LMMs Not LLMs: Large Mumbling Models

I’m sure in the past few months, you have heard a lot of hype about AI reasoning. Starting with OpenAI's o-series of models gaining human-like reasoning capabilities. Since then, all the major AI labs have released their own reasoning models or are about to release one. A few other reasoning models that attracted a lot of attention include the DeepSeek R1 and Claude 3.7 thinking mode. We have gone from Large Langauge Model to Large Reasoning Models (LRMs). But like me, I think many of you still have your doubts about the true capabilities of these models beyond the fancy demos.

So, today, we are diving deeper into what’s actually happening out there in the space of LLMs. Are these truly reasoning models, or are they just mumbling models?

Table of Contents

• Introduction To Reasoning
• Types Of Reasoning
• Search Algorithms + LLMs
• Test Time Scaling
• DeepSeek’s Reasoning Is Different
• “Prompt Augmentations” Not Chains Of Thoughts!
• LLM “Reasoning Traces” Or Large Mumbling Models
• Conclusion

Photo by pure julia on Unsplash

Introduction To Reasoning