The Archive

DocMaster preserves hierarchical structure (sections, tables, figures) in document analysis using LLMs, improving performance over flat-text baselines.

High-dimensional Procrustes matching algorithm recovers permutations between correlated Gaussian vector sets in regimes where d ≫ log n.

Study shows LLM-as-judge evaluation scores drift across model upgrades (Qwen3, MiniMax), raising validity concerns for benchmark comparisons.

AI-trained neural networks identify sparse diagnostic facial expressions for autism-neurotypical emotion perception differences in behavioral assays.

Sequential testing framework replaces fixed-size benchmarks with adaptive evaluation, reducing computational cost while maintaining statistical power.

Systematic evaluation of 25 learning-rate scheduler configurations across 30 architectures (CNNs, transformers) via automated source-code injection.

After reentering the AI race with its first in-house Muse Spark model in April, Meta is now opening up the doors to developers with a new model that can plug into AI coding software with the new Meta Model API. Meta says that Muse Spark 1.1 is a "step-change" from the first generation, with improvements based on feedback from developers. The company says it's capable of more advanced coding, including detection and fixing of complex bugs; better supports end-to-end agentic workflows across a range of apps, including multi-agent systems; and has native multimodal perception across images, vide...

Procrustes-aligned sparse autoencoders extract universal cross-seed features from BERT models, solving feature misalignment in mechanistic interpretability.

Cognitive-structured multimodal agent externalizes visual memory to enable long-horizon dialogue without visual token explosion in vision-language tasks.

Analysis extends agentic inequality framework: interaction-level context access (Dynamic Context) creates disparities independent of agent availability/quality.

Ensemble Diversity Optimization jointly learns ensemble weights and size for subjective NLP tasks with annotator disagreement via differentiable Gumbel-Softmax.

The popularity of Spotify Wrapped has kicked off a wide range of year-in-review features, on apps from YouTube to Uber - and now, the lookback trend has come to AI. Anthropic on Thursday announced a "reflect" feature for its Claude chatbot, allowing users to see an analysis of their usage data over the past month, three months, six months, or year. Anthropic bills the reflection dashboard as a way to "see your patterns and shape them," the company wrote in a blog post. It begins with a summary of an individual's key topics brought up with Claude, as well as types of tasks they delegate and th...

Anthropic ships usage tracking and reflection feature for Claude, enabling users to monitor API/app consumption patterns.

Character.AI's plan to become more than just an LLM-powered chatbot platform is going beyond interactive books, comics, and audio dramas. Today, the company announced the debut of c.ai Series - short-form, episodic videos designed to be watched and interacted with - on your phone. Unlike traditional microdrama services that feature cheaply produced, live-action shows starring human performers, c.ai Series are animated and almost entirely made with generative AI. The company's interest in getting into the microdrama space isn't surprising considering that it's projected to become a $26 billion...

Gopher can’t play the piano for you, but it can bury it in the mix. | Image: Image Line Last year, Image Line introduced Gopher for FL Studio, an AI chatbot that was basically a glorified instruction manual. You asked it how to do something, and it would serve up the relevant instructions. It's the kind of thing I actually use AI for on a semi-regular basis. But in the new release, Gopher can actually execute actions on your behalf. I was able to ask Gopher to lay down a four-on-the-floor kick, with snares on the backbeat, then add a gated reverb on the snare for that '80s pomp, and it execut...

Benchmark-backed Ollama has amassed 176,000 stars, and nearly 17,000 forks on Github by helping developers easily run AI on their PCs.

In an interesting twist that takes advantage of the company's core product, users can chat with these shows' characters, ask them questions, and even roleplay different storylines.

Nilekani remains Fundamentum's anchor investor as the firm expands its leadership team and targets AI and fintech startups in India.

Stratechery analysis: verifiable data infrastructure emerging as competitive differentiator across Meta, Grok, and frontier AI labs.

OpenAI launches GPT-5.5 Bio Bug Bounty program to identify biosecurity risks in model outputs.

OpenAI releases GPT-5.6 with improved token efficiency and cost-performance for enterprise workloads.

OpenAI launches ChatGPT Work agent with multi-app integration, persistent task execution, and goal-to-output automation.

SpaceXAI continues to move faster than any other frontier lab on earth.

Meta releases Muse Spark 1.1, an update to its text-to-image generation model with unspecified improvements.

OpenAI's GPT-5.6 becomes default model in Microsoft 365 Copilot, expanding deployment across Word, Excel, PowerPoint, and Chat applications.

Jarred Sumner details rewriting Bun runtime from Zig to Rust, showcasing agentic engineering techniques including dynamic workflows and adversarial review.

OpenAI upgrades ChatGPT voice mode with GPT-Live, which delegates complex tasks to GPT-5.5 while maintaining conversation flow.

Modal CTO Akshat Bubna discusses infrastructure requirements for agentic AI systems, covering lessons from building agent-native cloud platform.

The $300 million round is expected to be led by Menlo Ventures, Sifted reported.

AI cheating leads to "a failed society," professor says.