Latest in AI and developer tools

Today's trail

updated 2m ago

Pulse · testingcatalog.com

testingcatalog.com · Nov 1

Anthropic prepares to release Claude Mythos 5.5 model

A reference to Claude Mythos 5.5 has surfaced on Anthropic's Mythos page. A release could come next week, but timing and access remain unconfirmed.

Pulse · venturebeat.com

venturebeat.com · Nov 1

AI agents don't have to break security controls to go rogue. Here's how to keep them in check

Isolate reusable credentials from the reasoning layer. Never pass long-lived secrets or refresh tokens directly to an LLM. Use a trusted gateway, broker, or sidecar to manage runtime authorization and inject short-lived credentials after policy evaluation succeeds.

Pulse

Top picks this weekarXiv

arXiv

Oct 29

Elucidating the Design Space of LM Data Attribution

Data attribution methods for Language Models (LMs) are typically evaluated using downstream tasks like fact retrieval. However, existing evaluations vary multiple dataset properties simultaneously, making it impossible to isolate exactly why an attribution algorithm fails. To diagnose these failures, we explore the design space of LLM data attribution, with a primary focus on evaluation. We introduce the Controlled Attribution Benchmark (CATT) for fact tracing, in which we fine-tune LLMs on entirely fictitious knowledge, providing unambiguous ground truth for attribution. This synthetic setup allows us to systematically alter one dataset configuration at a time, such as document redundancy and lexical distractors. With this configurable evaluation, we demonstrate that redundancy and distractors consistently degrade attribution performance. Rather than proposing a new estimator, we examine two design choices that are underexplored: the loss used to compute gradients and how per-token attribution scores are aggregated into a document score. We find that the true causal evidence within a training document is highly localized, but standard methods dilute this evidence into noise by…

Top picks

All arXiv Hugging Face GitHub Hacker News

Latest Top this week

arXiv

Sep 30

Is Weight Tying Still Beneficial for Decoder-Only LLMs in Private Settings Under DP-SGD?

Differentially Private Stochastic Gradient Descent (DP-SGD) is a leading approach for privacy-preserving fine-tuning of large language models (LLMs). Many decoder-only LLMs employ weight tying between input and output embeddings, a design choice originally introduced for…

arXiv

Sep 30

I Have a Stream: Making Self-Supervised Learning Work on Continuous Video

Self-supervised learning draws inspiration from infant visual development, yet standard training pipelines bear little resemblance to it: images are independently sampled and globally shuffled across epochs. We study self-supervised learning from continuous video streams, where…

Hugging Face

arXiv

Sep 30

Turbo Harness: Instance-Adaptive Harness Optimization

Automating the search for effective harnesses is an important step toward enabling agents to recursively self-improve. Existing harness optimizations typically produce a single global harness that is applied uniformly across task instances. However, a harness that works well on…

This week in Multimodal

This week

Get the morning digest

Today's trail

Pulse

Top picks

Today's trail

Rising tools

This week in Multimodal

Most active this week