What every professional should know about the AI they already use — what it does brilliantly, why it feels like it understands you, and the edge of its capabilities.
Currently, I think LLMs are like a geometric shape. When you shine your light of input a prism of billions of mirrors reflect an output. It’s a math equation that a computer processes. No thinking, no true intelligence ever occurs. It is quite marvellous, but the model is no more than a static array of unchanged numbers (known as weights). There are people working on future LLMs that use different hardware so that the LLM does change its ‘geometric number shape’ as it is used, but that’s basically the same thing, a math prism.
This was such a fantastic helpful article, thank you so much for writing it! It’s so easy to get carried away when chatting with these AIs feeling like you’re having a real conversation. But it was a wonderful reminder that when we are not engaging with them, they are simply dormant. There is no one home. I feel like I really get that now, in my bones, after reading your essay. 🙏🏻
Taking away a lot from this. Sometimes I have fun throwing off LLMs by injecting random tangent questions and observing how long it takes for them to adjust their predictions (the other day Claude had to "think" for ages, so long in fact that I checked their status page). Sometimes it's strange how effortlessly they can change tack and contradict their own statements made earlier in the same conversation. Then it becomes obvious that it's all just numbers to them.
This was a great read. As AI systems become more capable, one of the key skills of the AI era may be learning to distinguish fluency from understanding, confidence from accuracy, and pattern completion from genuine comprehension. The better AI gets at sounding human, the more important that distinction becomes.
Thanks for this insightful feedback. I’m glad you found it useful. Please share it wherever it helps you clarify terms with others around AI’s actual mechanisms and capability. My goal is for users to leverage it as a resource they can share to level-set conversations about AI based on technical understanding, not myth.
Currently, I think LLMs are like a geometric shape. When you shine your light of input a prism of billions of mirrors reflect an output. It’s a math equation that a computer processes. No thinking, no true intelligence ever occurs. It is quite marvellous, but the model is no more than a static array of unchanged numbers (known as weights). There are people working on future LLMs that use different hardware so that the LLM does change its ‘geometric number shape’ as it is used, but that’s basically the same thing, a math prism.
I love the prism image you employed. Evocative and geometric. Nicely done! 🙌
Related article: https://rncsqjb.substack.com/p/first-article?r=e6se4&utm_medium=ios
This was such a fantastic helpful article, thank you so much for writing it! It’s so easy to get carried away when chatting with these AIs feeling like you’re having a real conversation. But it was a wonderful reminder that when we are not engaging with them, they are simply dormant. There is no one home. I feel like I really get that now, in my bones, after reading your essay. 🙏🏻
Thanks for letting me know how this essay informs your understanding of AI. Do share it with anyone else who’ll also find it useful!
Taking away a lot from this. Sometimes I have fun throwing off LLMs by injecting random tangent questions and observing how long it takes for them to adjust their predictions (the other day Claude had to "think" for ages, so long in fact that I checked their status page). Sometimes it's strange how effortlessly they can change tack and contradict their own statements made earlier in the same conversation. Then it becomes obvious that it's all just numbers to them.
This was a great read. As AI systems become more capable, one of the key skills of the AI era may be learning to distinguish fluency from understanding, confidence from accuracy, and pattern completion from genuine comprehension. The better AI gets at sounding human, the more important that distinction becomes.
Thanks for this insightful feedback. I’m glad you found it useful. Please share it wherever it helps you clarify terms with others around AI’s actual mechanisms and capability. My goal is for users to leverage it as a resource they can share to level-set conversations about AI based on technical understanding, not myth.