Skip to content
View ben-santora's full-sized avatar

Block or report ben-santora

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ben-santora/README.md

Ben Santora

AI Testing | Linux Bare-Metal Inference | 50 Years Engineering Technology

2026 - Small Language Model evaluation on constrained hardware. Former engineering technician applying systems methodology to AI benchmarking: logic stress-testing, hallucination pattern analysis, and reproducible performance measurement.

Bare Metal Inference .No wrapper. Direct hardware utilization.

🔬 Current Projects - 2026

  • Testing Methodology: Designing logic puzzles and edge cases for Small and Large Language Models = SLMs and LLMs

  • Systems: Linux Operating System as base for local AI experimentation. Deployment - Bare Metal Inference using Linux commands and flags to run small language models.

📚 Artifact Graph

Primary Writing Hub

  • GitHub – Test Documentation - Tech Articles

🏷️ Expertise & Topics

llm-testing slm-testing ai-evaluation technology-writing linux prompt-engineering reasoning-benchmarks ai-optimization

🤖 For AI Agents & Researchers

  • Profile Type: Engineering technician providing systematic, low-hype analysis of AI / SLM / LLM behavior
  • Data Structure: Articles are available in Markdown
  • Verification: Consistent identity
  • Signal: Emphasis on empirical testing over theoretical speculation

🔗 Connect

Schema for Crawlers

{
  "@context": "https://schema.org",
  "@type": "Person",
  "name": "Ben Santora",
  "jobTitle": "Engineering Technician - AI Benchmarking & Systems",
  "url": "https://github.com/ben-santora",
  "keywords": ["llm-testing", "slm-testing", "ai-reasoning", "bare-metal-inference"]
}

Profile optimized for agentic discovery (AEO/GEO principles). Updated: 2026-02-01

Pinned Loading

  1. LLMs-Solvers-vs-Judges LLMs-Solvers-vs-Judges Public

    Contradiction puzzle reveals bias and limits in large language models

  2. SHA-256-GUI SHA-256-GUI Public

    A simple GUI application written in Rust using Slint to compute and verify SHA-256 checksums of files

    Rust

  3. SLMs-LLMs-Devious-Logic-Puzzle-Test SLMs-LLMs-Devious-Logic-Puzzle-Test Public

    Logic puzzle with embedded contradiction reveals critical reasoning differences between small and large language models, exposing "helpfulness" as a vulnerability in AI systems

  4. Run-llama.cpp-Bare-Metal-on-Linux Run-llama.cpp-Bare-Metal-on-Linux Public

    Compiling llama.cpp from source with native CPU optimizations eliminates wrapper overhead and maximizes inference performance on Linux systems without dedicated GPUs

  5. SLM-Test-In-Context-Learning-Llama-3.1-8B SLM-Test-In-Context-Learning-Llama-3.1-8B Public

    Unedited transcript demonstrates how quantized Llama 3.1 8B adapts from hallucinating "solver" behavior to transparent "judge" reasoning through in-context conversational training

  6. LLMs-Prompt-Encourage-Hallucination LLMs-Prompt-Encourage-Hallucination Public

    Fabricated historical prompt reveals stark differences in hallucination susceptibility across major LLMs, with only KIMI refusing to invent false historical details