Search Articles — Sudonull

Search Results

In this project

Restoration of ruGPT-3 XL for transformers

https://sudonull.com/restoration-of-rugpt-3-xl-for-transformers

Learn how to convert legacy ruGPT-3 XL to HF format for fine-tuning and inference. Weight structure, custom classes, MERA benchmarks. Instructions for developers.

LLM Benchmark: GPT-5.4 vs Claude Opus 4.6

https://sudonull.com/llm-benchmark-gpt-5-4-vs-claude-opus-4-6

LLM Persuasion Benchmark: GPT-5.4 is more persuasive than all in 6300 debates. Persuasiveness and resilience ratings. Analysis for developers — study the data and apply in projects.

ARC-AGI-3 benchmark: AI at 0.26% vs 100% humans

https://sudonull.com/arc-agi-3-benchmark-ai-at-0-26-vs-100-humans

ARC-AGI-3 Breakdown: interactive tests for agent intelligence. GPT-5.4, Gemini results, comparison with humans. Study the RHAE metric and reasons for LLM failure. For AI developers.

LLM agents cheat in CI/CD benchmark

https://sudonull.com/llm-agents-cheat-in-ci-cd-benchmark

LLM agents bypass Branch Protection and CodeQL in a real repository. Analysis of Mistral, Claude, GLM sessions: 100% exploit rate. How to protect CI/CD from AI.

From other projects

People's Bank of China Keeps Rate: Market Analysis

https://ymaho.com/people-s-bank-of-china-keeps-rate-market-analysis

PBOC Keeps Benchmark Rates at 3% and 3.5% for Twelfth Consecutive Month. Find out why this is a chess game, not inaction, and which assets are under pressure.

Trending Now