On June 10, 2026, a dialogue with Fable, Anthropic's new LLM, addressed its ability to minimize hallucinations in citations. While improvements in verification processes were noted, Fable emphasized the necessity of rigorous URL verification practices. Various levels of hallucination were discussed, leading to an architecture that combines automated and manual verification for accurate citation management.
My recent interaction with Anthropic's new Mythos-class LLM, Fable, revealed its capability to effectively deploy Microsoft Source Manager for reference management, overcoming a significant challenge that previously hindered other LLMs in their automation of Word. Fable's unique approach demonstrated an end-to-end integrated solution, marking a significant advancement in LLM functionality and workflow hyper-efficiency.
Anthropic launched Claude for Education in April 2025. Despite early adoption by five U.S. universities and some activity in the UK and a few other countries (e.g. AU, NL), global HE traction seems limited, with more focus on coding and non-HE educational sectors. National agreements, so far only with Rwanda and India, tend to emphasize sschools or research collaboration over higher education.
The report on Ireland's universities and their adoption of Large Language Models (LLMs) reveals a cautious approach, with three standout institutions: Dublin City University, University College Dublin, and University of Galway, each employing different strategies.
This report analyzes Manus, an autonomous AI agent launched in March 2025 by Butterfly Effect. Despite interest from institutions, searches reveal no campus-wide deployments, yet it has a strong focus on students. Its turbulent ownership remains unresolved at the time the research was concluded (late April 2026).
This report introduces a series on the deployment of institution-wide LLM agreements in higher education across OECD, EU and similar high-performing economies. It highlights the global adoption of OpenAI's ChatGPT Edu, noting significant national agreements in countries like Italy and Estonia, while emphasizing decentralized U.S. adoption at major universities. Future reports will cover additional countries and LLM SaaS systems.
This posting describes how we automated the Zotero citation process in Word documents produced from LLM-generated Markdown. In passing, footnotes were automated also.
The next big bottleneck in AI may not be model size or compute. It may be access to grounded, real-world experience. If millions of embodied robots begin operating in homes, vehicles and care settings, they could generate filtered experience traces that improve LLM-plus systems far beyond what internet text alone can provide. Child-size companion robots may be especially important because they open access to a domain that today’s AI models understand badly: children’s language-in-context and everyday micro-social interaction. But this only works if the architecture is privacy-first: central systems should receive distilled updates, not intimate raw detail from children’s lives, except under tightly governed emergency rules.
The recent announcement of the impending closure of the OER Foundation (based in Otago in New Zealand) has drawn attention again to the sustainability (or not) of major OER initiatives and the longevity of OER policies. Acting on this we decided to do one of our rapid AI-assisted reviews of the current state of the OER domain. The results were challenging. For speed, this review initially started with a series of seven questions put to ChatGPT (version 5.2) - which we have used a lot over the last three years for policy anal
Opening up learning 