A question about AI chatbots

[scout]

This popped up in my feed.

Very funny take on what it will be like when AI chatbots start including ads — something Anthropic promises not to do, I guess? Along with some self-aware poking fun at AI chatbot peculiarities — the pause before responding, the sycophancy — which I assume Anthropic’s Claude exhibits as badly as the rest.

(It’s a playlist of four Anthropic ads. The first, about the business plan, is the best. The other three are not quite as good, all four make the same point.)

Because Anthropic is clearly poking fun at themselves along with the rest regarding chatbot behavior, I was less sure about the deliberateness of the other thing I noticed — the trivial nature of what people were asking the chatbot to do. How can I improve my abs? How can I communicate with my mom? Is my class project good enough? And the advice was vacuous. Even the one in the best ad, writing a business plan, is pretty mundane — there are thousands of ten dollar books that can help you with that, and any one of them will give better advice than the chatbot in the ad.

Is that all that chatbots can do for me? I ask as someone who has no experience with them and only a mild, abstract curiosity. I’m mostly past the point of seeking out life coaching in any form, I’m fully occupied doing the few things I know how to do and no longer sigh longingly about those I can’t. But I was still surprised. I would have figured that these ads would have cherry-picked the best possible scenarios for chatbot use. Instead I had flashbacks to the early 1980s when ads were telling me how great a personal computer would be for balancing my checkbook, managing recipes, keeping a calendar. All things I needed to do, but also tasks I had perfectly suitable tools to handle, maybe even superior (I’m talking about pencil and paper).

I’m not ready to dismiss these chatbot examples out of hand. True, I already know how to write a business plan, improve my abs, talk to my mom, assess a class project — but some people don’t, and perhaps this is a good way for them to learn, or at least the way learning will happen now. I was skeptical of online map services when they first showed up. Printing the directions was a headache, following turn-by-turn instructions worked until you missed a turn and you were at once totally lost, I was already pretty good at reading a map. But smart phones and GPS solved that problem, and even though it took me many years to get over my resistance I did, and Google Maps has made my life better. (Though I still study highway maps before I travel, and I think my hard-won navigational skills make me a better than average Google Maps user.)

So, tell me if you don’t mind: is my “is that all there is?” reaction mostly an “OK Boomer” thing, a resistance to doing reasonable things in a new and different but equally valid way? Or are AI chatbots the new checkbook balancers? Or did these ads use feeble examples by accident, or to fit into a 60-second tale, or to be funny, or … ?

[scout]

Two pertinent videos, one from Anthropic, one from the Wall Street Journal. I have nothing to add.

Claude ran a business in our office

We Let AI Run a Vending Machine. It Lost All the Money. | WSJ

[LeoNot]

I find chatbots better at processing information than retrieving or fetching. “Build me a business plan” is a retrieval. So is the use of ChatGPT et al as replacements for traditional search. AIs still get it wrong too often, but people don’t notice, as it’s so confidently wrong.

I use AI a lot in my work. Day to day, it’s things like “summarize this article” (followed by my article text), or “write a teaser for this article that will encourage people to click through and read”. I edit the heck out of what it provides. I also use it to generate the article equivalent of thumbnails (aka “hero images”), and if I’m stuck for ideas, it’s “describe 5 hero images appropriate for the following article”. Often, there’s something I can use, or something that sparks an idea.

One of the approaches I should use more is “tell me what this article is missing or how it could be expanded”. The response always begins with “It’s a great article!” or equivalent, but it often then proceeds to list things I hadn’t thought of, which I vet and incorporate some of in my original. AI doesn’t do the writing, it just gives me ideas.

This is a slightly out-of-date article on the topic: How I Use AI at Ask Leo! - Ask Leo!

I think many of the examples used in public-facing marketing (like the ads) aim for concepts and tasks that are easier for the mass market to grasp quickly. Everyone knows what a business plan is; few have created one. It sounds daunting, and wouldn’t it be nice to have someone else do it for you? (Spoiler: no.) The real value, on which many, including myself, have only scratched the surface, is applying it to very specific scenarios common in whatever it is you do.

Tangent: another scenario. I recently had emergency surgery for a neck injury. I was able to access the medical records – mostly in raw form. In other words, messy. I uploaded to an LLM (NotebookLM), and started asking questions against my record. The summaries, analysis, and chronology it then spit out were a) in plain English rather than med-speak (at my request), and b) pretty much spot on and so much easier to digest. It was a fascinating exercise.

Oh, and finally, no, not an “OK Boomer” scenario. I’m 68, deep in the heart of Boomer.

-Leo

[scout]

@LeoNot

The real value, on which many, including myself, have only scratched the surface, is applying it to very specific scenarios common in whatever it is you do.

Thanks, that helped. It sounds parallel to my Google Maps usage, which is for very specific situations — real-time reporting of traffic conditions around my location, last-mile turn-by-turn directions on a trip I’ve never driven before, and so on. The technology may be oversold — I could easily get by without it, did so for most of my life — but some features are new and welcome, and since they’re there I’m glad to make use of them.

[Gwhiz]

Thank you those were a riot!

[danielmclaughlan]

Thanks for sharing. Those were entertaining videos but also depressingly accurate in terms of what it’s like to communicate with a chatbot.. and I say that as someone who has been using ChatGPT frequently for the past year. I’m not sure it’s the best advert for why you’d actually want to use a chatbot.

That said, I have been using ChatGPT and - while my thinking has drastically changed recently - I think they can be very helpful as a sounding board for ideas, working through problems and exploring topics we’re not comfortable discussing openly.

I know not everyone has the luxury of having a real life, living, breathing human they can consult, and our experience on online services can be mixed. Some communities, particularly in tech, can be pretty hostile towards even the most well-researched question. That’s where having a collaborator like a chatbot can be useful.

Allow me to share some examples of how I’ve personally used AI chatbots like ChatGPT (I’m actually now trying to step away from them but can elaborate on that in a follow up):

To design lesson plans

In my previous role I was a full time trainer at a university. While I’m a subject matter expert (digital accessibility), this was my first time in a role where I was solely responsible for the instructional design. I had been studying several learning design courses in an effort to move my learning experiences beyond ‘death by PowerPoint’, but I had no collaborator when it came to the implementation.

It was helpful to be able to work through with ChatGPT how I might apply the learning design theories I wanted to use to my material, as well as ideate on activities that I could adapt. It definitively helped me see how I could apply some of the abstract theory to what I was doing.

To explore fields I’m unfamiliar with

Last year, I rediscovered my interest in animation. I’d been making some demonstration videos to support my training and I wanted to use some motion graphics like title cards and animated icons. I started learning a little bit of After Effects and Premiere Pro, but also, in a wildly misguided late night crisis of confidence, I latched onto the idea that I could retrain as a motion designer.

It was helpful to be able to ask ChatGPT for a list of gateway topics I should focus on to discover further areas for research, and I also asked for help making myself a curriculum.

Needless to say I did not pursue a new career.. but I do still plan to learn motion design to help create educational materials, even though my (still digital accessibility) is more UX focused. Ultimately I found my way back to stop motion as a hobby too so it was still valuable.

To help me plan and build a family history website

This was my heaviest use of ChatGPT towards the second half of last year. I’ve long been an armchair genealogist but, coinciding with my rediscovery of the personal web last year, I’ve grown increasingly concerned with documenting my research and preserving our family history. While I do use the major platforms like Ancestry and Find My Past, these platforms are not infallible and I know only a handful of people in my wider family are interested in exploring those.

I also want to capture more than just facts: I want a way to share our stories and encourage family to share their memories too. I see a lot of my wider family sharing photos and stories on Facebook, seemingly unaware that Facebook could pull the plug tomorrow and it would all be lost.

Basically I’ve embarked on a project that will help me document my research more thoroughly, form closer connections with my wider family, and hopefully preserve our stories for many years to come.

Where ChatGPT was useful was discussing the architecture of said project, finding a web framework with which to build it, and then helping me figure out how to code it. It is a massive project and I can honestly say I don’t think I would have attempted this project without the (saccharine) encouragement of ChatGPT.

However, I did not do what the kids call “vibe coding”: I didn’t describe what I wanted and ask ChatGPT to build it for me. I built it myself by using ChatGPT to ask questions about the web development techniques I was using. At times I got lazy and copied whole sections of code, but for the most part I asked ChatGPT to explain what we were working with and then I wrote it in my own way (that’s my preference for coding and learning overall: I need to understand how things work before I will use them).

So that’s three, genuinely advantageous uses of ChatGPT that I found. However, as I said, I personally want to stop using generative AI. I’m happy to elaborate why if you’re interested, but seeing as I’ve already written quite the tome here I think I better stop while I’m ahead.

To be clear: I’m not an AI-hater and I wouldn’t shame anyone for using AI, but I don’t want it to feature in my life moving forwards, though that’s been easier said than done so far.

[scout]

First, a tangential technical note. This reply is worthy of being a full-blown topic, in my opinion, because it is meaty enough to support a standalone discussion. And topics are easier for users to find at a later date, particularly when this one, which is bumped in the “latest” section due to your reply, scrolls out of sight once again.

Discourse has a way to “reply as new topic” … but I haven’t figured out yet how to do it! I’ve scoured the documentation and other forum discussions, seen it mentioned but not clearly explained. I do know that someone with more privileges than me (maybe a Regular, certainly a Leader) can take a reply and make it a standalone topic. But I think the user writing the reply can at least create it as a topic, if only they know the proper magic. Perhaps someone here more knowledgable about Discourse can enlighten me!

Back to your actual reply — excellent reading! Especially for me, someone who has no opinion about the potential of (and potential dangers lurking within) AI, is curious about how chatbot usage will settle out, even open to becoming a user eventually, but uninterested in adopting them early or even in the midterm. I don’t need a chatbot any more than I needed a smartphone (or so I think, anyway).

I didn’t start carrying a smartphone until 2023, fifteen years after they were introduced, even though I have been an early adopter of many other tech innovations. What made me get one? The convergence of several small tipping points, I suppose. I wasn’t holding out on principle, just lack of interest. But I found myself needing to chat with family members while away from home, not excited about lugging a tablet or laptop to do that. I rode in cars with people whose phones connected automatically via bluetooth when they got in, and thought about the music and podcasts and audiobooks I could have access to while driving. Most of all, I rode in cars with people who used Google Maps to make the experience much smoother. I have a phone now in my pants pocket, and those three functions are the ones I use the most. It’s worth a $150 iPhone SE and a $10/mo US Mobile plan to me to have those functions.

All that to set the stage for this: nothing has been more helpful to me lately regarding chatbot usage than normal people giving straightforward accounts of their attempts to use a chatbot in everyday life. So I’ll be pondering what you wrote. Thanks!

[danielmclaughlan]

I did not know about the “reply as new topic” feature, but I just figured it out:

Start a reply as usual, but then on the left hand side, next to the icon of the person you’re replying to, click on the right-facing arrow. This gives you the option to “reply as linked topic” or “reply to topic”.

In your case, you want the former option (if you still want to continue the conversation, of course). However, that second option is going to be extremely useful to me: i’m a member of another Discourse forum and I often find myself accidentally replying to individuals, rather than the topic as a whole.

I learned two very useful tips today, thanks to your suggestion. Thank you.

[scout]

And thank you for tracking that down! I can’t wait for a chance to try it out. Of course, I have a hard enough time choosing the proper reply button, so who knows whether I’ll actually be able to use this feature.