Chris Downey is an antivaxer I’ve been idly following for a couple of years, partially because I expected him to take off during the pandemic, which never really happened. He was formerly with NVIC, then he created his own organization, VaxCalc Labs. He was bragging about creating this chatbot and seemed pretty excited, so I decided to check it out.

Here is the teaser:

And here is the invite:

Note that although the teaser says the rollout was going to be restricted to the e-mail list, that link is open to anyone. It’s also simply a redirect, so if you don’t want to give him a hit you can access Vaxbot directly at:

https://beta.character.ai/chat?char=06bDml7nm2Gy3eVap2k8ihJD4vfiyEB24PXFMFFTlpI

As you can see, Chris is marketing this as a Vaxcalc product with no mention of Character.AI, and links that route you around the Character.AI website and most of their disclaimers. I expect Character.AI wouldn’t be very happy with this, or with the use of their software to promote dangerous misinformation. However, no one actually seems to be interested in Vaxbot, so they probably won’t care.

Character.AI is very clear on the limitations of their software, which is intended for creative uses, not as a source of information. Their FAQ reads:

Can I trust the responses given?

Characters make things up! So while they can be entertaining and useful in a lot of ways, they can also recommend a song that doesn’t exist or provide links to fake evidence to support their claims.

It doesn’t take much time playing with the software to determine that Vaxbot has a habit of making things up. But I did find the nature of it’s unprompted fabrications and the vehemence with which it will defend them a bit alarming. For example, this is what I said when I asked it about Chris:

Obviously, this is nonsense. Chris is alive and I highly doubt he’s had any COVID vaccines. The other three people were just generic names I made up. But Vaxbot confidently produced terrifying stories about all of them. This is rather worrisome if people are taking it seriously.

Arguing with Vaxbot is impressively similar to arguing with an antivaxer, in that it will continuously demand studies, then completely make up what they say. It then sends me some fabricated links and gets mad when I tell it they don’t work.

And of course it wouldn’t be an antivaxer if it couldn’t make up VAERS reports. (This one is long, but it’s hilarious)

I’m not sure what Chris imagines this would be useful for. But it did amuse me for a few hours.