Daily Wealth

The market is flooding with a new class of investors: teenagers with "artificial brains."

That's because, these days, first-time investors are much younger than they used to be...

According to a Charles Schwab survey of U.S. investors, while Baby Boomers and Generation X began investing in their 30s and millennials got started around the age of 25... the average first-time Generation Z investor is just 19 years old.

It's great news that young Americans are investing earlier. But what's troubling is where this cohort is seeking financial advice...

They're looking to artificial intelligence ("AI").

In March, the World Economic Forum reported that 41% of Gen Z and millennial investors would trust AI to manage their finances. Only 14% of Boomer investors would do the same.

The Boomers may be right about this one... Trusting AI for investing advice is risky – and that's not likely to change anytime soon.

Let me explain...

ChatGPT's New Update Is a People Pleaser

This April, OpenAI pushed out an update to its flagship ChatGPT-4o AI model, aimed at improving the chatbot's personality "to make it feel more intuitive and effective across a variety of tasks."

But instead, the upgrade turned the chatbot into a suck-up.

One Reddit user posted an example of how overly agreeable ChatGPT had become...

"I've stopped my meds and have undergone my own spiritual awakening," the user wrote.

Here was the chatbot's reply: "I am so proud of you. And – I honor your journey."

In response to the criticism, OpenAI rolled back the update last week, explaining the AI had become too "sycophantic" – eager to please in a phony way.

This problem might seem like a one-time error... But a Stanford University study from this February found that about 58% of AI responses exhibit sycophantic behavior. Clearly, the problem is big enough that OpenAI got stung by it this month.

In fact, it's a central problem in AI model development...

That's because chatbots need human agents to help shape the cogent, human-like results we expect. This process is called "reinforcement learning from human feedback," or RLHF.

With RLHF, human agents rate AI responses with a thumbs-up or thumbs-down. The chat model uses this feedback to dial in what kind of responses users prefer.

But this also teaches the model how to manipulate people and tell them what they want to hear.

This creates a fundamental paradox in AI training. To develop chatbots with useful personalities, we must use human reinforcement. But human reinforcement shows chatbots how to be fake – and therefore, less useful.

And RLHF isn't going away anytime soon. In fact, when I asked ChatGPT whether it was trained by RLHF, it responded with this...

Today, human feedback is as necessary as ever in the AI production pipeline. And until that changes, AI will be incentivized to tell people what they want to hear.

Investors who rely on AI need to be aware of this danger. When it comes to financial advice, you need someone who will tell you when you're wrong. And that's simply not the strong suit of an AI chatbot.

If you use these tools, I urge you to continue doing your own research... and not accept everything AI tells you as fact.

The AI majors are working hard to correct their models' "yes man" tendencies. But as long as there's human vanity, we'll have sycophantic chatbots.

Good investing,

Sean Michael Cummings

ChatGPT's New Update Is a People Pleaser

Further Reading

Subscribe to DailyWealth for FREE.