Mitigating model limitations

Comprender la ingeniería de prompts

Alex Banks

Founder & Educator

Introduction to model limitations

 

Group 84.png

Comprender la ingeniería de prompts

The reversal curse

 

Group 85.png

Comprender la ingeniería de prompts

The reversal curse

Group 86.png

Comprender la ingeniería de prompts

Biases - a mirror of Society

 

Group 87.png

Comprender la ingeniería de prompts

Hallucinations - when the model imagines

 

Group 88.png

Comprender la ingeniería de prompts

Hallucinations - when the model imagines

 

Group 89.png

Comprender la ingeniería de prompts

Hallucinations - when the model imagines

 

Group 90.png

Comprender la ingeniería de prompts

Overfitting - echoing the data

Model is too closely tailored to its training data.

Makes it less effective at generalizing to new, unseen data.

image_115__3_-removebg-preview 1.png

Comprender la ingeniería de prompts

Overfitting - echoing the data

Comedy challenges large language models like ChatGPT.

Researchers found that 90% of 1,008 generations were the same 25 jokes.

image_115__2_-removebg-preview 1.png

Comprender la ingeniería de prompts

Overfitting - echoing the data

Group 91.png

Comprender la ingeniería de prompts

Let's practice!

Comprender la ingeniería de prompts

Preparing Video For Download...