Claude solved my childcare puzzle

The trouble with distilling 2.5 weeks of my scattered thoughts on AI is that things change so fast that the relatable, interesting, ‘newsworthy’ stuff that I thought would be genuinely interesting to talk about feels so blasé a few weeks later. 

The good news is that means you are only getting the best of the best in this newsletter. So… enjoy! And if you like what you read, forward this to a friend (or let me know in a reply to this email!)

My husband and I are in the process of figuring out our child care situation for when we're both fully back at work. We're thankful to have two grandmothers that are very excited to help. However, this does mean we have to balance four different calendars and trying to equitably assign caregiving times to them within the constraints of their own capacity (and ours). It is a lot to think through.

Instead of trying to figure it out on my own with my Google Calendar, I decided to talk to Claude. I shared with Claude all the constraints and considerations and asked it to help me optimize the calendar according to them, and it found a solution. The thing that really wowed me was that it visualized it for me so I could clearly see it at a glance and then refine from there.

These “Visuals” within Claude chats launched a couple of weeks ago. At first I thought “neat” but didn't immediately think of a way to leverage them. I love how often Claude surprises me with a solution I would not have thought to ask for! Once the schedule was made, I also had Claude turn it into an .ics file so I could upload it directly to my google calendar.

If you are among the lucky few using Claude for work, and you also happen to be working with Google Drive, my favorite discovery recently is just how good the Claude Google Drive connector is at finding stuff. I had a situation recently where I was trying very hard to find a file from early 2022. The Gemini AI in Google Drive couldn’t find it. All the Drive filters in the world couldn’t either. 

So, as a final “Hail Mary”, I opened Claude, turned on its “Research” mode, and made sure “Drive” was toggled on as a connector. I told it everything I could remember about the document (roughly when it was made, who I think owned it), and more critically the detail within it I was trying to find– a very particular performance metric for a very particular campaign. 

And within a minute or two, BOOM! It linked me straight to the document and even told me what the metric was within the doc!

“AI paralysis.” A real term? Not really. A real thing? Absolutely. This is what my husband and I jokingly call the sensation of what you can do SO MUCH with AI at once that you ultimately end up getting very little done. Or at least, progressing very little forward. Think of it as “analysis paralysis” but multiplied, and way sneakier.

Because if you can crank out an insanely in-depth research project, turn that research project into a well-organized 20-page document, and then convert all of that into a 10-slide deck all in an hour in the background while you are eating lunch, you’ll probably feel very productive.

But if you don’t ultimately do anything meaningful with all that output, because there’s just too much to look at and go “Wow! How interesting”… you’re not really better off. I don’t have a solution for this yet. But the first step to solving a problem is always “Awareness,” right?

Canva has been a go-to tool of mine since 2015. And while it doesn’t get all the press of ChatGPT/Gemini/Claude, etc. it’s a tool worth knowing about for its AI capabilities. Especially this latest release.

Canva recently released “Magic Layers,” a feature that promises to take any AI generated image and separate it out into, you guessed it, design layers. That you can edit. Now this is a big deal because generating images with ChatGPT or Google Gemini (as for a quick digital invitation, or a fun party menu– just some fun examples I’ve seen) is exceptionally fun and easy (no design skills required, just a creative vision), BUT usually there’s some spacing that looks off, or some funky wording, or a mispelling, or a font you just don’t like.

So I was thrilled when I tested it out and… it worked! Unfortunately, only paid users can try this. But if you’re looking for something that can solve your AI image problem: This is the one to use!

What happened: In late March, OpenAI shut down Sora, an AI Video model and an app that allowed its users to, for free, make truly wild multi-shot AI 8 second video clips with sound. But now, the app and the API access for the model is… kaput. If you are thinking “I didn’t even know this existed”, check out the Sora Instagram to get a glimpse at a sample of all this tool can do. (It’s also an education in AI video generally. Remember: The realism of it will only go up from here. )

My take: Sora was genuinely impressive when it launched. I’ll even admit I had fun playing with the app, at first. But at least for me…there’s only so much AI slop I can tolerate. And the app allowed anyone to make short video clips very easily, so the things that were genuinely clever or funny got totally usurped by everything else. And almost immediately after launch, all those 10-second videos started migrating into all of my other social feeds. It was just so easy, and so inexpensive (FREE) that anyone wanting to rapidly grow a following to make a quick buck could.

Unfortunately, now that it’s gone we don’t necessarily get to look forward to less “AI slop.” The barrier to making it will just be monetary again. What we CAN hopefully look forward to is a more focused effort from OpenAI to improve the ChatGPT experience. Sora was a big distraction, and lately…Claude has been eating ChatGPT’s lunch. 

While I did not encounter any glaring hallucinations this past week, it does bring to mind one that I remember very distinctly from last summer. I was very pregnant and on vacation at a seaside town, and I pulled up one of my favorite pre-dinner AI tricks: screenshot the menu, upload it to ChatGPT, ask it to explain every dish (that ‘fancy food’ lingo leaves me scratching my head way too often!)  and give me recommendations based on whatever constraints I'm dealing with at the moment.

At the time, I was managing blood sugar and needed to balance my macros more carefully than usual. (Pregnancy is fun.) Ultimately, it recommended high-protein swordfish. Eek.

You may be wondering, "Where’s the hallucination?"  Well, Swordfish is high in mercury. And high-mercury fish and pregnancy is a bad combo. My ChatGPT knew that I was pregnant. Despite this, it focused its answer more towards my dietary needs rather than that reality.

This is a good example of why expertise is still needed when engaging with an AI for important topics. Thankfully, I knew this tidbit about swordfish. I imagine many don’t. When I pushed back on it and said, "Shouldn’t I NOT eat this because of my pregnancy?”, it apologized profusely and said I was right. But at this point, for most people, the damage would have already been done.

So this is a reminder to you that AI can be a great help with many things, but for important stuff, it's always best to consult with a professional too. And remember to always be on the lookout for any AI hallucinations or any “advice” that may have been given without the appropriate context considered.

My husband Neil is a Software Engineer who writes about what he's actually trying with AI over on his free Substack. His latest covers three things worth knowing: a new way to supervise your AI agent from your phone, a plugin that makes your AI smarter every time you use it, and a research paper that hints at where all of this is heading. Check it out here!

Thank you for reading! Please feel free to reply to this email with any feedback on issue #2—what resonated, what didn’t, and what else you’d like my take on. And if you are looking for someone to teach you or others about AI, check out www.smartworkai.com .

Until next time,
Celia