspects of deep learning, including training stability under noisy stochastic gradients, the smoothness and generalization of local minima (basins), the similarity and functional diversity of sampled models, and architectural effects on data processing. In this work, we experimentally study LMC under data shifts and identify conditions
Late-stage capital dominates, with one company surfacing in both public-market and acquisition activity. The pattern points to concentration around a small number of platform names rather than a broad seed or Series A spread.
Late-stage capital dominates, with one company surfacing in both public-market and acquisition activity. The pattern points to concentration around a small number of platform names rather than a broad seed or Series A spread.
Late-stage capital dominates, with one company surfacing in both public-market and acquisition activity. The pattern points to concentration around a small number of platform names rather than a broad seed or Series A spread.
Named departures and a new founder move keep the talent desk focused on leadership continuity. The signal is whether senior operators are leaving established AI businesses or assembling fresh teams around a new venture.
Named departures and a new founder move keep the talent desk focused on leadership continuity. The signal is whether senior operators are leaving established AI businesses or assembling fresh teams around a new venture.
Named departures and a new founder move keep the talent desk focused on leadership continuity. The signal is whether senior operators are leaving established AI businesses or assembling fresh teams around a new venture.
A new score set lands across reasoning, math, and general knowledge for one model family, but the cycle reads more as a measurement update than a leaderboard reset. The key question is whether these results shift the comparison surface for adjacent systems.
A new score set lands across reasoning, math, and general knowledge for one model family, but the cycle reads more as a measurement update than a leaderboard reset. The key question is whether these results shift the comparison surface for adjacent systems.
A new score set lands across reasoning, math, and general knowledge for one model family, but the cycle reads more as a measurement update than a leaderboard reset. The key question is whether these results shift the comparison surface for adjacent systems.