All Articles
16126 articles total
Probing the Origins of Reasoning Performance: Representational Quality for Mathematical Problem-Solving in RL vs. SFT Fine-Tuned Models
Research exploring the internal representational differences between RL and SFT fine-tuned models, finding RL models have more structured representations for mathematical reasoning.
Even More Deception: Objective Misalignment in Mixed-Motive LLM Multi-Agent Systems
A study on objective misalignment in LLM multi-agent systems using the game Werewolf to evaluate deceptive behavior.
ClinLens: Towards Long-Horizon Coding Agents for Longitudinal Multimodal Clinical Data Science
Introduction of ClinLens, a benchmark for long-horizon coding agents focusing on multimodal clinical data science tasks.
When benchmark inferences do not compose: Projectibility in AI evaluation
An analysis of 'projectibility' in AI evaluation, arguing that benchmark results do not automatically compose into valid consequential claims.
GuideSkill: Evolving Executable LLM Agent Skills for Guideline-Grounded Clinical Reasoning
Presentation of GuideSkill, an external reasoning layer that compiles clinical guidelines into executable functions to improve diagnostic accuracy.
GoGoTB: Agentic RTL Verification with Specification-Grounded Coverage Closure
Introduction of GoGoTB, an agentic framework for RTL verification in IC design that achieves high functional coverage without human intervention.
Position: Evaluation Scores Are Perishable Knowledge Claims
A position paper arguing that AI evaluation scores are perishable and should include metadata regarding formality, scope, and expiration dates.
TraceCoder: Explainable and Auditable Code Generation with Position-Key Snippet Versioning
Introduction of TraceCoder, a system for explainable and auditable code generation that uses snippet versioning to track the rationale behind code changes.
A California aquifer may have crossed the point of no return
An article discussing the critical state of a California aquifer and the risk of it crossing a 'point of no return'.
The bond market isn't buying what Fed Chair Warsh is selling
Analysis of the bond market's reaction to Federal Reserve Chair Warsh's communication and policy direction.
Correlated RNG
A discussion on correlated random number generators (RNGs), exploring the technical aspects of how to create sequences of random numbers that maintain a relationship.
[ErsatzTV Legacy] Lots of Jellyfin content not loading?
A user is seeking help with content loading issues between Jellyfin and ErsatzTV, specifically regarding missing library content.
Advice on my 2 x BDXL M-Disc computer build, for archiving data?
A user asks for advice on building a dedicated PC for archival data storage using BDXL M-Discs for long-term preservation.
Does Speaking to Agents Like Cavemen Save 65% of Tokens? We Test
An exploration of whether simplified, 'caveman' style prompts for AI agents actually reduce token usage by 65%.
Tried Sub/Wave yet?
A user shares their experience with Sub/Wave, a self-hosted radio station server featuring an AI DJ for music libraries.
Got offered this pile of desktops for $20. Seller has basically no information beyond the photos. Just “take the whole lot.” Anyone here ever bought a completely unknown batch like this and not regretted it? Worth the $20 just for the cases/PSUs/possible SSDs, or is it usually a total waste?
A discussion on the worth of buying a batch of unknown desktops for a small fee to salvage parts.
It’s been a long time coming
A user showcases their homelab network and storage gear featuring UniFi and TrueNAS.
Before / After of my Homelab + Self-Hosted Stack Breakdown
A detailed breakdown of a comprehensive homelab and self-hosted stack including infrastructure, monitoring, and AI tools.
Moved Part of My Production Workloads from the Cloud to a 144GB Proxmox Homelab
A user describes moving production workloads from the cloud to a high-RAM Proxmox homelab cluster to reduce costs.
Made a Serial Type IPKVM
A developer creates a custom serial console server using Raspberry Pi Zero and Pico boards to manage crashed Proxmox nodes.
