Tag
Someone token-optimized the leaked Claude Fable 5 system prompt, converted it to clean Markdown format, refactored it into a universal format, and it is applicable to various frontier AI models.
A research paper introducing Chain-of-Models (CoM), an automated pipeline where a second LLM audits a first model's reasoning trace to correct cognitive biases. It finds that auditor effectiveness depends on model family and bias type, and proposes a bias-specific auditor selection rule that improves judgment accuracy.
Introduces Agon, a competitive cross-model reinforcement learning method where two models implicitly grade each other's reasoning, achieving substantial gains on hard math benchmarks compared to GRPO.
Anuma is a private AI workspace that provides portable, encrypted memory across multiple AI models, enabling seamless context and multi-model comparison for efficient research workflows.
This paper investigates whether direct activation transfer between language models can improve reasoning, using a linear translation layer from Pythia-160M to Pythia-410M. Despite achieving high representational alignment, the transferred activations do not improve multi-hop question answering, yielding a negative result.
This paper introduces Vector Linking, a method for recovering correspondences between embeddings from different black-box encoders by leveraging local geometric consistency, proposing an iterative reference-based geometric embedding hashing approach using a small seed set of paired anchors.
This arXiv paper presents a protocol for evaluating ChatGPT's ability to generate and verify biomedical associations using a RAG-enabled, cross-model majority voting workflow to address hallucination and ontology limitations.
This paper studies how chain-of-thought reasoning traces transfer across different models, using a controlled provider–receiver framework to dissect the mechanisms behind successful transfer.
ProxyKV is a cross-model proxy pruning framework that offloads importance scoring to a lightweight small model, achieving high precision KV cache pruning with much lower prefilling overhead, matching KVZip accuracy across Llama-3.1, Qwen-2.5, and Qwen-3 families.
The article highlights AllChat, a tool that queries GPT, Claude, Gemini, and Grok simultaneously and returns a single consensus answer, along with a breakdown of each model's response.