713,564 prompts show training did not help use as Alpha School plans 50 campuses
A preprint published on arXiv on 29 August examined 713,564 prompts from about 4,000 employees at one firm over eight months; formal training showed no lasting gain. Alpha School, which hands most teaching to AI software, plans to reach about 50 campuses this autumn and to test the model in public schools in Houston and Springfield. Corporate usage data and school expansion put the promise of teaching with AI to the test at different scales.
Artificial Intelligence··Midday
Senior staff used the tool more skilfully
Four researchers examined 713,564 prompts and the model responses to them from nearly 4,000 back-office employees in 15 functional areas over eight months of 2025. More senior employees used the tool in more sophisticated ways, which the authors read as domain expertise carrying over. Strategy, Digital Innovation and Project Management scored highest. The data show neither a rise in sophistication over the eight months nor a lasting gain after formal training, which the authors take as a sign that habits of use are hard to shift. They offer their measures as something managers can apply and other researchers can reuse. The study is a preprint on arXiv and has not been peer-reviewed, and the authors frame the eight-month window as long enough to test whether scheduled training sessions change everyday prompt patterns rather than only first-week curiosity. The reported findings come from the authors' own runs, and the work is presented as a preprint that has not been peer-reviewed.[1]
Alpha School expands campus and public-school plans
Alpha School, which hands most of the teaching to AI software, plans to reach about 50 campuses this autumn, 27 of them newly announced. It is also starting to test the model in public schools: two in Houston and one in Springfield, Massachusetts. According to Carl Hendrick, the organisation's senior learning scientist, Joe Liemandt, who runs the school, wants to reach a billion children. Alpha has published measurements based on the Northwest Evaluation Association's MAP framework and reports strong maths and reading results; researchers warn those results may reflect the student profile rather than the programme's effect. A 404 Media investigation published this year pointed to faulty lessons in the model, and a former employee said students were being treated like test subjects. The expansion plan arrives while both outcome data and classroom practice remain contested in public reporting. The reported findings come from the authors' own runs, and the work is presented as a preprint that has not been peer-reviewed.[2]
Corporate habit and school scale ask the same question with different data
The preprint, drawing on a large prompt sample at one firm, shows formal training did not durably raise usage skill; seniority gaps suggest domain knowledge carried over instead. On the Alpha School side expansion plans say the AI teaching model will move into more campuses and public schools; MAP-based measures report strong results, yet researchers warn the effect may track student profile. The two reports test the same promise — teaching or embedding AI — in different institutions: one through eight months of back-office prompts across 15 functional areas, the other through a growing school network that now targets about 50 campuses. Hendrick describes a billion-child ambition for the school model while the corporate study finds no lasting lift from formal training after 713,564 prompts. For readers the shared frame is that evidence remains contested on how much training alone changes use as tools spread. The reported findings come from the authors' own runs, and the work is presented as a preprint that has not been peer-reviewed.[1], [2]
