The Hillsberg Report

Quote of the week

“A jack of all trades is a master of none, but oftentimes better than a master of one.”

- Unknown

More MCP News

You may recall my brief on Model Context Protocol (MCP) from Edition 11. This week brought major news: OpenAI has now integrated MCP into their Agents SDK.

So, what does this mean? MCP is now solidified as the standard mechanism for inter-AI application communication. In practical terms, it will become the go-to protocol for how applications worldwide interact with LLMs and other AI models—and potentially how AI systems communicate with each other. While this could shift some use cases away from legacy REST APIs, they will likely remain relevant in many applications. MCP is likely to unlock an entirely new paradigm of AI-driven solutions.

MCP enables structured interactions between AI models and business tools, software platforms, and content repositories, enhancing their ability to complete tasks. It allows developers to establish two-way connections between data sources and AI-powered applications, such as chatbots.

With MCP, developers can expose data through MCP servers while building MCP clients—such as apps and workflows—that connect to those servers on demand. This shift opens the door to more dynamic, AI-integrated systems across industries.

Anger Towards OpenAI

With the release of image generation into ChatGPT came a flood of mixed feelings. On one hand, you have a fantastic and engaging new feature in the widest-adopted AI product in the world. On the other hand, you introduce some wild capabilities that could be abused.

For example, many of the OpenAI leaders updated their profile pictures on X to an animated style which reflects the stylings of a Japanese animation studio, Studio Ghibli. You can see some of the outrage and reasoning below.

This is interesting to me because it made me think about the following question: Is there a way to protect creators from having their digital image/video content or styles stolen, copied, or repurposed - even if it is accidental or without knowing?

Similar to how teachers used plagiarism detection technologies before the GenAI wave came along, can we develop a similar algorithm to detect plagiarism in output of image/video from these new models?

First, how did those old plagiarism-detection applications work? I'm just going to guess:

• Parse the student's text content
• Chunk the text content into pieces
• Run each piece against a database of text resources
• Run each piece against a Google search for results and use a web crawler to check results
• Use an algorithm with certain weights to determine a probability of plagiarism
• A quick search reveals: Tools like Turnitin and Copyscape use more than just chunking and web searches. They employ deep linguistic analysis and fingerprinting techniques.

Now, how would you do this for these new AI models? Use a very similar process, but with a few tweaks:

• Instead of a teacher using the application to check a student's work, anyone (usually the original content creator) can use the application to check the AI model's work
• Determine the criteria to check against. For text, it's incredibly simple. For image or video, you have many more variables (styles, flow, audio, pixel matching, color schemes, etc.)
• Build multiple algorithms to check each variable and then combine them into a master algorithm to determine the probability of a match. Each country would likely have to set the % likeness to resemble plagiarism
• The user would upload or link to their original content and run the check
• This would ideally be an open-source thing to prevent centralized control and regulation would need to be put in place to enforce it.
• A quick search reveals: Some methods already exist, such as perceptual hashing, reverse image search (like Google’s), and AI-powered style detection.

This is me, a non-engineer, spitballing at 10PM. The main reason this might not work is that AI model providers would be able to use this application in their own algorithms and generation code to ensure output never crosses the threshold. But maybe that's actually the goal! If AI companies implemented this kind of detection, would it be enough to protect creators, or would it just shift the problem elsewhere? What do you think?

Meme of the week

Good News

Nepal has nearly eliminated extreme poverty. In 1995, about 55% of Nepalis lived on less than $2.15 per day. By 2023, that number had dropped dramatically to just 0.37%. A major driver of this progress has been the surge in remittances from large-scale emigration. By 2023, remittances made up roughly a quarter of Nepal’s GDP, fueling economic growth and helping lift many out of poverty.

Checking in on Optimus

Tesla first announced Optimus back in 2021. Since then, they've poured millions into R&D. Elon Musk recently stated that Tesla hopes to produce ~5,000 in 2025. This technology wave feels inevitable, given how many companies are working on similar projects. The real question is: How long will it take for the majority of US households to own a robot?

My official prediction: By 2030, 10% of households will have robots, and by 2032, over half of U.S. households will own a personal chore machine. What a wild world we live in!

Perplexitok

Perplexity AI, one of the new AI companies, is making a hard push towards acquiring Tiktok. You can read about their vision, which was published on March 21, here.

Share with friends

Enjoying The Hillsberg Report? Share it with friends who might find it valuable!

Haven't signed up for the weekly notification?