ChatTJB Wants You to Stop Trusting LLMs A San Francisco poet is answering your prompts badly, slowly, and earnestly.
A San Francisco poet is answering your prompts badly, slowly, and earnestly.
In the ’60s, the Bay Area almost got a high-speed train from Marin to San Jose. What could happen if we dreamed big again?
A San Francisco poet is answering your prompts badly, slowly, and earnestly.
If you happen to be in the vicinity of 6th and Folsom in San Francisco, keep an eye out for the ChatTJB.org billboard that looks like every other billboard in the city — except for one small difference. The tagline: “The leading chat interface powered by AI.” In small print, a footnote reads, “Average Individual.” (The footnote happens to be totally obscured by a tree, so you might miss that part.) Like ChatGPT, Claude, Grok, and all the other LLMs out there, you can go to the site and punch a prompt or query into the chat box. “What should I make for dinner?” “Generate a picture of my cat.” “Do you think I’m being unreasonable in this text exchange with my mom?”
Unlike LLMs, the answer is coming from a single human being: experimental poet and public speaker Tucker Bryant, who has been answering more than 1,000 queries each day via his phone. The answers aren’t instant — sometimes it takes hours, because he is a human that needs to sleep. (And the Tucker-generated images are pretty loose, MS Paint-level scribbles.)
“You will have a bad experience with ChatTJB,” Bryant, a Stanford grad who spent years working in tech, wrote on his LinkedIn post announcing the project back in May. “That's kind of the point.” His thesis: What many people actually need when they go to these services is to have someone to bounce things off of. And what’s ensued after launch seems to confirm it.
However, the site picked up big-time this week when he posted a video about it on TikTok and Instagram. It’s currently down thanks to all the incoming traffic, and Bryant is figuring out how to accept the help of people who have volunteered to answer prompts.
Bryant deserves to have his thumbs back, but underneath the real human limitation of the app is a compelling argument. “I want you to have such an inconvenient experience with something that looks like AI that the next time an LLM gives you what sounds like an acceptable answer, you remember the time ChatTJB responded to your tax question with a haiku about crows,” he writes.
Last year, an MIT study of 54 participants found that repeated use of large language models (LLMs) carried
Anyone who has witnessed a person engaging with an LLM in public could attest to that. Last month, San Francisco supervisor Matt Dorsey went on a social media tailspin while prompting xAI’s Grok to produce justifications for Israel’s targeting of civilians in Gaza. In another case, California State Senator Scott Wiener’s election campaign recently launched a gut-churning “parody” LLM,
Using these energy-wasting tools to manipulate public opinion and manufacture quotes and other “evidence” should be disqualifying for any elected official. And it’s worth asking who ultimately benefits when people are trained to stop thinking for themselves. Trust your gut if you feel that something is wrong here. Thousands of people have reached out to ChatTJB — an Average Individual — for poetic and not-very-accurate advice because they know it, too. Deep down, we want friction; we want to do the work of living.
Despite his claim of wanting to do it badly for you, Bryant, ever the human being, can’t seem to resist the urge to give good advice. When I asked, on behalf of a friend looking at arts residencies, what advice he’d have for someone struggling with imposter syndrome, Bryant extended empathy and commiserated over the cost and effort involved. But in the end, he wrote, “I think you should do it.”
Soleil Ho is a cultural critic, cookbook writer, and food journalist who has a nasty habit of founding media projects instead of going to therapy: from the feminist literary magazine Quaint to food podcast Racist Sandwich to our dear COYOTE.