@mahgister You mentioned mathematics at the engineering level ... if you or anyone would like to see how an LLM works at that granular of a level check out the video Transformer Neural Networks ... on YouTube channel StatQuest with Josh Starmer. It shows you the actual mathematical formulas used within an LLM.
Thanks for the excellent recommendation. The first video you suggested is the best i ever seen if few minutes describing the LLM in his core idea.
I asked perplexity engine which maths fields are necessary to master N.N. LLM and i underline the main point for my argument (A.I. is a technological collective "tour de force" not a mathematical peak feat it is relatively basic maths applied ) :
«To master neural network-based Large Language Models (LLMs), several key areas of mathematics are essential:
- Linear Algebra: This forms the foundation for understanding the structure and operations of neural networks
- 2. Key concepts include:
- Vectors and matrices
- Matrix operations
- Eigenvalues and eigenvectors
- Vector spaces and linear transformations
- Calculus: Crucial for optimization in neural networks:
- Derivatives and gradients
- Multivariable calculus
- Chain rule (for backpropagation)
- Probability and Statistics: Fundamental for understanding how LLMs learn and make predictions:
- Probability theory
- Random variables and distributions
- Bayesian inference
- Statistical hypothesis testing
- Optimization: Essential for training LLMs:
- Gradient descent and its variants
- Loss functions
- Regularization techniques
- Information Theory: Useful for understanding aspects of language modeling:
- Entropy
- Cross-entropy
- Kullback-Leibler divergence
- Discrete Mathematics: Helpful for algorithmic aspects:
- Graph theory
- Combinatorics
- Real Analysis: While optional, it provides a deeper mathematical foundation.
Additionally, understanding the transformer architecture, attention mechanisms, and tokenization is crucial for LLMs specifically
. Proficiency in coding and data manipulation is also essential for practical implementation
. It’s worth noting that while a strong mathematical background is beneficial, many practitioners start with high school-level math and gradually build their knowledge as they gain practical experience»
.Back to music. I’ve been playing around with AI music generators (just Google that exact term if interested) and as with all language models the quality of what you get out is heavily dependent upon the prompts you put in. I play guitar and I’ve found the more I use precise and concise musical terminology, the closer I get to what I want. Prompt engineering is still an art, although eventually I think AI agents will create music for you based upon your listening history. In the near future, there will be no physical media.
You are here crucially right! but it illustrate my point well: all a user of A.I. will have to do will be learning to master the art of questioning and suggesting... Called the prompt art...
Do you think that it will motivate children to write in cursive and learn poetry ?
or to create more by themselves ?
No i dont think so...
I have two nephews. One was thinking of going to school for computer science, and since I’m a former software engineer, he asked me about it. I told him not to do it because it’s a dead field (as is every field where you interact with a computer all day). The other was thinking about playing guitar, and since I play guitar he asked me about it. I told him to go for it. For one, it’s a whole lot cheaper than wasting your money on a dead field in college. And secondly, even if AI can do it better, I think you can still achieve a sense of accomplishment and it only costs you time and about $700.
This last point confirm my point about demotivation and job loss...Transforming people more and more in "consumers" ...
In all that the worst consequences are not even touched :
Control of all flows ( money,information,people ) by monster corporations
Lost of truth in a virtual world of fictions: how do you know real from fake from events, images, identities ?
Control of "truth" and ministry of "truth" become mandatory and unconscious crowd will ask for it ... Is this ringing a bell ?
The worst is not so far : complete divisions of mankind in 2 groups : The few workers and numerous passive consumers (worthless one )slaves natural human and the cyborgs and their few masters ( Gates and Musk children ? )
And yes there exist miraculously positive use as in medecine , diagnostic and chirurgy...
But those who see this coming with pink glasses are heavily deluded ...
« IT in BITS from the PIT of Babel »--Groucho Marx 
