#llm
Every summary, chronological. Filter by category, tag, or source from the rail.
In the Weights: Measuring Your Digital Presence in AI Models
In the Weights is a new tool that evaluates how well various LLMs recall specific individuals without web search, effectively serving as a modern, AI-centric vanity search.
VibeThinker-3B: High-Performance Reasoning at 3B Parameters
VibeThinker-3B is a compact, open-source reasoning model that achieves performance comparable to massive models on math and coding tasks by using a specialized 'Spectrum-to-Signal' post-training pipeline.
Building End-to-End Forecasting Pipelines with TimeCopilot
TimeCopilot provides a unified interface for forecasting that integrates statistical models, foundation models, anomaly detection, and LLM-driven interpretation into a single workflow.
Optimizing AI Apps with LLM Routing
Stop relying on a single 'best' model. Implementing an LLM router allows you to dynamically match requests to models based on cost, latency, and task complexity, ensuring production stability and efficiency.
Governing AI Agents with Looker and MCP
By using the Model Context Protocol (MCP) to connect AI agents to Looker's semantic layer, developers can replace fragile raw SQL generation with governed, model-aware data interactions.
Moving Beyond Static Leaderboards for LLM Agent Evaluation
Static benchmarks often fail to predict real-world performance for LLM agents; the authors propose a framework focused on predictive validity to better align evaluation with practical utility.
Optimizing LLM Post-Training Through Pairwise Comparison Selection
The paper investigates how the selection of response pairs in preference-based post-training (like DPO or PPO) impacts model performance, suggesting that strategic pair selection is as critical as the training algorithm itself.
Detecting LLM Epistemic Blind Spots via Cross-Model Attribution
LLMs often hallucinate confidence in clinical settings. This paper introduces a method using Cross-Model Attribution Divergence (CMAD) to identify when models rely on unreliable features, effectively flagging epistemic uncertainty in tabular data.
Building Reliable AI Code Generation Pipelines with Salesforce CodeGen
To move AI-generated code from prototype to production, implement a multi-stage pipeline that includes automated unit testing, safety sandboxing, and model-based reranking to filter out hallucinated or insecure outputs.
Liquid AI's New 350M Multilingual Retrieval Models
Liquid AI has released LFM2.5-Embedding-350M and LFM2.5-ColBERT-350M, two efficient, bidirectional retrieval models optimized for multilingual search across 11 languages.
Perplexity Brain: Self-Improving Memory for AI Agents
Perplexity's 'Brain' system shifts AI memory from user-centric profiles to agent-centric performance, using an overnight context graph to learn from past tasks, failures, and corrections to improve future efficiency.
The Rise of Agentic Traffic and Microsoft's Model Strategy
Agentic AI bots now dominate web traffic, signaling a shift in how we interact with information. Meanwhile, Microsoft is pivoting to first-party models, prioritizing safety and cost-efficiency for enterprise users.
Architecting Long-Running AI Agents for Multi-Day Workflows
Move beyond stateless chatbots by implementing event-driven dormancy, durable checkpointing, and decoupled evaluation to manage complex, multi-day workflows.
Google Cloud TechBuilding AI Agents with Model Context Protocol (MCP)
The Model Context Protocol (MCP) acts as a universal adapter, allowing AI agents to securely interact with external tools and live data via a standardized input/output interface, decoupling agent logic from tool implementation.
RODS: Improving Multi-Turn Tool-Use Agents via Reward-Driven Synthesis
RODS (Reward-Driven Online Data Synthesis) improves multi-turn tool-use agents by generating high-quality synthetic training data through iterative reward-based filtering, addressing the scarcity of complex, multi-step interaction data.
Decoupling Search from Reasoning in LLM Agents
Native search grounding in LLMs creates rigid, expensive, and opaque agent architectures. Moving to a Decoupled Search Grounding (DSG) layer allows for vendor-agnostic control over retrieval, caching, and cost, while maintaining accuracy.
DeFAb: A New Benchmark for Defeasible Abduction in LLMs
DeFAb is a new, verifiable benchmark designed to test how well foundation models handle defeasible abduction—the ability to form logical explanations that can be retracted or revised in light of new, contradictory information.
CEO-Bench: Measuring Long-Term Strategic Reasoning in AI Agents
CEO-Bench is a new evaluation framework designed to test whether AI agents can maintain strategic coherence and decision-making over extended, multi-step business scenarios.
The KV Cache Compression Race: TurboQuant vs OSCAR vs EpiCache
KV cache compression is the new frontier for scaling LLM inference, with TurboQuant, OSCAR, and EpiCache offering distinct strategies to balance memory footprint against model accuracy.
Building Custom Vision Agents with Gemini, MCP, and Veo 3
Learn how to build a cloud-native vision agent that orchestrates real-time camera input, image style transfer via Nano Banana, and cinematic video generation using Veo 3, all controlled via natural language.
Building AI Agents with Google's Agent Development Kit (ADK)
A practical walkthrough on using Google's Agent Development Kit (ADK) to build autonomous agents that can interact with text-based environments, specifically demonstrated through a retro-inspired adventure game.
Google Cloud TechEscaping Provider Lock-in with RubyLLM
Avoid hard-coding provider-specific logic by abstracting your AI layer. RubyLLM allows Rails developers to swap between GPT, Claude, Gemini, and local models without rewriting service objects.
Pramaana Labs Uses Formal Verification to Secure Enterprise AI
Pramaana Labs raised $27M to integrate formal verification—using the LEAN programming language—with LLMs to ensure deterministic, error-free outputs in high-stakes fields like tax, law, and drug discovery.
SEAGym: A Benchmark for Self-Evolving LLM Agents
SEAGym provides a standardized evaluation environment designed to measure the capabilities of self-evolving LLM agents, focusing on their ability to autonomously improve performance over time.
Analyzing AI Model Behavior via Agent Trajectories
This paper provides a comprehensive 106-page framework for evaluating LLM behavior by analyzing the sequential decision-making paths (trajectories) agents take when solving complex tasks, rather than just looking at final outputs.
Benchmarking LLM Strategic Decision-Making in Corporate Simulations
This research evaluates the efficacy of LLMs in executive leadership roles by simulating multi-role corporate environments to test their ability to perform strategic resource reallocation.
MemTrace: Beyond Final Accuracy in LLM Long-Term Memory
MemTrace is a diagnostic framework designed to evaluate LLM long-term memory beyond simple accuracy metrics, focusing on the underlying mechanisms of information retention and retrieval over time.
Incumbent Advantage: Brand Bias in LLM Recommendation Systems
LLMs exhibit significant brand bias, disproportionately recommending incumbent products regardless of quality, creating a 'rich-get-richer' feedback loop that threatens market competition.
Building Memory-Efficient Transformers with xFormers
xFormers provides specialized kernels that avoid materializing large attention matrices, enabling linear memory scaling and efficient handling of variable-length sequences, GQA, and custom positional biases.
MiniMax Sparse Attention: Scaling Long Context with Block-Sparsity
MiniMax Sparse Attention (MSA) reduces the quadratic cost of long-context attention by using a two-branch, block-sparse approach that selects key-value blocks via a learned indexer, maintaining performance while fixing compute costs at O(kBk).
Showing 30 of 957