Autor Cointelegraph By Andrew Fenton

Apple wants to put an AI in your pocket

Apple has been playing its cards close to its chest when it comes to AI. While rival Microsoft has jumped on the ChatGPT bandwagon and is integrating AI into everything despite the bugs and hallucinations, the acronym didn’t even get a mention at Apple’s Worldwide Developers Conference in June.

Reports emerged in July, however, that Apple was working on its own generative AI tool, dubbed internally “Apple GPT,” which uses a large language model (LLM) framework called Ajax. On this week’s quarterly earnings call, CEO Tim Cook said Apple was enthusiastic about the technology and has incorporated AI into forthcoming iOS17 features like Personal Voice (voice cloning and text-to-speech) and Live Voicemail (live transcription). He added:

“We’ve been doing research across a wide range of AI technologies, including generative AI, for years. We’re going to continue investing and innovating and responsibly advancing our products with these technologies, with the goal of enriching people’s lives. That’s what it’s all about for us. As you know, we tend to announce things as they come to market, that’s our M.O., and I’d like to stick to that.”

Of course, what everyday users want to know is whether Siri will be getting an AI upgrade. And they certainly appear to be working on it, with the Financial Times reporting that Apple is hiring dozens of researchers and engineers to work on “compressing existing language models so they can run efficiently on mobile devices, rather than in the cloud.” The ads indicated the company is fully focused on bringing LLM technology to mobiles.

Also read: Experts want to give AI human ‘souls’ so they don’t kill us all

There are speed, privacy and security reasons to run the AI locally on the phone hardware rather than in the cloud, given concerns over OpenAI and Claude hoovering up all your personal and business data. Back in 2020, Apple spent $200 million snapping up Seattle startup Xnor, which focuses on this exact problem.

Apple’s Personal Voice is coming in iOS17. (Apple)

Passwords even more useless due to AI

Even prior to the advent of AI, computing technology had progressed to the point where the average eight-character password — using a combination of numbers, upper and lower case letters and a special character as recommended — could be cracked in around five minutes. New research indicates that AI password crackers like PassGAN can crack more than half of all commonly used passwords in less than a minute.

ChatGPT eats cannibals

ChatGPT hype is starting to wane, with Google searches for “ChatGPT” down 40% from its peak in April, while web traffic to OpenAI’s ChatGPT website has been down almost 10% in the past month. 

This is only to be expected — however GPT-4 users are also reporting the model seems considerably dumber (but faster) than it was previously.One theory is that OpenAI has broken it up into multiple smaller models trained in specific areas that can act in tandem, but not quite at the same level.

But a more intriguing possibility may also be playing a role: AI cannibalism.

The web is now swamped with AI-generated text and images, and this synthetic data gets scraped up as data to train AIs, causing a negative feedback loop. The more AI data a model ingests, the worse the output gets for coherence and quality. It’s a bit like what happens when you make a photocopy of a photocopy, and the image gets progressively worse.

Can you book flights and hotels using AI?

The short answer is… kind of, but none of the AI chatbots are reliable, so you’ll still need to do your own research at this stage.

Having recently spent hours researching flights and accommodation for a three-week trip to Japan, I decided to compare my results to Bard and ChatGPT’s suggestions.

It turns out that Bard is actually surprisingly good at finding flights. A simple request for flights from Melbourne and Tokyo on a particular day returned options with major carriers like Qantas and Japan Airlines, which is probably what many people would be after.

Bard was then able to further refine the results to “cheapest direct flight, with seat selection, a minimum 15 kilograms of luggage and a meal,” finding an Air Asia flight from Melbourne to Osaka that was cheaper than the one I’d booked to Tokyo.

Bard found a very good value flight after the search query was refined.

The AI was also pretty good at determining the seat width, pitch and recline angle for the Air Asia flight to work out if actually flying with the airline was going to be a nightmare.

Overall pretty impressive, though it’s unable to provide a link to book that particular flight. I checked, however, and the prices and details on the site matched.

On the opposite end of the spectrum, ChatGPT was a total fail, despite its new Kayak travel agent plugin. It offered me a 29-hour flight via Atlanta and Detroit, which is about three times as long as a direct flight would take. And while there are plenty of direct flights available, it insisted there were none. As it’s a U.S.-focused site, your mileage may vary.

In terms of hotels, the Kayak plugin won but only by default. Prompted to find an affordable double room in Shibuya with a review score above 7, it suggested the Shinagawa Prince Hotel for $155 a night and provided a direct link to book it. It turned out the hotel was an hour’s walk from Shibuya, and none of the other options were located in Shibuya either.

This was still an order of magnitude better than Bard, which suggested the Hotel Gracery Shibuya at $120 a night. The only problem is that no such hotel exists.

Bing Image Creator was able to generate a nice pic of the fake Hotel Gracery Shibuya.

It then offered the Shibuya Excel Hotel at $100 per night, but the actual cost was $220 a night when I tried to book. After I pointed this out, Bard apologized profusely and again suggested the non-existent Hotel Gracery Shibuya.

Frustrated, I gave up and asked Bard for a transcript of our conversation to help write this column.

Hilariously, Bard provided a totally fictional transcript of our conversation in which the AI successfully booked me into the nonexistent Hotel Gracery Shibuya at $100 a night, with the reservation number 123456789. The hallucinated transcript ended with the fake me being delighted with Bard’s superlative performance:

User: Thank you, Bard, that was very helpful.

Bard: You’re welcome. Is there anything else I can help you with today?

User: No, that’s all. Thanks again.

Bard: You’re welcome. Have a great day.

Clearly, AI assistants are going to revolutionize travel booking, but they’re not there just yet — and neither are their imaginary hotels.

Bard invents a fictional scenario in which I was pleased with its travel booking abilities.

All killer, no filler AI news

— Toyota has unveiled generative AI tools for designers to create new car concepts. Designers can throw up a rough sketch and a few text prompts like “sleek” or “SUV-like” and the AI will transform it into a finished design. 

— Vimeo is introducing AI script generation to its video editing tools. Users simply type in the subject matter, the tone (funny, inspiring etc) and the length, and the AI will churn out a script.