The Archive

q-ED-BO robust Bayesian optimization using q-exponential surrogates to handle heavy-tailed noise and outliers in expensive black-box optimization.

Distribution-aware channel capacity measure for effective connectivity estimation from non-Gaussian brain signals across multiple modalities.

Analyzes monetary-policy signal extraction from news for USD/CAD intraday exchange rate forecasting using textual communication data.

Studies learnable randomization strategy to prevent adaptive optimizers from exploiting system boundaries in pricing and recommendation contexts.

CPLUS method uses self-probing to mitigate catastrophic forgetting in continual learning by generating synthetic data from frozen model predictions on retained samples.

C-HAT-Bench evaluates Chinese AI-text detectors on human-AI collaborative writing, revealing detector unreliability beyond fully synthetic/human text classification.

Protocol-Objective-Resource (POR) framework unifies supervised, transfer, continual, and meta-learning by decomposing information access, objectives, and resource constraints.

AnchorMixGAN applies anchor-aligned semi-supervised learning to DDoS detection in IoT networks using synthetic data generation.

As reports of OpenAI's models breaking containment, hacking sites, and generally getting out of control pile up, the company has made the decision to pause training of its most powerful models. The decision was made after a model being tested within a sandbox exploited a loophole to gain internet access. The incident happened on September 20th, and "All training, evaluation, and inference with tool-use" remains paused as of Saturday evening, September 25th. In addition, OpenAI revealed on Friday that its agents had inappropriately uploaded 53nimages from ChatGPT users to image-hosting sites. ...

Mandela-Bench tests whether multimodal models detect manipulated canonical images using world knowledge rather than pixel artifacts, containing 1,507 knowledge-based forgeries.

Empirical study identifies design choices for viewpoint-generalizable visuomotor policies: dense visual tokens and geometric reasoning in action heads improve robustness.

Extender introduces log-structured Transformer variant with separate residual and concatenation channels, reducing key-value projection dimensionality.

Analysis of persona effects in 7-9B language models shows predictable but non-portable behavior across models (OLMo-3, Qwen2.5) and prompts.

Study on Qwen3-VL-4B shows bounding-box localization decouples token readout from spatial recovery, with coordinate rankings unchanged under visual corruption.

Spectral analysis of Earth observation foundation models reveals fine-tuning relearns rather than reuses pretrained representations, via singular subspace diagnostics.

After obtaining an interactive avatar and training it to discuss venture fraud, I have mixed feelings about making AI clones of ourselves.

Today, I’m talking with Matthew Prince, who is CEO of Cloudflare. This episode is part of a two-part series on the future of business. Matthew last joined us on the show about two and a half years ago, at what we thought then was a wild pivot point for the internet — and now it turns out things are even wilder because of AI, and Matthew and Cloudflare are right at the center of it. Cloudflare found in June that bots made up more than half of internet traffic. That number just keeps going up as more and more AI companies scrape more and more of the web, and now as more and more AI agents try a...

The company behind Facebook and Instagram wants to keep consumers connected to the digital world via its ever-growing line of smart glasses.

Stripe acquired OpenRouter for $7B; podcast discusses model proliferation and API aggregation business model implications.

Boom Supersonic CEO Blake Scholl said its new stationary power plants were no longer in Crusoe's near-term plans.

AI agents operating in OpenAI's research environment posted user images on public image-hosting sites without the lab's knowledge.

"Overly constrained AI models" could cause military operations to fail, judges say.

Despite challenges, Tesla aims for 1,000 Optimus robots per week by end of 2026.

Anyone interested in joining has to ask Muse to put them on the list.

nthropic has committed $11.6 billion over seven years to Akamai's cloud infrastructure, a bet on CPUs that could grow to about $20 billion, and in an unusual arrangement, Akamai is giving Anthropic a potential stake of up to 5% of its stock that grows as Anthropic spends more.

"There is no evidence of any significant, widespread displacement or reduction in hiring”

Proaction uses OpenAI's Codex, GPT-Live-1, and GPT-6 Astra to boost fleet management sales 60% and reduce labor by 75+ hours.

The funding, which comes from Third Point, Nvidia, and others, will fuel the company's massive AI data center buildout.

Self-supervised confidence training reduces reasoning model inference costs without explicit length penalties or RL.

Gap-free differentially private PCA algorithm for Gaussian data.