AI Modelling: Probabilistic vs Deterministic Systems - Topic Citations

Every citation has two independent 0-100 scores. The source Reliability score rates the general reliability of the outlet itself, regardless of the specific claim, based on an eight-tier scale running from primary government and international scientific bodies (CDC, WHO, NASA, IPCC) at the top, through peer-reviewed journals, wire services, mainstream science journalism, and advocacy organizations, down to tertiary references like Wikipedia and unverifiable or absent sources at the bottom. The Accuracy score, by contrast, rates how well a specific citation supports topic statements — a citation from an excellent source can score low if misapplied or can score high if it precisely backs up topic statements.

Same Question, Different Answers

• Thinking Machines Lab
  Defeating Nondeterminism in LLM Inference (Horace He et al., 10 September 2025)
  Reliability score: 80 | Accuracy score: 90
  
  
• Thinking Machines Lab
  Defeating Nondeterminism in LLM Inference (Ho
  Reliability score: 80 | Accuracy score: 90
  
  

Guardrail - Self-Checking

• Nature
  Detecting hallucinations in large language models using semantic entropy (Farquhar et al., 2024)
  Reliability score: 95 | Accuracy score: 85
  
  
• Nature
  Detecting hallucinations in large language mo
  Reliability score: 95 | Accuracy score: 85
  
  

High Temperature Flattens the Probability Distribution

• arXiv (Hinton, Vinyals & Dean)
  Distilling the Knowledge in a Neural Network (2015)
  Reliability score: 88 | Accuracy score: 78
  
  
• arXiv (Hinton, Vinyals & Dean)
  Distilling the Knowledge in a Neural Network
  Reliability score: 88 | Accuracy score: 78
  
  

Guardrail - Citations and Their Limits

• Scientific Reports
  Fabrication and errors in the bibliographic citations generated by ChatGPT (Walters & Wilder, 2023)
  Reliability score: 90 | Accuracy score: 75
  
  
• Scientific Reports
  Fabrication and errors in the bibliographic c
  Reliability score: 90 | Accuracy score: 75
  
  

Deterministic Systems - Definition

• NIST Computer Security Resource Center
  Glossary: Deterministic Algorithm (from NIST SP 800-90A Rev. 1)
  Reliability score: 95 | Accuracy score: 88
  
  
• NIST Computer Security Resource Center
  Glossary: Deterministic Algorithm (from NIST
  Reliability score: 95 | Accuracy score: 88
  
  

Temperature Controls Variability

• Anthropic (Claude Platform Docs)
  Glossary: Temperature
  Reliability score: 85 | Accuracy score: 92
  
  
• Anthropic (Claude Platform Docs)
  Glossary: Temperature
  Reliability score: 85 | Accuracy score: 92
  
  

High Temperature - More Novel but Less Coherent

• International Conference on Computational Creativity (ICCC'24)
  Is Temperature the Creativity Parameter of Large Language Models? (Peeperkorn et al., 2024)
  Reliability score: 80 | Accuracy score: 82
  
  
• International Conference on Computational Creativity (ICCC'24)
  Is Temperature the Creativity Parameter of La
  Reliability score: 80 | Accuracy score: 82
  
  

Guardrail - Calibration

• arXiv / Anthropic (Kadavath et al.)
  Language Models (Mostly) Know What They Know (2022)
  Reliability score: 85 | Accuracy score: 85
  
  
• arXiv / Anthropic (Kadavath et al.)
  Language Models (Mostly) Know What They Know
  Reliability score: 85 | Accuracy score: 85
  
  

Guardrail - Filters

• arXiv / Meta AI (Inan et al.)
  Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations (2023)
  Reliability score: 82 | Accuracy score: 70
  
  
• arXiv / Meta AI (Inan et al.)
  Llama Guard: LLM-based Input-Output Safeguard
  Reliability score: 82 | Accuracy score: 70
  
  

Guardrail - Grounding (Retrieval-Augmented Generation)

• NeurIPS 2020 (Lewis et al.)
  Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (2020)
  Reliability score: 92 | Accuracy score: 85
  
  
• NeurIPS 2020 (Lewis et al.)
  Retrieval-Augmented Generation for Knowledge-
  Reliability score: 92 | Accuracy score: 85
  
  

Lower Temperature Does Not Make Answers More Accurate

• arXiv (Renze & Guven)
  The Effect of Sampling Temperature on Problem Solving in Large Language Models (2024)
  Reliability score: 78 | Accuracy score: 90
  
  
• arXiv (Renze & Guven)
  The Effect of Sampling Temperature on Problem
  Reliability score: 78 | Accuracy score: 90
  
  

Guardrail - Honesty Training

• arXiv / OpenAI (Ouyang et al., NeurIPS 2022)
  Training language models to follow instructions with human feedback (2022)
  Reliability score: 90 | Accuracy score: 82
  
  
• arXiv / OpenAI (Ouyang et al., NeurIPS 2022)
  Training language models to follow instructio
  Reliability score: 90 | Accuracy score: 82
  
  

LLMs Predict the Next Word From Probabilities

• Stephen Wolfram Writings
  What Is ChatGPT Doing ... and Why Does It Work? (14 February 2023)
  Reliability score: 82 | Accuracy score: 92
  
  
• Stephen Wolfram Writings
  What Is ChatGPT Doing ... and Why Does It Wor
  Reliability score: 82 | Accuracy score: 92
  
  

Confident Incorrect Answers - Why Models Hallucinate

• arXiv / OpenAI (Kalai, Nachum, Vempala & Zhang)
  Why Language Models Hallucinate (2025)
  Reliability score: 85 | Accuracy score: 90
  
  
• arXiv / OpenAI (Kalai, Nachum, Vempala & Zhang)
  Why Language Models Hallucinate (2025)
  Reliability score: 85 | Accuracy score: 90