Generative Artificial Intelligence: More Than You Asked For

Generative artificial intelligence has caught up to years of hype. As a result, companies are scrambling to adopt and deploy generative models. Yet, there are also concerns surrounding this innovative technology. Businesses must approach this opportunity with both optimism and caution. Generative Artificial Intelligence: More Than You Asked For helps decision-makers anticipate foreseeable risks created by the strange technological artifacts of generative technology. This includes the data used to train these models, the tendency of generative models to memorize training data, vulnerabilities to direct and indirect adversarial prompting, its tendency to leak sensitive information, and the rapidly evolving regulatory and legal concerns. Simply put, Richard Heimann and Clayton Pummill promote safer and more secure implementations for businesses with practical knowledge and sensible suggestions.
From the Publisher

Generative Artificial Intelligence: More Than You Asked For

Generated image with Getty watermark.

Generated image with Getty watermark.

Using membership Inference to extract copies from Stable Diffusion.

Using membership Inference to extract copies from Stable Diffusion.

Combining models with retrieval-oriented applications can lead to new security vulnerabilities.

Combining models with retrieval-oriented applications can lead to new security vulnerabilities.

Art by James Daly III (left). Images generated by Stable Diffusion were fine-tuned in his work.

Art by James Daly III (left). Images generated by Stable Diffusion were fine-tuned in his work.

Chapter 1

Generative models are trained on uncurated samples of the internet. These massive and often undocumented datasets contain private and personal data, copyrighted materials, and biased, offensive, and inappropriate content. Unlike other machine learning projects, generative models present technical, ethical, and legal challenges to businesses resulting from third-party black box models pre-trained on third-party inscrutable datasets.

Chapter 2

Generative models inexplicably memorize training data and inadvertently expose sensitive, copyrighted, or private information. Consequently, hurdles exist for organizations deploying these models, especially when trained on sensitive corporate data. The authors describe the problem and some practical strategies to address these challenges.

Chapter 3

Deploying generative technology means being responsible for keeping it secure. Unfortunately, despite the quality of training data, model architecture, or fine-tuning procedure, generative technology is prone to attacks because it is designed to follow user instructions. This vulnerability makes generative models susceptible to data leakage through direct and indirect prompt injections and adversarial attacks. It is crucial for companies to be aware of these risks and adopt strategies to keep their deployments safe.

Chapter 4

Generative technology in the enterprise has potential but faces legal hurdles due to uncurated training data, memorization, and security issues. The fluid regulatory and legal landscapes need constant attention by organizations considering deploying generative technology. The authors share some observations and offer some suggestions.

ASIN ‏ : ‎ B0CB742YW4
Publication date ‏ : ‎ July 26, 2023
Language ‏ : ‎ English
File size ‏ : ‎ 17678 KB
Text-to-Speech ‏ : ‎ Enabled
Enhanced typesetting ‏ : ‎ Enabled
X-Ray ‏ : ‎ Not Enabled
Word Wise ‏ : ‎ Not Enabled
Sticky notes ‏ : ‎ On Kindle Scribe
Print length ‏ : ‎ 138 pages

Price: $9.99
(as of Jul 27, 2023 02:16:43 UTC – Details)

buy now

Generative artificial intelligence has caught up to years of hype. As a result, companies are scrambling to adopt and deploy generative models. Yet, there are also concerns surrounding this innovative technology. Businesses must approach this opportunity with both optimism and caution. Generative Artificial Intelligence: More Than You Asked For helps decision-makers anticipate foreseeable risks created by the strange technological artifacts of generative technology. This includes the data used to train these models, the tendency of generative models to memorize training data, vulnerabilities to direct and indirect adversarial prompting, its tendency to leak sensitive information, and the rapidly evolving regulatory and legal concerns. Simply put, Richard Heimann and Clayton Pummill promote safer and more secure implementations for businesses with practical knowledge and sensible suggestions.
From the Publisher

Generative Artificial Intelligence: More Than You Asked For

Generated image with Getty watermark.

Generated image with Getty watermark.

Using membership Inference to extract copies from Stable Diffusion.

Using membership Inference to extract copies from Stable Diffusion.

Combining models with retrieval-oriented applications can lead to new security vulnerabilities.

Combining models with retrieval-oriented applications can lead to new security vulnerabilities.

Art by James Daly III (left). Images generated by Stable Diffusion were fine-tuned in his work.

Art by James Daly III (left). Images generated by Stable Diffusion were fine-tuned in his work.

Chapter 1

Generative models are trained on uncurated samples of the internet. These massive and often undocumented datasets contain private and personal data, copyrighted materials, and biased, offensive, and inappropriate content. Unlike other machine learning projects, generative models present technical, ethical, and legal challenges to businesses resulting from third-party black box models pre-trained on third-party inscrutable datasets.

Chapter 2

Generative models inexplicably memorize training data and inadvertently expose sensitive, copyrighted, or private information. Consequently, hurdles exist for organizations deploying these models, especially when trained on sensitive corporate data. The authors describe the problem and some practical strategies to address these challenges.

Chapter 3

Deploying generative technology means being responsible for keeping it secure. Unfortunately, despite the quality of training data, model architecture, or fine-tuning procedure, generative technology is prone to attacks because it is designed to follow user instructions. This vulnerability makes generative models susceptible to data leakage through direct and indirect prompt injections and adversarial attacks. It is crucial for companies to be aware of these risks and adopt strategies to keep their deployments safe.

Chapter 4

Generative technology in the enterprise has potential but faces legal hurdles due to uncurated training data, memorization, and security issues. The fluid regulatory and legal landscapes need constant attention by organizations considering deploying generative technology. The authors share some observations and offer some suggestions.

ASIN ‏ : ‎ B0CB742YW4
Publication date ‏ : ‎ July 26, 2023
Language ‏ : ‎ English
File size ‏ : ‎ 17678 KB
Text-to-Speech ‏ : ‎ Enabled
Enhanced typesetting ‏ : ‎ Enabled
X-Ray ‏ : ‎ Not Enabled
Word Wise ‏ : ‎ Not Enabled
Sticky notes ‏ : ‎ On Kindle Scribe
Print length ‏ : ‎ 138 pages

Generative Artificial Intelligence: More Than You Asked For

Hot Deal
Logo