AI Bytes Newsletter Issue #35

Welcome to AI Bytes #35! This week, we’re covering the latest advancements in AI, including Apple’s bold move with the iPhone 16 and its AI integration, along with ethical concerns surrounding data transparency, such as Grok AI's EU ban. We’ll also highlight the newest features from Claude 3.5 Sonnet and share expert tips on optimizing OpenAI's API calls. Plus, don’t miss important insights from MIT on data transparency in AI and YouTube’s innovative AI detection tools. Time to catch up on the tech that’s doing all the heavy lifting (so you don’t have to)!

The Latest in AI

A Look into the Heart of AI

Featured Innovation
Apple’s iPhone 16: A Leap into AI, But at What Creative Cost?

As much grief as Rico has given Apple about not innovating and producing a new iPhone with more than just an upgraded camera and slightly faster processor, here comes Apple with A-iPhone! Don’t worry, they aren’t calling it that.

Apple has just unveiled its highly anticipated iPhone 16, marking a bold step into the future of smartphone technology with deep AI integration. Dubbed Apple Intelligence, the new AI software promises to revolutionize how users interact with their devices, making Siri more personal, contextually aware, and efficient. With the powerful A18 chip at its core, the iPhone 16 supports advanced AI features like object recognition and even auto-suggesting photo setups, streamlining everyday tasks and making the user experience smoother. This innovation, however, comes with a hefty price tag, with the base model starting at $799 and the Pro Max at $1,199.

Yet, while these features enhance convenience, they also raise concerns about how much AI involvement is too much. For some, the idea of AI guiding every step, such as suggesting photo setups, might stifle creativity by reducing the user’s role in the artistic process. By automating choices like framing or lighting, the AI could discourage experimentation, limiting the unique, personal touch that users bring to their creative endeavors. Over time, reliance on such AI features might lead to a homogenization of content, with many photos and videos adhering to the same AI-prescribed formulas rather than showcasing individual creativity. This, coupled with privacy concerns over the vast amounts of personal data AI processes, adds a layer of complexity to Apple’s bold leap into AI-driven devices.

Ultimately, Apple’s move reflects a broader trend as tech companies and other industries race to integrate AI into smartphones (and everything else), transforming them from mere communication tools into powerful AI companions. While Apple Intelligence offers significant benefits in terms of convenience and efficiency, its potential impact on creativity and privacy serves as a reminder that balancing innovation with personal agency will be essential in this AI-powered future. Does this move you closer to upgrading to the newest Apple Intelligence device?

Ethical Considerations & Real-World Impact 
The Ethical Quandary of AI and Data: Grok AI's EU Ban Highlights Larger Concerns

In a recent landmark decision, X (formerly Twitter) has halted its Grok AI from processing data from EU users after a court action led by Ireland’s Data Protection Commissioner (DPC). Grok AI, an initiative heavily promoted by Elon Musk, was training its models using data from X’s European users without their consent. This ruling exposes a growing tension between technological innovation and individual privacy, as platforms increasingly rely on vast amounts of personal data to develop AI systems.

The case brings into focus significant concerns over informed consent. Grok AI was using public tweets to train its models, but users were opted in without permission. This kind of data collection, even in public digital spaces, risks undermining personal privacy and autonomy, as individuals may not be aware their posts are being used to fuel AI development. Such practices challenge the notion of user consent and transparency, raising broader questions about the ethical responsibilities of tech companies when handling personal information.

This decision doesn’t just affect Grok AI—it sets a precedent for how AI development will unfold in Europe and beyond. With the DPC referring key regulatory issues to the European Data Protection Board, a more unified approach to governing AI data usage may emerge. As the debate over how to balance technological advancement with the protection of individual rights continues and intensifies, it’s clear that AI’s future must carefully weigh the costs and benefits of leveraging user data without compromising fundamental privacy. What privacy are we further ready to relinquish in the name of technological ease and benefit to the user? We would love to hear your take on this.

AI Tool of the Week - Claude 3.5 Sonnet

The Toolbox for using AI effectively

We’ve featured Claude on multiple episodes and as Tool of the Week before. Today we want to specifically call out Claude’s newer “Artifacts” interface. Artifacts allow you to generate and share substantial, interactive content in dedicated windows. One great thing about artifacts this is that Claude allows you to quickly iterate by letting you preview your result as an application, document etc.

In addition, you not only can test the results out yourself, you can share with others. So let’s say you had Claude build the bill splitting app in the example

You could choose to link this to a friend, by clicking “Publish” button at the bottom, right hand side of the screen.

Now, it looks like there's a bug! The issue is that the last person's slider can't be moved, and percentages can even go negative, which shouldn't happen. No worries, though—we asked Claude to fix it! Here was our prompt:

We gave Claude a quick prompt, and voilà—just like that, the issue was resolved. With a smart adjustment, the slider now works as expected, and negative percentages are no longer possible!

Further down, in the Mike Favorites section, we link a video that goes deep into Artifacts, we’ve also got a link to Anthropic’s official article about artifacts, see below:

Rico's Roundup

Critical Insights and Curated Content from Rico

Skeptics Corner
The Hidden Costs of Poor Data Transparency in AI

How many times have you heard or read us expressing just how important transparency will be through the development and evolution of AI? Probably more times than any of you wish to count, but here we are, talking transparency again. Thankfully, MIT is on the case!

This week I stumbled upon an interesting project out of MIT that is focused upon addressing the lack of transparency in datasets used to train large language models (LLMs). As it turns out, a lot of these datasets lose track of their origins, leading to confusion over licenses, limitations, and potential bias. When you don't know where your data came from, you might be using it wrong…and that can cause all sorts of problems, like training models on data not meant for certain tasks, or worse, perpetuating unfair biases in real-world applications, which we have covered before.

The MIT team audited over 1,800 datasets and found that 70% lacked key licensing information, and 50% had errors. In response, they developed a tool called the Data Provenance Explorer to help practitioners trace the origins and terms of datasets. We have been wondering about how much time would pass before someone put some structure around this, and here we are. Without knowing your data, your AI model is basically flying blind—and no one wants to see what happens when that crash lands.

One big problem is that much of the data being used for fine-tuning large models is coming from the Global North, which means that models built for use in different regions may miss important cultural contexts. For example, a dataset created by developers in the U.S. might not fully capture nuances of the Turkish language or culture, leading to poor performance when deployed in a Turkish setting. As more and more of our lives are affected by AI decisions, this kind of blind spot has serious consequences.

Mike and I, time and time again, have expressed the importance of privacy, data management, and above all else, transparency. Transparency is more than just a buzzword—it’s the key to ensuring that AI works fairly and effectively for everyone who uses it or it is used upon. Without proper data provenance, building an AI model is like navigating with a map but no compass—you might get somewhere, but it’s unlikely to be where you want to be. And anyone who has ever been on a land navigation course knows that having your azimuth off by a degree or two can take you a long way off course. Ultimately, knowing your data’s origins is as important as knowing how to use it, and we will see this play out more as legal cases arise and are settled in the ever-changing AI and machine learning landscape ahead.

Must-Read Articles

Mike's Musings

Coding Corner
5 Tips for coding against OpenAI’s APIs

1. Optimize API Calls

When building an app that communicates with ChatGPT, efficiency matters—especially when managing costs and performance. Instead of sending multiple small requests, you can optimize your app by grouping inputs together into a single API call where possible.

Example: If your app needs to handle multiple user questions at once, rather than sending individual API requests, you can batch them:

This reduces the number of API requests, improves performance, and reduces cost, especially if you’re on a token-based pricing plan.

2. Use Token Management

ChatGPT has a token limit, so if you're generating large responses, it's essential to manage the number of tokens used in both your input and output. This ensures your requests are valid and that responses aren’t getting cut off.

Example: If you're building a summarization tool, you can preemptively control token usage by truncating user input:

This prevents hitting the API’s token limit and ensures that the input is processed effectively. Additionally, you can manage response length by setting max_tokens in your API call:

3. Error Handling and Timeouts

APIs are prone to occasional timeouts, failures, or rate limiting. Without robust error handling, your app could provide a poor user experience, such as freezing or displaying incomplete information.

Example: Implementing error handling ensures your app doesn’t break when something goes wrong:

This way, even if the API fails or times out, the app handles the situation gracefully without crashing.

4. Contextual Chaining

For more complex conversations where ChatGPT needs to retain context across multiple interactions, you need to "chain" the inputs and responses by passing previous responses as part of new API requests.

Example: Let’s say you’re building a customer service chatbot. You’ll want the bot to remember past interactions in the conversation:

By appending the previous conversation to each new API call, the assistant can remember the conversation's flow and respond in context, making it more effective.

5. Secure User Input

If your app allows users to enter text (especially when generating content or processing external data), it’s crucial to sanitize this input to prevent malicious content or script injection.

Example: If users are submitting data that will be processed or displayed elsewhere, you need to ensure it’s safe by sanitizing the input:

For an app where ChatGPT generates code, such as a coding assistant, you might also want to validate input to avoid security risks like command injection:

These precautions ensure that your app remains secure and doesn’t inadvertently execute harmful commands or display dangerous data.

Mike’s Favorite
Walkthrough of Claude’s Artifacts

Claude is our tool of the week, and Claude Artifacts are great, but… how can you get started? Rather than creating a whole new tutorial, I’m embedding a great video tutorial, check it out below:

If you’ve got something you think I’d like, hit me up at [email protected] 

Latest Podcast Episode

• Listen on:

  • Apple

  • Spotify 

  • All other major podcast networks

Connect & Share

• Have a unique AI story or innovation? Share with us on X.com or LinkedIn.

• Collaborate with us: Email us.

Stay Updated

• Subscribe on YouTube for more AI Bytes.

• Follow on LinkedIn for insights.

• Catch every podcast episode on streaming platforms.

Thank You!

Thanks to our listeners and followers! Continue to explore AI with us. More at Artificial Antics (antics.tv).

Quote of the week: "Automate the repetitive, hire for the exceptional." – Unknown