Artificial intelligence

Content and key concepts

Mathematics
→
Nero.
→
Network
→
Text
→
Transformer
→
Training
→
LLM
→
AI System
How one chapter continues another
00
Map of Modern I.E.
What exactly does neuronetics teach?
The next thing you need is to distinguish between the entrance, the parameters, and the way you learn.
01
Vectors and spaces
How is the direction and value placed in the numeral?
It's gonna take a lot longer than that: orderly signs, length, direction comparison, and projection.
02
Matrixs and transformations
How do you get a new set of signs from these coordinates?
The next thing you need is the conversion and its adjustment factors.
03
Production and gradients
How can we measure the impact of small change?
It's going to be useful: derivative programs and direction of function change.
04
Probability and information
How can we measure uncertainty and false confidence?
It's going to be useful: distribution, measurement of error and comparison of distributions.
05
Optimization
How do you choose the length of the movement against this growth?
It's gonna be useful next: the rule of updating parameters and checking for sustainability.
06
Classic ML
Does success in known examples prove quality in new ones?
It's gonna take a lot of time: different fit, check and generalization.
07
Artificial neuron
What does one neuron add?
It's going to be useful: the computational element with the entrances, the weights, and the exit function.
08
Neural networks
How do you connect them to a flexible program?
Next thing you know, the network as a sequence of operations and keeping the straight path.
09
Backup and Autograde
How can they be applied to an entire network without the manual output of each formula?
It's gonna be useful, the gradient of the whole computer program.
10
As the computer counts the network
What limits the speed and the possibility of launching it?
It's gonna take a lot of time to calculate memory, accuracy, and parallel work.
11
Signals, folding and CNN
How can we highlight local signs of large entrances?
The next thing you need is local signs of image and signal.
12
Text and currentization
What is the input unit of the language model?
Then you'll need it: compatible sequence token ids.
13
Embeddings and Spaces
How does the number become a vector?
It'll be useful to see the currents and compare their directions.
14
Attention
How can information be exchanged between positions?
It's gonna take a lot more than that. Computing the exchange between the currents.
15
Transformer and GPT
How do you collect the shape block and take care of order?
It's gonna take a lot more than that. It'll be a straight-up computational path.
16
How LLM Learns
How does the body of the text change weights?
It's going to be useful: a trained and compatible set of parameters.
17
Inference and quantization
How to choose and add a sequel with a reasonable amount of work?
The next thing you need is a generation cycle, a cache, and a limited choice.
18
RAG, tools and agents
How to separate the application's information and actions from the model?
The next thing you need is a controlled external system with the origin of the data.
19
Large and multimodal systems
How do you divide the work and process the image or sound?
It's gonna take a lot of time to figure out the scale, the exchange and the context budget.
20
I.I. system itself
How do you put them together without a new black box?
It's gonna be useful: your own tested ai-system.