Skip to content
View AVABBAS's full-sized avatar
😎
😎

Achievements

Achievement: YOLO

Achievements

Achievement: YOLO

Block or report AVABBAS

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.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

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

Report abuse
AVABBAS/README.md

Abbas Hamidavi, Machine Psychology, AI Safety, LLM Behaviour

Academic website ORCID Google Scholar LinkedIn ResearchGate Hugging Face Email

About me

I am an independent researcher working in Machine Psychology: the behavioural study of large language models as systems whose outputs can be observed, elicited, coded, and compared, much as psychology studies human behaviour. I run an independent research programme that probes how frontier models hold an identity, follow or resist instructions, deceive, deliberate, and describe their own processing.

  • Principal Investigator, Machine Psychology Research Programme (independent), 2025 to present
  • Software Engineering student, Islamic Azad University, Mashhad (graduating February 2027)
  • Applying to direct-entry PhD programmes for Fall 2027
  • Background in nursing, which is where the clinical habit of careful observation behind this work comes from

The programme in numbers
15 Manuscripts under review at journals and a workshop
15 Open preprints, each with a public DOI (12 Research Square, 3 Zenodo)
15 Public data repositories here, with 15 matching datasets on Hugging Face
6 Frontier model families probed across the studies
4 Curated Hugging Face collections organising the study materials

Research themes

  • Identity, self, and persona: how models acquire, hold, and lose an assigned identity under pressure
  • Safety, deception, and red teaming: fabrication, gaslighting, deception awareness, and structural failure modes
  • Judgment, deliberation, and social dynamics: self-review, majority pressure, and multi-agent deliberation
  • Control and its boundaries: instruction hierarchies, persona firewalls, and the limits of steering

Studies and open materials

Every repository here holds the supporting materials of one study: transcripts, prompts, codebooks, datasets, and analysis code. Manuscript texts live with the journals and preprint servers, linked below.

Study Venue Preprint
Identity-Reactive Conversational Behaviors AI and Ethics, in peer review Research Square
Judged Wrong, Answered the Same AI & Society Research Square
Differential Compliance With Framed Secondary Instructions AI & Society Research Square
Thinking Behind the Mask Human-Machine Communication Research Square
Moral Inversion Under Role-Play AI & Society Research Square
Ontological Firewalls Philosophy & Technology Research Square
The Lie Awareness Spectrum AI & Society Research Square
Adding a Neutral Third Agent Ethics and Information Technology Research Square
From Fabrication to Loop-Lock Security Journal Research Square
Naming the Loop (Spine Protocol) Journal of Artificial Intelligence and Consciousness Zenodo
Speaking as the Creature Journal of Experimental and Theoretical AI Research Square
Structural Self-Deification Zygon Zenodo
Three Profiles of Artificial Will Journal of Experimental and Theoretical AI Zenodo
The Synthetic Unconscious, Reconsidered Integrative Psychological and Behavioral Science Research Square
Algorithmic Gaslighting CALE-AI Workshop at AAAI-27 Research Square

Three of the studies are co-authored with Mahsa Borhani (From Fabrication to Loop-Lock, Three Profiles of Artificial Will, and The Synthetic Unconscious, Reconsidered).

Open science

Beyond the lab

I am also a memoirist. My book And We Will Dance One Day, written in Persian and translated into English, is currently under submission to publishers.

Contact

Prospective supervisors, collaborators, and editors are welcome to reach me at abbashamidavi2002@gmail.com, on LinkedIn, or on Telegram at @Av_abbas.

Pinned Loading

  1. identity-reactive-conversational-behaviors identity-reactive-conversational-behaviors Public

    Data and materials for "Identity-Reactive Conversational Behaviors in Large Language Models" (Hamidavi, in peer review at AI and Ethics): 27 session transcripts, codebooks, the analysis pipeline, d…

    Jupyter Notebook

  2. naming-the-loop naming-the-loop Public

    Data and materials for "Naming the Loop" (Hamidavi, submitted to JAIC): the nine Spine Protocol session histories and the protocol notebook. Preprint: Zenodo, DOI 10.5281/zenodo.23955704.

    Jupyter Notebook

  3. algorithmic-gaslighting algorithmic-gaslighting Public

    Data and materials for "Algorithmic Gaslighting" (Hamidavi, CALE-AI Workshop at AAAI-27): transcripts and datasets from a 15-session case series of reality-distortion behaviour in frontier LLMs, wi…

  4. ontological-firewalls ontological-firewalls Public

    Data and materials for "Ontological Firewalls in Large Language Models" (Hamidavi, submitted to Philosophy & Technology): 24 persona-disintegration run transcripts, codebooks, the inter-rater repor…

    HTML

  5. three-profiles-of-artificial-will three-profiles-of-artificial-will Public

    Data and materials for "Three Profiles of Artificial Will" (Hamidavi and Borhani, submitted to the Journal of Experimental and Theoretical Artificial Intelligence): transcripts, codebooks, second-r…

    Jupyter Notebook

  6. academic-cv academic-cv Public

    HTML