Apple is baking AI features into its upcoming iOS18

- Simplified AI -
Your gateway to discovering the world of Artificial Intelligence!

Happy Cinco De Mayo, amigos! You aren’t living if you aren’t eating chips and guac every once in a while. I need to buy stock in that Avocados from Mexico company that is always advertising during the Super Bowl.

Turns out, I jumped the gun on the margaritas, celebrating a day early. Should've been wielding a lightsaber yesterday instead, right?

A sneak peek of what’s below:

• How much data do you need to train AI?

• iOS 18 is going to bring some new AI usability to iPhones

• Companies are spending how much on AI!?!?

• Guess the real picture - Cinco de Mayo edition!

How much data do you need to train AI models?

Turning AI topics into a walk in the park, not a mind-boggling maze

You're probably familiar with megabytes (MB) and gigabytes (GB). You encounter MBs when checking email attachment sizes and likely have a smartphone with storage ranging from 64 to 256 GBs. It's easy to grasp these sizes since they're part of your daily tech interactions.

If you’re downloading a few movies and TV shows for a trip, it might take up 5-10 GBs on your phone or tablet. But what about larger storage units? Ever wondered how big a terabyte (TB) or even a petabyte (PB) is? And no, we're not talking about peanut butter!

AI models rely on substantial amounts of data for training to function effectively. Some of these models are massive and require extensive datasets. Let's explore how much data is needed for various sizes of AI models:

• Small model with a specific task: Let's say you're training an email filter to weed out spam. For this, you'd likely need a few thousand examples of text, which might take up around 1-5GB of data.

• Medium-sized, more complex model: If your goal is to develop an AI tool for recognizing stop signs and red lights in images, you'll require tens of thousands to hundreds of thousands of examples. This could occupy around 100GB of data.

• Large language models (LLMs): Now, when we talk about training huge models like ChatGPT, we're talking about a whole different scale. ChatGPT 3.5, for example, was trained on approximately 45 terabytes of data. As for ChatGPT 4, it's reported to have been trained on a staggering 1 petabyte.

When training these large language models, they aggregate data from diverse sources like textbooks, research articles, audio, and video, accumulating vast quantities of information. Just to put it in perspective, one terabyte equals 1000 gigabytes, and a petabyte equals 1000 terabytes. If you were to look at 1 Petabyte in front of you, it would be roughly 500 billion pages of printed text. That's a lot of trees.

Fun fact - the total stored information worldwide as of 2020 is estimated to be around 33 zettabytes, which equals 33 trillion gigabytes. That would require one very large phone to fit all of that! I think we need to build a super-sized popsocket for it. Who's up for helping me set a new world record?

AI News Flash

Catch up on the latest AI news from the week!

• Apple is gearing up to unveil new on-device AI features alongside the upcoming iOS 18 release

• Say goodbye to those pesky exit checkers at places like Sam's Club; their days are numbered thanks to AI

• As expected, AI is becoming increasingly entangled in scams and hacks. The US Government is now stepping in with plans to introduce new AI security guidelines for safeguarding critical infrastructure

• Amazon Q is now available for the masses, promising to make developers 80% more productive on Amazon Web Services (AWS), aka the “cloud”

AI Pictures of the Week

I made four of these Cinco de Mayo themed pictures with AI, and one is a real photo. Can you guess which one is real?

Answer at the bottom of the newsletter!

AI Spotlight

Explore how AI is actively shaping our digital future

Big companies are spending BIG on AI

Already worth hundreds of billions of dollars annually, the impact of AI is poised to soar to trillions of dollars per year by 2030. One estimate from PwC shares that we may reach a $15.7 trillion global impact.

Given this projected growth, it's no surprise that major corporations are investing heavily in developing AI models and applications. These initiatives aim to enhance productivity and serve as substantial revenue generators.

• OpenAI, the maker of ChatGPT, has raised over $10 billion in funding to date and is currently seeking an additional $8-10 billion in investments to continue training their new LLMs.

• Microsoft, one of the original backers of OpenAI, has pledged $10 billion to the company. This funding will aid in the integration of ChatGPT4 into Microsoft's suite of products, branded as "Microsoft Copilot".

• Google anticipates allocating $12 billion or more each quarter in 2024 towards the training of Gemini and its integration into their ecosystem.

• Amazon is planning to invest $14 billion or more per quarter to further develop their own AI products and services.

  • They also announced a $4 billion investment into Anthropic, a ChatGPT competitor, marking their largest outside investment ever

Take a step back: Global investment in AI has actually fallen recently - so what's happening here?

According to a report from Stanford's Institute for Human-Centered Artificial Intelligence, from 2022 to 2023, there was a noticeable decrease in AI-related mergers and acquisitions, plummeting from $117 billion to $81 billion. Similarly, private investments in AI experienced a downturn, dropping from $103 billion to $96 billion during the same period.

It seems that the landscape of AI investment is evolving, with established players like ChatGPT and Anthropic solidifying their positions in the AI arena. However, investors may be growing wary of the substantial investment requirements of AI, especially considering the ongoing challenges within the field.

Gone are the days when investors poured funds into companies like WeWork and Uber without seeing any profits. A prime illustration of this shift is Stability AI, a company specializing in AI tools for image and video generation. Despite its impressive technology, the company faced challenges as it boasted a billion-dollar valuation while generating just $11 million in revenue in 2023, while spending a staggering $153 million on operating expenses. For you non-finance folks out there….that's NOT a good ratio.

With the hefty costs involved in creating large AI models, reduced investments might create some hurdles for smaller players looking to join the competition.