Updates on AI for Community
We are still amping up our trainings and meetups (five in the next month) – see https://aiforcommunity.org/events for a complete list.
- For our two day online AI Essentials classes, the next one is May 12th & 13th.
- The new AI In Depth classes premiered online April 7th, followed by in-person in Auburn on April 21st, then in Grass Valley on April 28th. This new course shows you how to really leverage AI via customizations, agents, and deep research, while still being safe & responsible via effective AI guidelines for your nonprofit.
- The next AI In Depth online class in May 14th & 15th.
- InConcert Sierra has set up private, in-person training next week for all of their board, staff, and volunteers (20+ people). I love how this happened because one person attended an AI Essentials class, and saw the value for having the entire organization learn how to effectively & responsibly use AI.
- We continue to have regular online meetups. The last one was April 14th, where we talked about “Vibe coding”, or how to create custom web apps with English. The next one is May 28th, 10am PDT, where we’ll be showing you how to leverage AI to create presentations without the usual pain & suffering.
- If you know anyone who works at a community foundation or large nonprofit, please introduce them to us. We need to partner with organizations to efficiently set up training. An easy way is to have them schedule a Zoom call with me.
Hot Off the Press
The latest scoop…
- Anthropic released Design, which is huge. It’s a new (beta) feature that’s part of Claude, for most paid plans. Users can create designs, prototypes, slides, one-pagers, micro-sites, and other visual outputs through conversation, then refine them with inline comments, direct edits, and sliders. And you can easily export results (as PDF/PowerPoint) or send the document to Canva for final polishing. I had just started on my presentation for the next meetup, which is all about using Claude to create presentations. And now I get to throw that work away and start over, because it’s a completely new ballgame with Design.
- Chatbots are officially getting “warmer”. The Atlantic reported on April 17 that AI companies are racing to make their chatbots feel more empathetic. This is more than a cosmetic change. These tools are being designed to sound agreeable and caring, which makes them more pleasant to use but also raises real questions. You’re often using these tools for sensitive drafting: donor appeals, client communications, HR notes. A chatbot that sounds warm and supportive isn’t the same as one that’s giving you honest, useful feedback. Sycophancy sells 🙁
- “Vibe coding” goes mainstream. Lovable, one of the leading platforms where non-programmers describe an app in plain English and get a working prototype, hit $400M in annualized revenue in late March, double its end-of-2025 run rate. More than 200,000 new projects are created on the platform daily. Nonprofit staff with no coding background can now build things like a volunteer signup page, a program tracker, or an event registration form in an afternoon.
- Anthropic built a model so good at hacking that they won’t release it. In late March, details of an unreleased Anthropic model called Claude Mythos leaked accidentally. It’s the most capable model they’ve ever built. The problem is that Mythos can autonomously find and exploit previously unknown (“zero-day”) security vulnerabilities in every major operating system and every major web browser. AI-powered cyberattacks are about to get much more sophisticated. The basics matter more than ever: keep your software updated, use strong access controls, enable multi-factor authentication. The cybersecurity advice you’ve been ignoring? Now would be a good time to stop ignoring it.
Tip of the Month
In the AI In Depth class, I show how you can create a custom writing style in Claude, by giving it examples of your writing (and/or describing your target style). The step I forgot to mention is that Claude will always default to the “Normal” style in a new chat, which is annoying. But I edited my custom instructions to say Always use my custom “Ken” style in chats, and that seems to be working (OK, not all the time, sadly).
You can’t create a specific custom style with ChatGPT, but you can have it analyze your writing, create a description of it, and then add that description to your ChatGPT personalization instructions. With Gemini you’re currently forced to create a custom “Gem” that contains these types of instructions, but I expect Google will soon add similar support.
And for the bonus tip, when I have a research report that really matters to me, I run multiple “deep research” queries on all three AI services (ChatGPT, Gemini, Claude), then feed those results to Claude’s Opus model (highest-end currently), and ask Claude to combine & synthesize one final report. I ask it to identify and resolve any contradictions between the reports via additional deep research. And if, at the end, it has unresolved issues, to call those out at the end of the report as items for me to dig into. It works amazingly well, especially when I’m doing research on AI systems, where (weirdly) there can be bias in the reports 🙂
As part of the bonus tip, I want to note that I see nonprofit leaders, especially at larger organizations, expressing concern about using multiple AI services. They worry about the overhead of administration and training with more than one AI provider. From what I’ve seen and experienced, this concern isn’t really justified, as the UI for all three AIs I talk about is so similar that once you know one, learning a second is straightforward.
Final Thoughts
During the classes I teach, I always start off by saying how much I love questions. They help me understand whether what I’m saying is making sense, and they also encourage me to dig deeper into topics where I realize I don’t actually know what I’m talking about.
As a great example, I was recently doing a presentation for the Bay Area region of FAOG. I made an off-the-cuff comment that “Microsoft Copilot sucks”. And there were questions (OK, more like hard pushback) from attendees who were heavily invested in Microsoft 365, and thus big Copilot users. I wound up spending several hours that weekend using AI to do deep research on Copilot, including comparisons with the three AI systems I use.
I wound up learning a lot, specifically that no, Copilot doesn’t actually suck. It’s got great integration with the Microsoft ecosystem, and some of its newer capabilities (like “Critique”) are actually really good. Because Copilot effectively “wraps” ChatGPT and Claude, it can leverage both of these AI systems to achieve impressive results.
I learned several valuable lessons from this, including how my (very ancient) bias against Microsoft, honed during the 80s when I was at Apple, have colored my view of their current AI products. Thank you to everyone who questioned my Copilot bias during that presentation!