I was right again
📚 Main Topics
Images vs. Text in AI Models
- Recent research suggests that images may be more efficient than text for AI models.
- DeepSeek's paper indicates that one image token can yield ten text tokens with high accuracy.
Efficiency and Cost Reduction
- Using images can lead to a 59-70% reduction in end-to-end costs.
- The speaker shares personal experience with token usage in experiments.
Technical Explanation
- Text tokens are discrete, while image tokens are continuous, allowing for more information to be represented in fewer tokens.
- The inefficiency of text tokens in AI models is discussed, highlighting the potential advantages of using images.
Practical Experimentation
- The speaker conducted experiments using a game context file, comparing performance between text and image inputs.
- Results showed that while image-based inputs took longer to process, they could potentially yield better results in terms of damage dealt in the game.
Cautions and Recommendations
- The speaker advises against rushing to convert all data to images without thorough testing.
- Emphasizes the importance of experimentation and optimization in AI applications.
✨ Key Takeaways
- Images may outperform textin certain AI applications, particularly in terms of efficiency and cost.
- Experimentation is crucialResults can vary significantly based on the context and method used.
- Understanding the underlying technologyis essential before making changes to data input methods.
🧠 Lessons Learned
- Don't follow trends blindlyAlways validate claims with personal experimentation and research.
- Be aware of the limitationsof current technologies, especially when transitioning from text to image-based inputs.
- Optimization is an ongoing processContinuous testing and adaptation are necessary to find the best solutions in AI development.
Transcript excerpt
0:04 You see that face? You know what that face is? That that is the smug face of somebody who was proven correct. And the best part, it's right after caught in 8K. You're even told by your friends you're wrong. So, you're probably thinking, "Okay, what what are we talking about? What are you even trying to say? What were you proven right about?" Well, my friends, it turns out after many hurt feelings and being told I was wrong, images might actually be better for models than text. I knew it.
0:35 Don't make it do image recognition on that, dude. >> No, no, you just do this. What are you talking about? >> Hey, >> now this isn't just me saying this, of course. This is actually DeepSeek releasing a new paper that suggests that if you provide images, you could pull out 10 text token from a single image token with near 100% accuracy. In other words, a model's internal representation of an image is 10 times efficient as its internal representation of text.
🔒 The full, searchable transcript is available with Pro.
Questions & Answers
Common questions about this video
What recent discovery suggests that images might be more efficient than text for models?
How do image tokens compare to text tokens in terms of information density?
What is one reason images might encode more information than text in models?
What were the results of the experiment comparing context as a string versus as an image in a game-playing AI?
Why might using images as context be more challenging for models, according to the experiment?
What does the speaker suggest about experimenting with image inputs for models?
What is the overall takeaway regarding the use of images versus text for model inputs?
Related summaries
This Might Be the Best AI Release of 2026
Learning is Cool
The Token Black Market
Why You Should AVOID Linked Lists
This is the only AI I want to use
