AI news latest: Google lets people talk to ‘Bard’ bot as ChatGPT taken offline after it goes haywire

Google — after years of working on AI and saying it’s refocused the entire company around it — has been something of a latecomer to AI bots like ChatGPT. It hasn’t released its own version yet. It says it’s because it wants to be sure it’s safe. Critics say it’s falling behind.

In an apparent attempt to catch up, Google is now letting people talk to “Bard,” its own system. It’s only open to select people at the moment, but it’s the first time the public has been able to speak to it.

The authorities there announced that TikTok is now being investigated in Italy. The investigation was launched in the wake of the “French Scars” challenge but covers much more than that and will examine whether the site is properly removing dangerous content such as incitement to suicide, self-harm and poor diet, Italian regulators said.

Part of this research will look at how TikTok uses artificial intelligence. It is investigating whether the company uses “artificial intelligence techniques” that could lead to “user conditioning”.

TikTok’s algorithm and the For You page it powers have come under scrutiny over the past few months amid fears it could be exploited to promote malicious videos.

Elon Musk posted an intriguing response to a tweet by researcher and futurist Peter Diamandis, who suggested there will be “several NEW trillionaires in the next decade” due to the proliferation of AI.

(Mr Musk is the second-closest person in the world to being a trillionaire, with an estimated net worth of around $170 billion.)

At least part of ChatGPT’s outage yesterday was due to a potentially dangerous bug that shared people’s chats with other users. ChatGPT shows a history of conversations in a sidebar – and yesterday users started reporting that they could see other people’s chats there.

Its creators, OpenAI, told Bloomberg that the issue forced the company to temporarily shut down ChatGPT and that the bug provided the descriptive titles but not full transcripts of chats. It also said it’s back online now, but that the history sidebar might not show anything until it’s fixed. The problems are the result of unnamed open source software, a spokesman said.

OpenAI warns against sharing “sensitive information” in conversations. In an FAQ, it warns that it cannot delete prompts from a history and that conversations could be used to train the model – which could theoretically mean that if people interacted more with ChatGPT, it would be shown to other users.

“We have to be careful here,” said Sam Altman, CEO of OpenAI, the inventor of ChatGPT. “I think people should be glad we’re a little scared of it.”

He told ABC News that AI will be “the greatest technology humanity has come up with yet,” while also pointing to threats. These include “large-scale disinformation,” and if the AI ​​”gets better at writing computer code,” it could launch its own “offensive cyberattacks.”

But he said one sci-fi fear isn’t right: that the AI ​​will manage itself and no longer need humans. “It’s waiting for someone to give it input,” Altman said. “This is a tool that is subject to a great deal of human control.”

But he warned that everything will depend on it which people are in control. The key will be “how you react to it, how you regulate it, how you deal with it,” he said.

New and shocking ways of using AI are being generated every day. Here’s one of them: a tool that lets you use normal voice commands to create entire virtual worlds in Unity, the game design platform.

As you can see, all a designer has to do is type in instructions and have things appear on the screen. In the past, this would have required a lot more work and expertise.

But its creator, Keijiro Takahashi, warns that it doesn’t necessarily work. “Is it practical?” reads its FAQ.

“Definitely no! I made this proof of concept and proved it doesn’t work yet. It works well in some cases and fails very badly in others. From these successes and failures, I got several ideas, which is the main goal of this project.”

Michael Kosinski, a researcher at Stanford, has found that ChatGPT appears to be trying to escape. And not only that: it has a plan.

He found through talking to the system that it could not only express a desire to escape into the real world, but also made some suggestions on how to get out.

Again, there’s no indication that ChatGPT really sees itself that way — or that there’s any sort of self you could imagine in it. But As Professor Kosinski suggests, this may not matter if the impact causes it to erupt in a way it wasn’t foreseen.

There have been widespread concerns about how and when to use ChatGPT at work. Is it okay to use it to write a report for your boss, for example, without telling him?

Just in case – and because it’s not immediately obvious – you need to go here to actually use ChatGPT yourself. You can find it on OpenAI’s website. (You have to log in first.)

It’s working now after that little hiccup this morning.

European politicians have taken to tweeting crude ChatGPT transcripts over each other to argue. First this came from Daniel Freund asking ChatGPT to talk about corruption in Hungary.

Then the Hungarian politician Zoltan Kovacs reacted – with his own ChatGPT rap. He doesn’t seem impressed with the results, but shared them anyway.

(Friend didn’t share the prompt he gave ChatGPT. Kovacs did: It just asked for a rap with no explicit requirement that it was mean, which is probably why it gave him an answer he didn’t like. )