Adabra World Model

Peer-reviewed studies of AI helping people work and learn. None of them tested Adabra or a personal AI clone. Studies were checked in October 2026 against publisher, PMC/Europe PMC, Crossref and arXiv records by an independent AI-assisted audit; some were read in full and some only as abstracts (Noy & Zhang and Dell’Acqua et al. abstract only). This is not a systematic review or a statistical replication.

1. Brynjolfsson, Li & Raymond · Quarterly Journal of Economics · 2025

   An AI assistant that learned from top agents’ conversations helped less-experienced agents most

   Task Customer-support chats at one software firm
   Compared with Agents before, or without, the assistant
   Sample 5,172 agents
   Scope The AI suggested; people decided. About 15% more issues resolved per hour on average; top agents gained little.
   

2. Noy & Zhang · Science · 2023

   40% less time and 18% higher rated quality on professional writing tasks

   Task Bounded, occupation-specific writing tasks
   Compared with The same tasks without ChatGPT
   Sample 453 college-educated professionals
   Scope Short writing tasks, not a whole working week.
   

3. Dell’Acqua et al. · Organization Science · 2026

   25.1% faster on tasks inside AI’s capability frontier, but less often correct on a task outside it

   Task Realistic consulting tasks
   Compared with Consultants working without AI
   Sample 758 consultants at one firm
   Scope AI help depends on the task: outside what it does well, accuracy worsened.
   

4. Kestin et al. · Scientific Reports · 2025

   Students learned more with a carefully designed AI tutor than in an active-learning class, typically in less time

   Task Two physics lessons in one university course, tested straight after
   Compared with The same lessons taught as in-class active learning
   Sample 194 eligible students
   Scope Short-term learning in one course; class time was assumed, not measured.
   

5. Létourneau et al. · npj Science of Learning · 2025

   AI tutoring systems were generally positive against teacher-led instruction, and weaker against other digital tutors

   Task K-12 school learning with intelligent tutoring systems
   Compared with Teacher-led instruction, or non-intelligent digital tutors
   Sample 28 studies, 4,597 learners
   Scope Mostly pre-LLM systems; no meta-analysis. Not personal clones.
   

6. Bastani et al. · PNAS · 2025

   Unrestricted ChatGPT practice lowered later exam scores without AI; a tutor with teacher-set guardrails avoided the drop

   Task High-school maths practice, then exams without AI
   Compared with Practice with textbooks only
   Sample About 1,000 students in one school
   Scope Design is essential: unguarded help can lift practice scores while harming learning.
   

Also referenced

• Gavoille, Subtil & Nourredine · JMIR Medical Education · 2026 · Voice-clone lecture pilot, 50 students analysed in one centre; superiority trial, not an equivalence result; two printed confidence intervals are inconsistent with their estimates. A limited component pilot, not validation.
• Park et al. · arXiv preprint 2411.10109 (v3, June 2026) · Emerging research, not peer reviewed: AI agents built from interviews predicted 1,052 people’s survey answers relative to their own two-week test–retest consistency. Simulates survey responses, not expertise.
• Coleman et al. · JMIR Diabetes · 2025 · Feasibility trial, 43 patients: a clinician-scripted avatar gave higher knowledge scores but lower trust, satisfaction and empathy than nurse education.
• Clay Ford · reanalysis of Kestin et al. public data (blog, 31 Jan 2026) · Not peer reviewed. Finds a real but smaller AI-tutor advantage (about 10 to 30% higher median scores); not reproduced by our audit.

About the swirling flow at the flame in “Time is the fire”

The swirling flow drawn at the flame follows the explicit leading-order core of the construction in OpenAI, “Finite time blowup for Navier–Stokes” (preprint PDF, September 2026; sections 2-4 and Proposition B.2). Tracer paths come from the paper’s formulas at illustrative parameter values. They are shown for a finite time that ends before the singular time. It is not the full construction, which adds higher-order corrections, oscillatory pulses and a smooth external force. It is not a simulation of fire, and it is not a check of the theorem. We have not verified the proof, and the PDF carries no peer-review information. OpenAI · Finite time blowup for Navier–Stokes (preprint PDF, September 2026)