AI Evolution: Classical AI to GenAI - Cedric Williams | Library

Ɛhia

1

According to the video, what are the three core behaviors shared by classical A* search, game AI, and modern tools like ChatGPT?

Read, Write, and Speak

Logic, Reason, and Fact-Check

Explore, Score, and Pick

Input, Process, and Output

Ɛhia

1

In the classical A* search formula f(n)=g(n)+h(n), what does the h(n) represent?

The actual cost from the starting point to the current position

The total number of nodes in the search tree

The heuristic estimate (the "guess") of the cost to reach the goal

The speed at which the algorithm is running

Ɛhia

1

Why did classical AI hit a "wall" when trying to solve complex problems like the game of Chess?

The computers ran out of electricity.

The number of possible moves (states) grew exponentially, becoming larger than the number of atoms in the universe.

Humans forgot how to write the code for the game.

The AI became too smart and refused to play.

Ɛhia

1

What was the primary change during the "bridge era" when machine learning began to replace classical AI methods?

AI stopped using search algorithms entirely.

Hand-crafted rules were replaced by neural networks that learned patterns and heuristics from data.

Humans started writing even more complex "if-then" rules by hand.

Computers were trained to think exactly like human beings.

Ɛhia

1

When viewing a Large Language Model (LLM) as a search graph, what does each edge (the connection between nodes) represent?

A physical wire in the computer

A complete sentence generated by the user

The prediction of the next token (word) in a sequence

The memory stored in the database

Ɛhia

1

In the search space of an LLM, how is the "cost" of a path determined?

By the amount of money it costs to run the server

By the number of letters in a word

By negative log probability (tokens with lower probability have a higher cost)

By how many times a human has used that specific word before

Ɛhia

1

What is the primary advantage of Beam Search compared to simple "Greedy" decoding?

It is much faster and uses less memory.

It maintains multiple promising paths (the "K best" paths) simultaneously rather than just picking one.

It allows the AI to search the live internet for answers.

It prevents the AI from making any grammatical mistakes.

Ɛhia

1

If a developer sets an LLM’s temperature to a high value (such as 1.5), what effect does it have on the output?

It makes the model strictly follow only the highest-probability words.

It makes the model stop generating text to prevent overheating.

It increases diversity by making lower-probability tokens more likely to be selected.

It forces the model to only give short, one-word answers.

Ɛhia

1

The video mentions "Tree of Thoughts" as a modern technique. How does this technique change how an LLM works?

It forces the model to memorize a dictionary.

It builds an actual search tree where the model explores and evaluates different reasoning paths.

It allows the model to draw pictures of its "thoughts."

It connects the LLM to a biological brain.