No, this isn’t an article about the ins and outs of experimenting with Ayahuasca. You’ll need to go elsewhere for that. Or maybe you can ask Aaron Rogers.
We did talk about another kind of hallucination on the latest episode of Localogy’s This Week in Local podcast. Our conversation with Evocalize CEO Matthew Marx covered a lot of topics – from raising money in today’s environment to what it takes to make life simple for small business operators.
We also talked about hallucinations (the AI variety) and why they are a challenge for digital marketers.
In AI parlance, a hallucination is when a generative AI tool makes something up. This sounds like something that could be funny. Yet in a business setting, even comical hallucinations can carry a real cost.
It’s a widely publicized barrier for generative AI solutions like ChatGPT to gain full acceptance.
The root of the hallucination problem is something that Christian Ward has taught us during his excellent AI presentations at the last few Localogy conferences.
Large language models are essentially built to predict how a conversation should unfold. And sometimes the next logical statement in a dialogue isn’t actually true.
Ep. 35 Asks if Automation is the Answer with Evocalize CEO Matthew Marx
Growing Pains or Chronic Disease?
It’s one thing if hallucinations are just a sign of AI’s growing pains. Something that eventually goes away. It’s quite another matter if it’s a permanent feature. At this point, it seems the problem is getting better. Whether it goes away completely appears far less certain.
This is what Open AI’s Sam Altman (the AI go-to guy) had to say about the subject at a recent conference in India, according to an AP story earlier this month.
“I think we will get the hallucination problem to a much, much better place,” Altman said. “I think it will take us a year and a half, two years. Something like that. But at that point, we won’t still talk about these. There’s a balance between creativity and perfect accuracy, and the model will need to learn when you want one or the other.”
While this statement offers some hope, it falls short of a commitment to eliminate the problem. To us, a “better place” means the problem is much less severe but still exists.
This House Needs a Fireplace
Evocalize focuses on a few key verticals. One of them is real estate. Our brief conversation on the episode with Matt about hallucinations focused on real estate.
The biggest concern in this context involves hallucinations that inadvertently mislead people. For example, imagine if a hallucination changed the price of a home in an AI-generated listing. It doesn’t require much imagination to understand why that could be a problem.
But even smaller hallucinations can create challenges. For example, adding a fireplace to a home that doesn’t have a fireplace. According to Marx, this has happened.
The topic came up as Marx was describing how using generative AI to produce digital marketing copy was driving an impressive 30% improvement in performance.
Yet he conceded that lingering hallucinations were standing in the way of a fully automated creative process. Marx did note that the hallucination problem was diminishing quickly.
“At the beginning, it was perhaps five or 10%,” Marx said, referring to the share of AI posts that were impacted by hallucinations. “Now I think we’re at least in the single digits, if not below that.”
Ultimately, full automation is the goal, Marx said.
“I think striving for…100% automation is really the answer for these industries,” Marx said on the episode.
While AI seems to be the best way to get to full automation, lingering hallucinations appear to stand in the way of this. After all, even the slightest chance of a hallucination demands the presence of a human editor.
Listen to the Episode
