The Inception: The Advanced Statistics Class
It all began three years ago in an advanced statistics class. My teacher, someone I still remember vividly, made a passing comment that struck me deeply: “The star rating systems for products don’t really work unless you know the context of the reviews. Statistically, it’s flawed.” That one remark sparked a cascade of questions in my mind. Why aren’t these systems better? For movie rating systems, the flaws seemed even more glaring—a single score or star rating couldn’t account for the complexity of storytelling, character depth, or cinematic artistry. This realization shaped my vision for LM6 as a tool to analyze these nuances and provide a more holistic and fair evaluation. Why do we rely on numbers that fail to capture what makes a movie good or bad? That was the beginning of LM6.
The Idea Takes Shape: Movie Rating with AI
I had a vision: to create something that could genuinely appreciate art, not just for movies but also for books and music. Movies, however, stood out as the medium where current systems seemed the most broken. Unlike books or music, where personal tastes can guide selection and enjoyment, movies often involve a more complex blend of visual storytelling, acting, and production design. For example, a film with stunning cinematography might receive poor ratings simply because its plot was unconventional, while a movie with a predictable story but star power gets higher marks. These disparities highlighted the need for a system like LM6 to assess movies on multiple dimensions and capture their true artistic value. I started small, creating an Instagram account to share my appreciation for art. Each month, I’d post about the top five movies, one book, and five songs. This little project grew steadily, and by the end of the first year, I realized I needed to include movie clips. That addition changed everything.
The Naming of LM6
For a year, I didn’t know what exactly to do with Lowerated or LM6. Then one day, I came across a random Instagram quote that said, “You have to suck at first, but you have to start with something to be good at anything.” Another reel suggested copying ideas to grow initial followers. Inspired, I started posting consistently—finding movie clips, writing detailed paragraphs, and sharing them daily. Sometimes I posted one, two, three, or even four times a day. The content gained millions of views, but at that time, I wasn’t branding anything about LM6 or Lowerated.
During this phase, I reconnected with a university fellow, Fakhir Ali, who was working on an open-source project called OpenVoiceChat. His idea was to create an open-source solution for voice agents that could replace call centers with just one file of code. He accomplished this well before other hyped companies attempted similar solutions. This inspired me to make LM6 open-source. I developed a Python library for it and wrote the backend code. To analyze movie reviews, we needed a model to extract and classify information.
To achieve this, we wrote code to scrape reviews of the top IMDb movies. Using a $10 API token for GPT and GPT-4-mini, we extracted chunks of reviews like, 'The movie was good, but the cinematography was trash.' The model would then map these to seven aspects:
Cinematography: 'cinematography was trash'
Direction: (empty)
Story: (empty)
Unique Concept: (empty)
Production Design: (empty)
Characters: (empty)
Emotions: (empty)
Another model, DeBERTa, labeled each aspect as positive or negative. For instance:
Cinematography: 'cinematography was trash' (negative: 90%)
Direction: (empty)
Story: (empty)
Unique Concept: (empty)
Production Design: (empty)
Characters: (empty)
Emotions: (empty)
I also hired interns to clean and refine this dataset. Later, Hasanat Lodhi provided invaluable assistance in training the models and finalizing the system. I wanted to turn my growing idea into a project. At that time, Lowerated was just a concept—a movie rating algorithm. LM6, which stands for the sixth module of Lowerated, emerged from a simple spreadsheet where I was planning the different modules of the system. Each module had a specific purpose—some focused on data collection, others on audience engagement or content curation. LM6, being the sixth, was envisioned as the core analytical tool to deliver fair and insightful movie ratings, tying all the other components together. Ironically, the fifth module isn’t even developed yet! But LM6 took on a life of its own.
When I pitched the idea to my supervisor as a final year project, they dismissed it, saying, “It doesn’t have any business application.” Reluctantly, I moved on to something else, deciding to work on a more traditional project that focused on AI Agents for business applications. While it wasn’t my passion, the experience honed my technical skills and gave me a new perspective on how to structure complex systems—lessons that proved invaluable when I eventually returned to LM6. For an entire year, I wasn’t sure what to do with Lowerated or LM6. The idea lingered, but it felt stagnant.
What is LM6?
LM6 isn’t just another rating system; it’s a new way to evaluate movies. Traditional systems rely on stars or numbers that often lack depth or context. LM6 uses AI to analyze movies across seven critical aspects:
Cinematography 🎥
Direction 🎬
Story 📖
Characters 🎭
Unique Concepts 💡
Emotions ❤️
Production Design 🏛️
By breaking down these elements, LM6 delivers a more nuanced and accurate evaluation. It’s not just about whether a movie is good or bad—it’s about understanding why.
The 50 Videos Experiment
This year, we started working on an ambitious experiment: a series of 50 videos, each featuring a movie rated by LM6. Starting with the 50th best movie, we worked backward, explaining the calculations behind each rating and sharing insights about the films. Here’s what we accomplished:
50 Videos shared across platforms.
250 Movies analyzed, reviewed, and rated.
1 Research Paper detailing the methodology.
Countless Blog Posts explaining LM6’s philosophy.
100s of Hours invested in development and discussions.
Open-Source Library and Models released for free.
Why LM6 Matters
LM6 is more than just a tool; it’s a movement to bring depth and fairness to movie ratings. It’s open-source, free, and designed to evolve with community input. Anyone can integrate LM6 into their website or app, making it accessible to creators, critics, and movie lovers alike.
What’s Next?
As we close this chapter, LM6 is just the beginning. Lowerated will continue to grow, with more modules and innovations on the horizon. For now, we’re thrilled to share the journey and can’t wait to see how the community uses LM6 to shape the future of movie criticism.
The Conclusion: Goodhart’s Law and LM6
Goodhart’s Law states: “When a measure becomes a target, it ceases to be a good measure.” Traditional movie rating systems fall victim to this principle. By reducing the evaluation of a film to a single number, they incentivize creators to chase high ratings rather than focus on authentic storytelling. LM6 is designed to counter this by embracing subjectivity. Instead of forcing a single narrative, it evaluates movies across multiple dimensions, allowing for a richer and more nuanced understanding of art.
IMPORTANT:
"""
Ratings, at their core, are subjective—a reflection of individual experiences and values. LM6 doesn’t aim to eliminate this subjectivity but to structure it in a way that highlights the diverse elements contributing to a movie’s impact. With LM6, we’re not just rating movies; we’re appreciating the art behind them. This is Lowerated: where art is truly appreciated.
"""
Comments