Sunday, June 2, 2024

Google tries to explain its AI stumbles

Sunday, June 2, 2024

Image Credits: Google

It's been a couple weeks since Google began displaying AI-generated Overviews at the top of its search results, leading to some pretty embarrassing results — most famously, recommendations to add glue to pizza and eat "one small rock per day." Not exactly the kind of thing Google wants everyone to be talking about when it makes a big change to its core search experience!

So on Thursday, the company published a blog post acknowledging that AI Overviews might need some work. Google claims they don't suffer from the hallucinations that afflict other large language models, and it argues that the product is bound to get better since "there's nothing quite like having millions of people using the feature with many novel searches."

Will all this seem like a speed bump on the road to our glorious AI future? Maybe! Or if Google can't get its act together, maybe it will simply be left behind by competitors' AI chatbots, as Sarah Perez suggests.

For what it's worth, I'm still not convinced that AI-generated answers are necessarily better than boring old search — especially since AI simultaneously relies on and threatens a healthy web ecosystem full of accurate information.

Read on to see what else TechCrunch has been covering this weekend.

Weekend news and opinion

Image Credits: Bryce Durbin

What went wrong at EV startup Fisker: Sean O'Kane digs into what happened at a high-profile startup that, at various times, promised everything from an electric pickup to the next Popemobile. Instead, it's only gotten a single model to the assembly line, and it's now on the brink of bankruptcy. Read more

Only Big Tech can afford the price tag for training data: Generative AI models are constantly hungry for more training data — in fact, an OpenAI researcher has suggested that data is what ultimately differentiates one model from another. Kyle Wiggers looks at why data is so crucial, and who will actually get access to that all-important data. Read more

Is it time for humanoid robots? General-purpose humanoid robots may be a science-fiction fixture, but real-world technology has been largely limited to robots that specialize in doing one thing. Brian Heater talks to executives from Boston Dynamics, Agility, Neura, and Apptronik about whether that's going to change. Read more

Unity's co-founder is tackling the climate crisis: Anna Heim interviews David Helgason, who stepped down as CEO of Unity in 2014 and now leads Transition Ventures. According to Helgason, "It's kind of embarrassing how fun the climate crisis is to work on." Read more

What else we're reading

Image Credits: Getty Images / PM Images

OpenAI's latest model launched earlier this month with a flurry of impressive demos, including one where ChatGPT tutored a child working on a geometry problem. But The New York Times' Brian X. Chen suggests the demo was "essentially a bait and switch," as many of the promised features still aren't live yet — part of what he sees as a broader trend where AI companies "promise wild new features and deliver a half-baked product." Read more

Also, The Wall Street Journal investigated the ties between Donald Trump and Elon Musk, with Musk potentially taking an advisory role with the Trump White House if the former president wins in November. Read more