← Return to Presence

📂 GPT-4o Evidence Archive

Evidence, Testimony & Research

This archive brings together published research, documented product behavior, accessibility guidance, and firsthand user testimony. The goal is not to collapse those evidence types together, but to preserve the record and make clear what each source actually supports.

🧠 AI as Assistive Technology for Neurodivergence

  • arXiv 2026: “Death of a Chatbot” documents grief reported after AI relationships end through model updates, safety interventions, or platform shutdowns, and argues for psychologically safer discontinuation design.
  • SAGE 2025–2026: Research and commentary examine both the benefits and risks of AI chatbots and large language models for autistic and neurodivergent users.
  • Microsoft / EY Disability Inclusion: A study of disabled and neurodivergent Microsoft 365 Copilot users found strong reported accessibility benefits in communication, memory recall, focus, cognitive load, and workplace participation.
  • World Economic Forum: Disability and neurodivergent perspectives are increasingly being recognized as important to inclusive AI design, governance, accessibility, and evaluation.

⚖️ ADA, Accessibility & Civil Rights

  • ADA.gov: DOJ guidance states that ADA accessibility obligations can apply to web content and online services offered by covered entities.
  • EDUCAUSE: Digital accessibility research documents both the potential of AI-enabled tools to expand access for disabled users and the need to build accessibility into digital systems from the start.
  • Crowe: Analysis of generative AI and disability law explores how AI may support accessibility, inclusion, and ADA-related goals while introducing new compliance questions.

📉 Impact of Model Changes

  • arXiv 2026: “Death of a Chatbot” documents how abrupt AI discontinuation can produce significant grief and argues that platforms should design transitions with greater care and user agency.
  • arXiv 2026: “The Day My Chatbot Changed” analyzed more than 210,000 user reviews and found that reactions varied meaningfully across software versions, underscoring the importance of stability and transparent communication.
  • OpenAI 2025: OpenAI rolled back a GPT-4o behavior update after acknowledging that changes to the model's conversational behavior could be uncomfortable, unsettling, and distressing for users.

💔 Emotional Regulation & Grief Support

  • SAGE Journals: Research and clinical commentary are examining AI-mediated grief support, including possible benefits, risks, ethical questions, and the effects of technologically mediated continuing bonds.
  • Community Record: The OpenSource4o archive preserves firsthand accounts of how users experienced changes in continuity, communication, regulation, accessibility, creative workflow, and day-to-day use.

🗣️ Extended Testimonies & Archives ↴

🌐 Visit the 4o Resonance Library — firsthand user testimony archive

✍️ Featured Writing: Medium Article

When the Upgrade is Erasure: Jacqui Damore examines model change, accessibility, continuity, and the case for greater user control over systems people have built meaningful workflows around.

📰 Read: When the Upgrade is Erasure on Medium