v2 - five ai tools I tried and didn't like

here's your tl;dr:

here’s five AI tools I tried, and why didn’t like them. maybe you will.

I will be roasting these five tools for educational purposes.

Well, also for my own enjoyment. But mostly because I’ve found that negative reviews can be illuminating.

The things that I don’t like about these tools might just be the things you’ve been searching for in a tool for months. Alternatively, you might be inspired to address one of the issues that I had with a tool, and end up creating something even cooler. Maybe you created one of these tools, and now you can adjust your target audience to make sure you’re solving the problem you meant to.

Every tool shouldn’t be for every person. Ideally, we’d develop a diverse set of tools tailored to address specific problems better than any generalist tool ever could - in other words, we’d address real human needs. In reality, it tends to mean that there are a ton of AI tools that don’t seem grounded in a real-world problem in the first place.

I’ll only be including tools I’ve actually used, with a bias towards tools I’ve tried in the last six months or so. I’ve been doing this on Twitter as I try new tools, so you can follow along there if you want to hear my thoughts more than five of them. And you should absolutely send me the tools you’d roast, I’d love to hear about them.

Now, on to the roasts.

1. YourMove.ai (and all the other dating app chatbots)

I never thought I’d see a day where I’d miss 90% of guys opening with “hey” on dating apps, but with YourMove AI, that day has come.

YourMove is probably the most well-known app in the GenAI dating app chatbot space, offering users the ability to generate opening messages for new messages, as well as to analyze the chat history with your match and generate responses. It can even provide feedback on your profile, which is by far the only feature that I can appreciate.

In theory, these tools could help people overcome anxiety around dating apps and/or small talk by assisting them in crafting a message or giving feedback on something they’ve already written. In practice, the messages sound like they were written by someone who took a crash course in pick-up lines and then immediately regurgitated them to their latest match, with the bonus addition of a bizarre number of emojis that may or may not be relevant to the message itself.

Unsurprisingly, the main target customer for these tools is men. You can actually set your gender (Male, Female, or Other - very curious as to what happens with that last one), but in my experience, they all sound the same.

I’ve played with these things a few times, but the vast majority of my interactions have been as a recipient of what is so clearly an AI-generated message. Honestly, it makes me kind of sad when I see it. It feels low-effort, contrived, and completely detached from real interaction. Moreover, it inherently means that the person I’d meet on that first date wouldn’t be the person I was actually talking to. Like, if you need an AI to get to the date, is it even you I’m meeting?

2. Most AI-Generated Text Humanizer Tools

I posted a video about this on YouTube/Nebula before I sent this, so I’ll keep this section shorter - if you’d like an extended dive into why these humanizer tools suck, head over there.

Essentially, these tools (ex. GPTZero) aim to make AI-generated text “sound” more “human.” I don’t know what problem the developers of these tools claim to be addressing, but the primary use is to get around AI-generated text detection tools (ex. Grammarly, TurnItIn), which means the primary user is people who have homework of some kind. I’ve tried them largely for fun, since I don’t have homework anymore.

The problem is that they don’t sound human.

In fairness, ChatGPT et. al. don’t necessarily sound human either. And it makes sense, given that the problem that these tools are ultimately trying to solve is tricking the AI-generated text detectors, which has very little to do with what “human” writing looks like in the first place. But as you’ll see in that video (if you watch it), it’s pretty easy to prompt ChatGPT to generate text that also tricks the detectors and doesn’t sound so bizarre, so why bother using these humanizers at all?

3. Robokiller

RoboKiller is an app that, in theory, uses AI to screen your calls, aiming to block telemarketers and bots.

In my experience, it made my life much harder. I’m not sure how the app uses “AI” to flag spam calls, but let through more spam calls that the default settings on my phone did, and occasionally blocked calls from people I actually knew.

I tried it for about a month before cancelling, and that should have been the end of the story, but it wasn’t. When you use tools that intercept your phone calls, you do have to reconfigure some of the settings related to your cellular provider. So when you cancel them, you have to revert those settings back to what they were before. I followed the instructions to restore those settings upon cancellation, and thought I was all set, until I realized that I wasn’t getting any voicemails. This would have been annoying in any scenario, but it was particularly frustrating to me at the time because I was trying to schedule a few doctors appointments, and needed the voicemails to know who to contact for scheduling.

Searching for troubleshooting tips revealed that I was far from the first person to have this issue. Luckily, others had figured out how to fix it, so I was able to reset the random obscure setting that was blocking my phone calls. Since then, I just enable the “decline unknown callers” setting on my phone. Most calls I receive are from people I know—and I don’t love answering the phone anyway.

4. Sudowrite

If you don’t know, I’ve been writing a novel for the last few years. Earlier this year, I decided to restructure it, but it ended up being a bit of a nightmare to pull all the pieces apart in Scrivner so that I could figure out what made more sense.

Enter Sudowrite. I likely discovered it via an Instagram ad, because Meta is entirely too good at showing me ads for things that I’d totally buy. Sudowrite is an AI writing assistant for fiction writers, aiming to make the writing process faster and more manageable. Its features include AI-powered storyboarding to help you — wait for it —restructure your novel.

The role of AI in creative writing, and whether it should have a role at all, can be a rather polarizing topic. I’m of the opinion that there can and should be space for AI, primarily as a tool to assist people with their writing process, but not as a way of generating text wholesale. The first draft of this newsletter is almost always dictated — it helps me get my ideas out so I can reorganize them into something readable. It’s not unusual for me to dictate novel ideas/excerpts while on a walk either. And there are AI-based tools that I use to get feedback on my writing.

Sudowrite isn’t one of them, because in my experience, it’s bad at giving feedback. It’s also bad at writing. I played around with having it generate paragraphs when I had writers block, revise dialogue that I didn’t feel was quite clicking, and give feedback on what I’d already done. The generated passages lacked creativity or coherence; the dialogue sounded flat, and the feedback might as well have been from Clippy. I have to assume that at the time, the context window was not very large, which may not have been their fault depending on the backend API they were using, but did mean that it tended to lose track of the plot relatively quickly.

Like I said, I’m not opposed to the use of AI tools in creative writing. But I wouldn’t recommend using this one.

5. Character.AI

I know that I said I was doing this in no particular order, but this one is going to be a bummer.

Character AI lets users create virtual characters that interact with people based on customizable traits. The idea is similar to something like Sims, or using a D&D character sheet, but you can make that character available to the public for free or for a fee, and people can interact with it.

And this may sound fun, until you find out how often people use it to impersonate real people. I’ll link the WIRED article on this topic here, but in short, it focuses on people who found out that someone had made a character using their likeness (or that of someone they know). As it turns out, it’s much easier to have fictional characters removed from the website, as the user who created it is likely infringing on someone else’s intellectual property, than it is to have an impersonation of yourself removed. And all the while, people will be able to interact with an impersonation of you, and someone might be profiting off of it.

I tried using it, the website and character creation experience works well, but it doesn’t really matter because I don’t really care about the user experience in the first place. You are able to create a character based on your own likeness and make it available to others, something that some influencers have chosen to do as a monetization route. I don’t know that I’d recommend doing so, but it’s also not really my business. The idea of someone creating an AI version of me without my consent is enough to make me question why this tool exists in the first place.

Honorable Mention: Apple Intelligence Notification Summaries

I’ve been opted into the iOS/MacOS/iPadOS beta updates for several years now, because I enjoy not knowing whether some crucial function of my phone might break after an update is automatically installed while I’m sleeping. And there are legitimately a ton of Sequoia/Apple Intelligence features that I really like - perhaps that will become a video later once it officially releases.

One of these features aims to use AI to condense your notifications into neat summaries. Ideally, you wouldn’t have to dig through all those notifications when you pick up your phone after a few hours.

In practice, you get a set of hilarious and extremely unhelpful summaries. I have no idea what the “AI” summarizing my notifications is actually programmed to do, but so far it excels at getting me to actually go through my notifications, if only to see what the summary could possibly mean. It’s like being on the other end of a game of telephone, except the starting message is the 20 new emails you’ve gotten in the next hour, and the ending message is shorter than a tweet.

I’ve included this as an honorable mention because I have been trying it, and I don’t like it as it relates to the actual thing it is supposed to help me with. At the same time, it’s been fun to try to guess what each summary really means, and to compare the summaries with other friends who are in the same group chat being summarized. It’s a lighthearted AI experience that brings a smile to my face multiple times a day, and I don’t get a lot of those moments with my phone - AI generated or otherwise.

(I did have to turn it off on my emails and a couple other apps where I actually do need to know what the notification said though)

Things I Consumed This Week

It’s been a crazy few weeks, so this will be on the lighter side - I’d love to hear about what you’ve been reading!

• AI companies are trying to build god. Shouldn’t they get our permission first? (Vox / Sigal Samuel)

• Kevin Bacon, Kate McKinnon, and other creatives warn of ‘unjust’ AI threat (The Verge / Wes Davis)

• Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku (Anthropic)

• Google DeepMind leaders share Nobel Prize in chemistry for protein prediction AI (MIT Tech Review / Melissa Heikkilä)

• How Damaging Are AI News Summaries to Publishers? (Hollywood Reporter / Winston Cho)

The Last Loop

if you know, you know