😺 Google fails (again)

Welcome, humans.

We've been testing the ChatGPT Mac app this past week, and while having Chat on our desktop is convenient, we've found the UX somewhat disappointing—it's very slow and prone to freezing randomly.

It's such a drag on speed that we've mostly reverted to using the browser app. Very frustrating when you’re just looking for quick answers.

It feels like being stuck in a long Trader Joe's line with just an XL box of dark chocolate peanut butter cups. Where's the express lane when you need it?!

Here’s what you need to know about AI today:

• Google’s AI search has been spewing wildly incorrect answers.
• Elon Musk’s xAI secured $6B at a $24B valuation in its Series B.
• GPT-4 performed financial statement analyses as accurately as human analysts.
• Meta is planning a paid version of its AI assistant.

On Thursday’s podcast: Pete explains Anthropic’s big research breakthrough that maps the mind of AI models (Apple Podcasts, Spotify, YouTube).

P.S. If you run an AI newsletter, we might be interested in buying you and integrating your audience into our ecosystem! DM us if this sounds like you!

Google’s worst AI blunder yet.

AI Overviews

We jumped the gun on Google’s AI Overviews, but Google shot itself in the foot with that same gun.

Initially, we thought AI Overviews did a decent job providing quick summaries, but we soon realized we were just asking the wrong questions.

After Google I/O, more people received access to AI Overviews and discovered something unsettling: Google’s AI search lies a lot, and sometimes in dangerous ways.

Check out this thread on some of the wildly inaccurate things AI Overviews has been spitting out this past week:

• Doctors recommend smoking 2-3 cigarettes per day during pregnancy (link).
• “Yes, it’s possible to train eight days a week” (link).
• Adding non-toxic glue to pizza so the cheese sticks to the sauce (link).

So, how does Google go from the world's most trusted source for over 20 years to advising people to "eat at least one small rock per day"?

Well, AI Overviews work by scanning all of the internet’s content and then summarizing info from select articles into 2-3 sentences when you ask a question.

But the AI behind Overviews, Gemini, can’t differentiate true from false. So, it sometimes picks an inaccurate source and runs with it.

• The advice to “eat at least one small rock per day” came from a 2021 Onion article.
• And the glue on pizza rec came from a satirical Reddit comment posted 11 years ago by a user named “fucksmith”.
  • Ironically, Google just started paying Reddit $60M/year to access its content archives for training its AI models.

Why it matters: This isn't the first time Google has botched an AI tool launch. When Gemini (previously Bard) debuted, we reported it was riddled with errors. Then came the uproar over Google’s image generator producing racially diverse Nazis and Founding Fathers.

But there’s a difference between fake news on Gemini, which includes a disclaimer stating, “Gemini may display inaccurate info, including about people, so double-check its responses,” and misinformation on Google Search, a service with no disclaimer that billions of people trust for answers every day.

Google will fix this, but expect to see far fewer AI Overviews for most searches while it does (we’ve already noticed this).

In the meantime, many will disable AI Overviews (here’s how), revert back to Perplexity (us), or just wait for ChatGPT’s Search, which is coming… well, we have no clue glue when.

FROM OUR PARTNERS

Discover innovative ways to deploy AI into your business.

Figuring out how to deploy AI in your organization is one of the biggest challenges right now.

If you want innovative ways to get AI running in your business, you need to register for AWS’s webinar ASAP.

The webinar, titled “Fast-track generative AI workloads into production,” will share the latest innovations in AI strategy and implementation that you probably haven’t heard about.

You’ll also hear insights from AWS and Loka about solutions for overcoming the top 5 challenges in delivering genAI.

Get the jump on your competitors and watch AWS’s free webinar on strategies for AI deployment here!

Around the Horn.

• xAI, Elon Musk’s OpenAI competitor, raised $6B at a $24B valuation, becoming one of the most valuable AI companies overnight.
• Apple struck a deal with OpenAI to use its AI in iOS 18; Apple is still considering Google Gemini as an "option."
• Meta is also developing agents and considering a paid version of its AI assistant.
• Researchers have found that LLMs can perform financial statement analyses as well as, if not better than, professional analysts.

Treats To Try.

1. *Learn how to select and onboard the right open-source LLM in this live technical webinar from OctoAI on June 6th, or register to receive a recording!
2. Doly lets you generate 3D product videos on your iPhone (raised $2M).
3. LLM Course by Maxime Labonne has become a hugely popular technical course for developing LLMs and applications.
4. Iki acts as a personal Wikipedia that retrieves any website, PDF, YouTube, or note you've saved.

*This is sponsored content. Advertise in The Neuron here.

Monday Tuesday Meme.

A Cat's Commentary.