Dr. Philippa Hardman is spot on with this analysis. I suspect we are all moving very quickly from generic AI (simple included models in tools like Google Workspace and Microsoft 365, the Gemini and CoPilot models) to building our own models internally.

Her analysis via LinkedIn post:

The AI Illusion in Corporate L&D: how & why adoption has not led to impact (and how we might change this in 2025).

This week, I addressed CLOs and other L&D professionals at a Fortune 500. My task: to summarise the story of AI & L&D so far, and explore what we need to do next to optimise its impact on the efficiency and effectiveness of L&D.

Here’s the TLDR:

👉 In 2024, the adoption of AI in L&D soared from 5% (2023) to 84%.

👉 In 2024, thanks mainly to the introduction of enterprise-grade AI tools, the adoption of AI in L&D went from “closeted” to “permitted”, and from grassroots to centralised.

👉 Many assume centralised implementation & widespread adoption = impact, but in the vast majority of orgs AI has not lead to any measurable impact on speed & productivity.

👉 The two biggest blockers to impact:

1. Permission without Direction: Orgs have given access to AI without enough strategic direction and support.
2. The Generic AI Problem: L&D is highly specialised, but we’re using generic AI tools which are sub-optimal for our needs and often detrimental to our work.

👉 Recommendations:

1. Learn from the methods and findings of a small number of “bleeding edge” L&D orgs which have taken a strategic, R&D approach to implementing generic AI.
2. Focus less on trying to retrofit generic AI with prompt engineering and prompt libraries and start co-creating specialised AI tools for L&D.

Read the full analysis and access the case studies using the link

I have already had some friends in large schools ask, “How do we build our own?” The roadmap for AI LLMs under school district control and trained on district data that is “safe” from prying Big Tech snoops has its own appeal.

Will schools skip using generic AI tools and go straight for the jugular, running their own AI on District servers and avoid all the potential third-party AI amd data vendor issues? Of course, they will want to on-premise everything.

Too bad cybersecurity is at an all-time low in K-12 with data breaches happening daily. Just check out the K12 Security Information eXchange for details.