Comox Valley AI Meetup #1 @May 6, 2026

Summary

First official Comox Valley AI Meetup organized by BC AI Ecosystem Association, following a successful "meetup zero" a month ago that attracted 65 people
Chris from BC AI Ecosystem Association welcomed attendees, noting the nonprofit has been hosting AI events in BC for 28 months
About half the attendees were first-timers
Attendees came from Courtney, Comox, Denman, Hornby, Cortez, Campbell River, and Squamish
Mayor Bob welcomed everyone and shared background in technology, having worked on MLS.ca in '95, video websites for independent film, and Shaw Cable's internet department

Core Concept: Separating AI Models from Product Design

Steve Jones presented analogy comparing AI models to strawberries and product wrappers to chocolate coating - the unhealthy "chocolate" (manipulative product design) shouldn't make us reject the healthy "strawberries" (AI models)
Explained that frontier AI models (like GPT, Claude, Gemini) are like car engines - they need product wrappers to be usable, and the same model can power very different user experiences
Stressed importance of distinguishing between the engagement layer (product design) and the AI capabilities themselves
Noted it's complicated to unravel these - for example, CBC incorrectly said OpenAI trained "ChatGPT" when they meant the GPT model

SlowSpeak: Custom Voice Assistant

Steve built custom voice assistant because commercial voice agents are too focused on seeming human rather than providing quality answers
Used Blue Ocean Strategy to identify what to eliminate vs. increase:
- Eliminate: Real-time communication, imperfections (stumbles/giggles), persona, first-person language, memory of personal hopes/dreams, sympathy
- Increase: Accuracy, depth, verified sources, intellectual honesty, playback controls
Built in a few days using Claude Code by explaining requirements in plain English
Example: Asked about quantum computing commercial applications - took 5-10 minutes to research and returned 17-minute detailed answer with sources
Available as open source on GitHub

Student-Safe AI System

Proposed alternative to having students use ChatGPT directly, which Steve considers problematic
Key design principles:
- Email-based (not real-time) to slow down interactions
- One question per day limit to prevent dopamine-seeking and cognitive offloading
- Human-in-the-loop: answers go to parent/teacher for review before reaching student
- Encourages expanded curiosity by suggesting related topics
Example response: Student asked where food coloring in Alberta river would end up - system provided answer, linked to glacier loss research, and suggested related river topics to explore
Age-appropriate responses based on child's age set during setup
Also available as open source on GitHub

Key Takeaways from Steve's Talk

Anyone can now build software using agentic coding tools like Claude Code, Replit, Lovable
Frontier model companies provide API access, so developers can build custom experiences around the models
Trade-off: Custom interfaces may be less addictive/manipulative, but lack sophisticated safety systems for detecting crisis situations that commercial products have invested in
Steve emphasized we don't need to accept the "chocolate coating" - we have power to build experiences on our own terms

When to Use Local AI

Choose local AI when data is sensitive and privacy/control are priorities
Local AI gives you control over where data and compute reside - important for legal compliance with data sovereignty laws
If you run AI in your basement, it's your compute and data; if in cloud, you're exposed to laws of that jurisdiction

When to Use Cloud AI

Cloud AI (frontier models) is more powerful than open source models
Frontier models benefit from millions of users providing feedback and reinforcement learning that open source models lack
Choose cloud when data isn't sensitive and you need the power of frontier models

Important Caveats

Running local AI isn't a "panacea" - you must be aware of what you're doing and keep it at arm's length
Be cautious about what packages/software you install even when running locally
For AI to be rolled out safely in BC classrooms, it will likely need to be local AI where no information leaves the classroom, while still gaining 80-90% of AI capabilities

Voice AI and Personalization

Debate about whether human-like AI interactions are harmful or just different
Steve's main motivation was getting better answers rather than avoiding psychology risks - the less human-like interface was a beneficial side effect
Ys offered counterpoint: personalization can be controlled through system settings in foundation models, and real-time voice interaction is valuable for some use cases like brainstorming
Agreement that different applications suit different purposes - real-time voice is "magical" for tasks like navigating phone trees, but may not be appropriate for student tutoring

AI in Education

Concern raised about convincing regulators/school systems to adopt safer AI approaches
Steve noted BC is more open to conversation following Cumberland incident where student's dangerous conversations weren't flagged
Emphasized need to empower students with AI safely rather than banning it entirely - students graduating without AI experience will be disadvantaged
Grassroots advocacy recommended: individual voices at community level can influence change, especially when ideas go viral

Data Sovereignty and Privacy

Canadian organizations traditionally required data stored on Canadian servers only
Recent regulatory changes (Cloud Act) mean if it's an American company, US government has access regardless of physical server location
Many organizations may be in non-compliance with Canadian privacy laws without realizing it
Example: Cohere (Canadian AI company) has data centers mostly in Buffalo, so federal funding is flowing to New York

Practical Development Insights

Colin from Tree.io shared journey from skepticism about agentic coding to finding it genuinely useful in actual work - adoption came quicker than expected
Chris emphasized it's never been a better time to be non-technical: if you can articulate ideas well, you can build software
Natural Pastures representative shared how ChatGPT helps with grant applications and supporting farmers despite not being technical background
Tip: Build your own first version using agentic tools before bringing in developer - this makes requirements much clearer and costs about a third of the price

BC AI Ecosystem launched new Life Sciences chapter with ~20 attendees including physicians, geneticists
Animation Accelerator and Responsible AI Professional Certification programs starting May 22nd
Responsible AI certification teaches UNESCO and IEEE frameworks for ethical AI deployment
RIPEN program recommended for working with talented student developers from Canadian universities
Security Alert: Linux users should update immediately due to new zero-day exploit (Copy Fail) discovered by local Cumberland company
50% of meetup attendees were women, and BC AI Ecosystem membership is 37.5% female

Notes

Transcript