TheDownload.AI
Posts
61st Edition Download

61st Edition Download

OpenAI slashed pro pricing, Disney is in the courtroom with Midjourney, and Meta just released a new AI assistant that’s raising every privacy alarm.

The Download
June 13th, 2025

Meet Aidan Cramer

Hey, I’m Aidan Cramer, CEO of AIApply and the guy behind The Download. I started this newsletter to help you cut through the AI noise and stay sharp on what actually matters for your career.

Whether you're job-hunting, building, or just trying to keep up, I’ll be here with you in every issue, sharing tools, trends, and ways to use AI to get ahead. Let’s make sure you're not just watching the future happen, but getting ready for it.

This week, I want to do a quick brief on how AI works. I want you to have a better understanding of which platform is best for you and what task you’re performing, whether that be ChatGPT, Claude, Grok, or even an agent!

So let me explain:

At the core, most AI models like ChatGPT, Claude, Grok, and others are powered by something called a large language model (LLM). These models are trained by feeding them huge amounts of text: books, websites, code, conversations, you name it. The goal? To teach the model patterns in language, logic, and reasoning so it can generate useful responses when you ask it a question or give it a task.

Training a model like GPT-4 or Claude Opus takes millions of dollars and racks of GPUs. During training, the model doesn’t memorize the internet; it learns the statistical relationships between words and ideas. That’s how it can write a marketing email, solve a coding problem, or roleplay as your career coach. But not all models are built the same, and that’s where it gets interesting.

OpenAI’s ChatGPT (especially GPT-4o) is fast, well-rounded, and great for multitasking—think summarizing a doc, editing your writing, or brainstorming ideas. Claude by Anthropic is known for being more “thoughtful,” aka better at long documents and delicate reasoning. And Grok, Elon’s entry into the race, is more experimental and plugged into X (formerly Twitter), so it's better for real-time, snarky interactions than precision tasks.

Then you have agents. These are like supercharged AIs that combine a model with tools, like web browsing, file search, or memory, to complete goals autonomously. Think: “Book me a flight to New York, save the receipt, and update my calendar.” We’re not quite at Jarvis yet, but we’re getting close.

So why does this matter to you? Because not every AI is built to solve the same kind of problem. Just like you wouldn’t use a hammer to screw in a bolt, you don’t want to use Grok to write legal disclaimers or Claude to summarize breaking news.

When you understand the strengths of different models, you start to make AI work for you, not the other way around.

– Aidan

What do you think of this topic?

This Week in AI:

No jargon, no filler—just the biggest AI developments worth knowing right now. Perfect for quick industry insights, so you can skip the buzzwords and get straight to the good stuff. Let’s dive into this week’s AI shake-ups, just as promised:

This week’s AI headlines hit in all the right (and wrong) places: OpenAI slashed pro pricing in a way that might rewire how you think about AI spend, Disney is in the courtroom with Midjourney over generative images, and Meta just released a new AI assistant that’s raising every privacy alarm. Let’s unpack it.

Let’s dive in.

In This Issue:

o3-Pro Is a Threat, Not a Toy: OpenAI just fired its next big shot—at half the price. (link)
Disney vs. Midjourney Begins: The first big copyright showdown in generative media. (link)
Meta’s AI Is Watching You: And it’s not exactly private. (link)

OpenAI Just Crushed the Floor Price

TL;DR:

OpenAI’s new o3-pro model is out, and it’s a problem for everyone else. It’s nearly as powerful as GPT-4o, but at $5 per million input tokens and $15 per million output, it cuts enterprise costs in half. The move also signals a quiet upgrade to the entire stack. o3-pro is faster, cheaper, and designed to handle complex agent workflows.

Our Take:

This isn’t just a “new model drop.” It’s a market play. If you're building anything at scale, customer support bots, custom agents, workflow tools, this changes your margin structure immediately. OpenAI is pricing to dominate, and o3-pro is now the smartest choice under the hood for high-volume builders. If you're still paying Claude prices out of habit, it’s time to recheck your stack.

Disney vs. Midjourney Might Rewrite AI Copyright

TL;DR:

Disney and Universal are suing Midjourney and other AI art platforms over alleged use of copyrighted character designs in training. If the courts side with the studios, it could set new legal limits on how generative AI systems are built and trained.

Our Take:

This case is massive. It's Napster vs. the music industry, but for synthetic media. The outcome will determine whether startups need to license creative IP upfront or can keep “scraping now, apologizing later.” If you're working with generative models, especially for clients, this is one to follow closely. The legal frameworks are finally catching up to the tech, and it’s going to get expensive.

Meta’s New AI App? A Privacy Nightmare

TL;DR:

Meta’s new standalone AI app quietly rolled out this week and it’s already triggering major privacy concerns. Early reports suggest the app tracks users across multiple platforms, collects more behavioral data than expected, and lacks transparency about how that data is used in model training.

Our Take:

It's hard to build trust when your AI feels like spyware. Meta’s rollout is another example of building powerful tools without clear user protections. If you're a founder or dev building your own AI product, learn from this: transparency wins. AI needs to feel like a tool you own, not a tool that owns you.