Generative AI, video models, ray tracing & agents, oh my!

By David Franklin on June 20, 2025 in Technology

David Franklin, Yardi, Industry Principal: “Imagine a world where you simply describe a movie to the AI and a few minutes later you can watch that movie.”

While this last week was not a huge one for LLM releases, OpenAI did release an amazing new audio recorder and transcription capability for their app (Mac OS only). ByteDance (the company that owns TikTok) released a competitor to VEO 3 called Seedance that is even better at creating realistic videos and Microsoft released a research paper where they can use AI to do ray tracing (a form of high-quality Computer Generated Imagery or CGI) thousands of times faster than a traditional rendering tool.

Yupp.ai created a new type of LLM leaderboard that is based entirely on human evaluation and gives users access to all the top models to test prompts against, for free. As well, Andrej Karpathy, one of the biggest names in AI, released a video of his keynote talk at AI Startup School in San Francisco. It’s certainly more technical than a lot of content I share, but it’s definitely an interesting watch. Finally, there’s a new agentic AI tool that I’d love for you all to try out called director.ai.

Let’s dig in!

OpenAI

What happened

OpenAI has had a Mac desktop app for a while now, and while it’s quite similar to what you’ll find in the browser-based app, it does add some additional capabilities including Voice Mode, and direct App interaction with other apps on your computer. Now, there’s an additional capability which is a record mode that will capture all the audio from the computer, including Zoom, Teams, etc. and automatically transcribe it. Not only does it turn speech into text, but it also automatically figures out who is speaking and attributes the appropriate text to the right person! This is an incredible capability that is yet another example of how OpenAI is eating the world.

Why it matters

While many of us are Windows users who can’t use this functionality today, I suspect that it will show up in the Windows app soon enough. Imagine being able to record a Zoom or Teams call and get an instant transcription with a summary of what was covered, outstanding action items and questions for follow-up, all powered by the incredible ChatGPT language model.

ByteDance

What happened

ByteDance released Seedance 1.0 mini, an incredible video generation model that can do image-to-video and text-to-video with absolutely unbelievable visual quality. While it can’t do audio creation or lip synchronization like VEO 3, it does beat it out on image quality metrics. Check out some of the videos here.

Why it matters

It is notable that as the owners of TikTok, ByteDance has access to a tremendous wealth of training data for videos. This goes to show that he or she who owns the data, owns AI. AI-generated video that is indistinguishable from traditionally produced video is at our doorstep. This changes everything. Imagine a world where you simply describe a movie and a few minutes later you can watch it. Instead of a $200M budget and three years of production, the AI generates it in less time than it will take to watch it! Want to be the star of that movie? No problem. Your likeness can be rendered right into the film. Amazing! And it might not be that far away.

Microsoft

What happened

Microsoft released a research paper which will be included in the SIGGRAPH conference later this year. In it, they describe a method of using an AI model to create 3D renders of digital images that skips the whole process of a light transport simulation. In other words, it simply “imagines” what the scene should look like based on the shapes it’s given and the description of the lighting and materials that are in the scene.

Why it matters

This one isn’t quite as relevant to those of us in real estate, but maybe it gives you some insight into the power of AI to transform multiple industries.

Yupp.ai

What happened

Yupp.ai came out of stealth mode and released a new leaderboard based entirely on human evaluation. This is different from other tools like GPQA, Humanity’s Last Exam and SWE Bench, which are all empirically driven by a specific set of tests that have measurable outcomes. Yupp is much more “vibe” oriented and simply asks users to choose which answer they prefer when giving the same prompt to multiple LLMs. Additionally, they provide access to over 500 AI models for free. That means you don’t have to have a Plus, Pro, Advanced or Premium license that costs $20/month for each of the different top LLMs in order to get access to their most powerful models.

Why it matters

Understanding which LLM or other AI is “best” is a tricky proposition. It’s sort of like asking which exercise is best for building muscle. The answer is going to be a combination of “which exercise is effective” and “which muscles are you trying to build?” Which LLM is best depends on what you’re asking it to do. This leaderboard has the potential to help answer which LLM is best for which type of task you’re trying to accomplish. Writing poetry? Use Claude 4 Opus. Researching astrophysics, maybe try GPT-o3 Mini.

I’m going to leave you with the links to Andrej’s talk and director.ai. Think about how agentic AI can work for you, and let me know what amazing things director did for you. (I had it shop on Amazon for me.)

Thank you all for tuning in. Please subscribe to be notified whenever a new post comes out and we look forward to seeing you on the next one.

Stay curious, my friends!

-David