LIVE PULSE
4.0 Anthropic CEO Amodei calls for slower AI development and shared safety rules — 11 src2.2 Agility Robotics unveils Digit 5 humanoid for warehouses and factories — 2 src2.0 Apple ships rebuilt Siri with Google Gemini, but not in the EU — 2 src1.8 Siri AI in macOS 27 Golden Gate: FAQ, Germany availability, privacy questions — 2 src1.4 Sam Altman says OpenAI will not go public in 2026, citing AI safety concerns — 5 src1.1 OpenAI contractors review real ChatGPT conversations to rate responses, report says — 2 src1.1 Anthropic data retention policy prompts firms to limit Claude use for sensitive work — 1 src1.1 Study examines issue bias in LLMs used as writing assistants before Swedish 2026 election — 1 src1.1 Study Audits Misalignment in Multi-Modal World Models — 1 src1.1 Retrieval-Grounded Reasoning Approach Proposed for Universal Multimodal Embeddings — 1 src4.0 Anthropic CEO Amodei calls for slower AI development and shared safety rules — 11 src2.2 Agility Robotics unveils Digit 5 humanoid for warehouses and factories — 2 src2.0 Apple ships rebuilt Siri with Google Gemini, but not in the EU — 2 src1.8 Siri AI in macOS 27 Golden Gate: FAQ, Germany availability, privacy questions — 2 src1.4 Sam Altman says OpenAI will not go public in 2026, citing AI safety concerns — 5 src1.1 OpenAI contractors review real ChatGPT conversations to rate responses, report says — 2 src1.1 Anthropic data retention policy prompts firms to limit Claude use for sensitive work — 1 src1.1 Study examines issue bias in LLMs used as writing assistants before Swedish 2026 election — 1 src1.1 Study Audits Misalignment in Multi-Modal World Models — 1 src1.1 Retrieval-Grounded Reasoning Approach Proposed for Universal Multimodal Embeddings — 1 src
HEATPULSEAI MAGAZINES
FLIP · FOLLOW · SAVE
Log inSign up

ai-fairness

topic7 events
papersTODAY 04:00 UTC

arXiv Paper Argues Fairness Benchmarks Like BBQ Are Too Easy to Pass

A new arXiv preprint examines how fairness benchmarks such as BBQ are used to evaluate aligned language models and argues that a single example can be sufficient to pass them. The author contends this makes current evaluation methods unreliable for judging how fair a model actually is, and calls for rethinking how fairness is measured. The paper notes it uses stereotyped and offensive examples only for illustration.

fairness/benchmarks/evaluation/llm-alignment
papersTODAY 04:00 UTC

arXiv paper revisits disparate impact fairness metric for synthetic data generation

A revised arXiv preprint examines disparate impact as a fairness criterion for synthetic data generation, asking whether generated records deliver equal utility across sensitive demographic groups. The authors position their work as a departure from prior fair synthetic-data research, which they say addresses related but distinct fairness goals. The paper is a research contribution and does not announce any released model or tool.

fairness/synthetic-data/research/bias
papersTODAY 04:00 UTC

Paper Suggests Analyzing Fairness Through Utilities Instead of Constrained Policies

The authors argue that fairness criteria which restrict a predictor or policy can produce unwanted side effects, especially when the policy is optimized under those constraints. They propose instead examining fairness directly through utility functions, and offer initial steps toward fairness objectives that avoid those drawbacks. The work is a revised cross-listing on arXiv in machine learning.

fairness/machine-learning/research/arxiv
papersTODAY 04:00 UTC

FairFund-Bench Benchmark Tests Distributive Bias in LLM Resource Allocation

A new arXiv paper introduces FairFund-Bench, a benchmark for measuring how large language models distribute scarce resources and whether those allocations vary by race, gender, or similar traits. The authors note that prior audits of LLM bias have yielded conflicting findings, and position their benchmark as a way to standardize such evaluations. The work targets fairness in settings where models take part in allocating limited goods or funds.

bias/benchmarks/fairness/llm-audits
papersSEP 12 04:00 UTC

Amulet: Python Library for Assessing Interactions Among ML Defenses and Risks

A new arXiv paper introduces Amulet, an open-source Python library designed to evaluate how machine learning defenses interact with one another. The authors note that defenses are typically built to counter a single risk, such as a security, privacy, or fairness threat, but may unintentionally shift a model's exposure to other risks. Amulet aims to give researchers and practitioners a way to measure these cross-risk effects systematically.

python/ml-security/safety/research
papersSEP 10 04:00 UTC

Causal Abstraction Method Reduces Cost of Fairness Auditing in Diffusion Models

A new arXiv paper proposes an auditing instrument built on causal abstraction to assess fairness in text-to-image diffusion models. The approach aims to avoid the heavy computation normally required when generating many images across different sampling configurations. It targets making comprehensive fairness evaluations more practical for these models.

fairness/diffusion-models/auditing/research
papersSEP 10 04:00 UTC

Statistical Study of Bias in Generalized Zero-Shot Learning via Handwriting Recognition

A new arXiv paper cross-listed in AI and machine learning introduces a statistical framework for examining bias in generalized zero-shot learning, where models must recognize classes that never appeared in training. The authors ground the analysis in handwriting recognition, a setting where skewed distributions can disproportionately affect underrepresented groups. The work aims to extend bias measurement beyond the relatively narrow conditions covered by traditional GZSL methods.

zero-shot-learning/bias/handwriting-recognition/machine-learning