SAFETY - AI Dynamics

Training objectives: data reconstruction versus distribution learning

By

–

01 February 2023 11h21

Como comenté, y de donde sale ese tweet, el objetivo de entrenar para "reconstruir una copia de los datos de entrenamientos" es apuntar a plagiar esos datos. Y no es lo mismo que el objetivo al que se apunta con estos modelos generativos: "aprender la distribución de los datos".

→ View original post on X — @dotcsv

1 February 2023

15-Year-Old Creates AI-Powered Fire Safety App

By

@officialindiaai

–

01 February 2023 9h40

When it comes to AI and its potential to transform humanity, Sainath Manikandan has a revolutionary mindset. Read more about how this 15-year-old innovator created an AI-powered app for fire safety: https://
bit.ly/3WVYRtE

→ View original post on X — @officialindiaai

1 February 2023

Deceiving AI: How the Longest Cow Fooled Computer Vision

By

@pascal_bornet

–

01 February 2023 5h30

The longest cow in the world …
Or how to deceive an AI program! Credits: Ralph Aboujaoude Diaz
#ai #artificialintelligence #computervision #machinelearning #meme #funny

→ View original post on X — @pascal_bornet

1 February 2023

94 Million AI-Generated Images Raise Quality Concerns

By

@dotcsv

–

01 February 2023 2h02

Ciertamente, 94 imágenes de 175.000.000 generaciones no son nuevas…

→ View original post on X — @dotcsv

1 February 2023

AI Text Detection False Positives Impact Student Evaluation Fairness

By

@thom_wolf

–

31 January 2023 23h49

a tricky thing with integrating AI-text detection in grading or evaluation is that false positives have significantly worse impact on the writer than false negatives have on the evaluating entity

→ View original post on X — @thom_wolf

31 January 2023

Justice Department Scrutinizes Tesla Self-Driving Technology

By

@cademetz

–

31 January 2023 21h19

The Justice Department is scrutinizing Tesla's self-driving technology, @JackEwingNYT and I report:

→ View original post on X — @cademetz

31 January 2023

Bypassing AI Text Classifiers: From Detection to Human-Like Output

By

@saboo_shubham_

–

31 January 2023 19h43

Step-3: Check the newly generated text again by running it through OpenAI classifier. And Voila The text is now categorized as human-generated!

→ View original post on X — @saboo_shubham_

31 January 2023

Bypassing AI Classifiers with Synonym Replacement Techniques

By

@saboo_shubham_

–

31 January 2023 19h40

Step – 2: Copy the text and run it through GPT-Minus1 that will help you fool the classifier by randomly replacing words with synonyms in your text. Copied the previous text and got the newly generated text from GPT-Minus1 application.

→ View original post on X — @saboo_shubham_

31 January 2023

How to fool AI-generated text classifiers in three steps

By

@saboo_shubham_

–

31 January 2023 19h38

And here is how you can easily fool the classifier in three simple steps Step – 1: Get the AI-generated text and run it through the OpenAI classifier. In this example, the text is classified as "AI-generated"

→ View original post on X — @saboo_shubham_

31 January 2023