Humans are the Top Layer of the AI Stack

  • December 9th, 2024
  • by David Spuler, Ph.D.

Humans on Top

Sometimes it seems like humanity is missing from vendor discussions of AI architectures. After all, it is humans that are issuing the queries and reviewing the LLM's results. It is human productivity that we aim to improve, after all.

So, here's my vote for the most under-utilised neural network in the whole AI industry: your brain. LLMs and their agent brethren should be considered as productivity extensions of the human brain, rather than the almighty sprint to thrust silicon shoulders through the AGI tape at the finish line. The average user's brain has much better capabilities than any LLM, except when it comes to waxing lyrical about how to organize your closet, when phrased in Klingon vernacular.

Too much of the AI rhetoric is based on the idea of "human in the loop," which is, when you think about it, a rather dismissive stance. It relegates humans to the position of one cog in the engine, as a reviewer or approver, rather than even a worker. Really, humans should be considered to be an improver.

Humans are rather good at all the things that LLMs struggle to do, so there is an important aspect to the human-AI partnership. At least, Microsoft seems to have it right as a proponent of "copilot" uses of AI, although even here sometimes you wonder which of the two pilots has the conn. If you've ever fought with Microsoft Word over where to place an image, well, you get the idea.

Hence, one of the things we think about at Aussie AI is how to use LLMs to better human lives. This means that humans have the role as:

  • Initiator
  • Guide
  • Decider (reviewer/approver)

In practice, this philosophy offers a framework for the user interface for LLM apps. The user is the one who chooses the app to run, the inputs to offer, and the manner of execution (by an LLM, yes). Approval of an LLM's output can be explicit, or it can be implicit in the choice across multiple answers, or in the decision on whether or not to proceed with a given result.

Related Research

Free AI and C++ Books

Generative AI programming books:

  1. The Sweetest Lesson: Your Brain Versus AI, November 2025: full text online, free PDF available
  2. RAG Optimization: Accurate and Efficient LLM Applications, June 2025: full text online, free PDF available
  3. Generative AI Applications: Planning, Design and Implementation, November 2024: full text online, free PDF available
  4. Generative AI in C++ (Spuler, March 2024): full text online, free PDF available, table of contents, bonus materials, reference lists, source code

CUDA C++ GPU Programming Books:

  1. CUDA C++ Optimization: Coding Faster GPU Kernels, July 2024: full text online, bonus materials, free PDF available
  2. CUDA C++ Debugging: Safer GPU Kernel Programming, July 2024: full text online, free PDF available

Modern C++ Programming Books

  1. C++ AVX Optimization: CPU SIMD Vectorization, 2025: full text online, free PDF available
  2. C++ Ultra-Low Latency: Multithreading and Low-Level Optimizations, 2025: full text online, free PDF available
  3. Advanced C++ Memory Techniques: Efficiency and Safety, 2025: full text online, free PDF available
  4. Efficient C++ Multithreading: Modern Concurrency Optimization, 2025: free PDF available
  5. Efficient Modern C++ Data Structures: Container and Algorithm Optimizations, 2025: free PDF available
  6. C++ Low Latency: Multithreading and Hotpath Optimizations, 2025: free PDF available
  7. Safe C++: Fixing Memory Safety Issues, Oct 2024: full text online, free PDF available

More AI Research Topics

Read more about: