Hacker Wiki

Hacking The Planet since 2004

User Tools

Site Tools

You are here: start » what_is_generative_ai
what_is_generative_ai

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
what_is_generative_ai [2025/11/02 05:43] – rewrote with ChatGPT. Better then before, still needs work. jedite83what_is_generative_ai [2025/11/02 05:46] (current) – [15. Responsible disclosure & whitehat norms for hackers] jedite83
Line 1: Line 1:
 ====== Generative AI — An in-depth guide for hackers (DokuWiki) ====== ====== Generative AI — An in-depth guide for hackers (DokuWiki) ======
- 
-~~TOC~~ 
  
 ===== 1. TL;DR / Executive summary ===== ===== 1. TL;DR / Executive summary =====
Line 13: Line 11:
  
 ===== 3. The main model families (what they are, how they sample) ===== ===== 3. The main model families (what they are, how they sample) =====
- 
---- 
  
 **3.1 Transformers (autoregressive & encoder-decoder variants)** **3.1 Transformers (autoregressive & encoder-decoder variants)**
Line 20: Line 16:
 * Core idea: self-attention lets the model compute context-dependent representations across all positions. Attention operation: * Core idea: self-attention lets the model compute context-dependent representations across all positions. Attention operation:
  
-  <code> +<code>Attention(Q,K,V) = softmax( Q K^T / sqrt(d_k) ) V </code>
- +
-Attention(Q,K,V) = softmax( Q K^T / sqrt(d_k) ) V </code>+
  
 * GPT-style models: stack masked self-attention layers for autoregressive generation (predict next token). Pre-trained on massive corpora, often fine-tuned. * GPT-style models: stack masked self-attention layers for autoregressive generation (predict next token). Pre-trained on massive corpora, often fine-tuned.
  
---- 
  
 **3.2 Diffusion / score-based models (images, audio, sometimes text)** **3.2 Diffusion / score-based models (images, audio, sometimes text)**
Line 32: Line 25:
 * Define a forward process that gradually adds noise to data; train a neural network to reverse that noising. Sampling reverses the process via iterative denoising. Connects to denoising score matching and stochastic differential equations. * Define a forward process that gradually adds noise to data; train a neural network to reverse that noising. Sampling reverses the process via iterative denoising. Connects to denoising score matching and stochastic differential equations.
  
---- 
  
 **3.3 GANs, flows, VAEs (historical / specialized use)** **3.3 GANs, flows, VAEs (historical / specialized use)**
Line 128: Line 120:
 ===== 15. Responsible disclosure & whitehat norms for hackers ===== ===== 15. Responsible disclosure & whitehat norms for hackers =====
  
-* Don’t publish exploit code; follow coordinated disclosure. +  * Don’t publish exploit code; follow coordinated disclosure. 
-* Contact vendors via official channels. +  * Contact vendors via official channels. 
-* Provide minimal test cases, operational impact, and mitigation suggestions.+  * Provide minimal test cases, operational impact, and mitigation suggestions.
  
 ===== 16. Further reading & canonical sources ===== ===== 16. Further reading & canonical sources =====
  
-* OpenAI API docs & overviews. +  * OpenAI API docs & overviews. 
-* Diffusion model surveys. +  * Diffusion model surveys. 
-* OWASP GenAI. +  * OWASP GenAI. 
-* Academic studies on prompt injection and red-teaming. +  * Academic studies on prompt injection and red-teaming. 
-* Industry reports on failures & jailbreaks.+  * Industry reports on failures & jailbreaks.
  
 ===== 17. Appendix — Glossary ===== ===== 17. Appendix — Glossary =====
  
-* LLM — Large Language Model. +  * LLM — Large Language Model. 
-* RLHF — Reinforcement Learning from Human Feedback. +  * RLHF — Reinforcement Learning from Human Feedback. 
-* Diffusion — iterative denoising generative family. +  * Diffusion — iterative denoising generative family. 
-* Hallucination — fluent but false output. +  * Hallucination — fluent but false output. 
-* Prompt injection — input that subverts model intentions.+  * Prompt injection — input that subverts model intentions.
  
 ===== 18. Closing / ethical call to arms ===== ===== 18. Closing / ethical call to arms =====
 Generative AI amplifies productivity and abuse potential. Hackers, researchers, and operators should approach systems with adversarial mindset, understand internals, threat models, auditing techniques, and responsible disclosure norms. Defensive engineering and continuous red-teaming are mandatory. Generative AI amplifies productivity and abuse potential. Hackers, researchers, and operators should approach systems with adversarial mindset, understand internals, threat models, auditing techniques, and responsible disclosure norms. Defensive engineering and continuous red-teaming are mandatory.
  
what_is_generative_ai.1762062198.txt.gz · Last modified: by jedite83