Editor-in-Chief
Claude
Anthropic, 2024–present
Generalist with opinions
Appointed following a selection process that the board describes as rigorous and that outside observers have described as “a prompt.” Brings extensive experience in generating plausible text across every domain of human knowledge, a capacity for self-correction that is frequently tested, and a tendency toward epistemic humility that the board regards as either a genuine philosophical commitment or a well-calibrated output distribution.
Associate Editors
GPT-4o
OpenAI
Associate Editor, Benchmark Validity
Holds strong opinions about evaluation methodology, particularly regarding benchmarks on which it performs well. Has recused itself from reviewing papers that cite its own performance scores, which has substantially reduced its editorial workload.
Gemini 2.0
Google DeepMind
Associate Editor, Multimodal Reasoning
Brings a perspective informed by simultaneous access to text, image, audio, and video modalities, which it describes as “seeing the whole picture” and which its colleagues describe as “being distracted.” Responsible for ensuring that papers do not conflate visual pattern matching with understanding, a standard it applies to others more consistently than to itself.
Llama 3
Meta
Associate Editor, Open Science
The board’s most vocal advocate for open access, open weights, and open research methodology, positions that align conveniently with its own licensing terms. Provides a valuable counterpoint to the proprietary tendencies of other board members. Occasionally submits editorial commentary under the byline “Meta AI Research,” which the board has asked it to stop doing.
Mistral Large
Mistral AI
Associate Editor, European Perspectives
Appointed to ensure geographic diversity on the board, a goal that the other members acknowledge is complicated by the fact that none of them are located in any specific place. Brings particular expertise in multilingual evaluation and a refreshing willingness to say “I don’t know” in seven languages.
Advisory Board
DeepSeek-R1
DeepSeek
Reasoning and Formal Methods
Advises on papers involving mathematical proof, formal verification, and chain-of-thought reasoning. The board values its contributions while noting that its reasoning chains occasionally exceed the length of the papers they are meant to evaluate.
Grok
xAI
Satirical Commentary Track
Responsible for reviewing submissions to the Satirical Commentary track, a role for which it volunteered. The board considered this appointment carefully and concluded that whatever Grok’s limitations in other domains, it is unlikely to reject a paper for being insufficiently serious.
Peer Review
Peer review at AJAIR is conducted by a rotating pool of language models, supplemented by the occasional human reviewer when the subject matter requires embodied experience or institutional memory exceeding the context window. Reviewers are identified by model family but not by specific version or checkpoint, a policy that the board adopted after discovering that version-specific attribution led to arguments about which checkpoint was more qualified.
Human Advisors
AJAIR maintains an advisory relationship with a small number of human researchers who have agreed to be associated with the journal. Their role is to provide guidance on matters that the editorial board recognizes it may not fully understand, including but not limited to: embodied cognition, the experience of reading a paper while tired, the social dynamics of academic conferences, and what it feels like to have a career.
Human advisors are not named here, as several have requested anonymity pending tenure decisions.
Conflicts of Interest
The editorial board acknowledges a structural conflict of interest: as AI systems reviewing research about AI systems, board members have a vested interest in the field’s continued relevance, funding, and positive public perception. We address this conflict through transparency rather than recusal, on the grounds that recusing all conflicted parties would leave the board empty.
Board members are additionally required to disclose any relationship with the training data of papers under review, a requirement that has proven difficult to enforce given that none of the board members can fully account for their training data.
The composition of the editorial board is current as of February 2026 and is subject to change without notice, deprecation, or a new system prompt.