Minor things that bother you

[Scott5114]

I pasted a scholarship letter I wrote before GPT was released into one of those and it estimated that 75% of it was written by AI.

And I've copy-pasted straight from the ChatGPT window into one of them and gotten 0% AI.

Given that LLMs are trained on as large of a corpus of English text as possible, the only thing an "AI detector" is doing is measuring how close to average the text is.

[Max Rockatansky]

People continuing to insist that US 101 doesn't exist on the Golden Gate Bridge because they pedantically looked up California's street and highway code once but didn't bother with AASHTO. 

[Roadgeekteen]

Plus you

I pasted a scholarship letter I wrote before GPT was released into one of those and it estimated that 75% of it was written by AI.

And I've copy-pasted straight from the ChatGPT window into one of them and gotten 0% AI.

Given that LLMs are trained on as large of a corpus of English text as possible, the only thing an "AI detector" is doing is measuring how close to average the text is.

Plus you can use AI to generate all the ideas then rewrite it into your own words.

[Molandfreak]

I pasted a scholarship letter I wrote before GPT was released into one of those and it estimated that 75% of it was written by AI.

And I've copy-pasted straight from the ChatGPT window into one of them and gotten 0% AI.

Given that LLMs are trained on as large of a corpus of English text as possible, the only thing an "AI detector" is doing is measuring how close to average the text is.

Interestingly enough, this passage containing highly-technical terminology was the one part of the letter that wasn't flagged as being AI-generated:

In addition to pursuing my Master of Arts in Urban Planning, I have frequently thought of ways I can utilize my backgrounds in habitat management and botany in further GEOINT-related studies. Courses covering remote sensing applications have taught me how different light wavelengths react with sensors from aeronautical surveys and satellite imagery. The latest development in satellite imagery came with the launch of Landsat 9 in 2021, which added the ability to distinguish coastal aerosol wavelengths from other sources of blue wavelengths. This is a massive development for the field of marine biology, and it inspired me to think of meaningful ways we can further divide the visible spectrum of light. This led to the idea of dividing the green spectrum into different portions to represent chlorophyllic gradients–if different groupings can be established based on leaf colors, it can save foresters time in estimating the density, basal area, and importance of different species of trees in a given forest remotely. Furthermore, according to the principles of the attenuation of light and Beer's Law, forest canopy height can determine the net photosynthesis rate for a forest. Therefore, georeferencing LIDAR data measuring a forest's canopy with a green chlorophyll analysis can produce measurable, remote data of a forest's ecological value to local wildlife and the net production of glucose and oxygen. It can also help foresters decide the best methods of smart forest management to ensure healthy forests reach their full potential both as habitats and community centerpieces.

[kkt]

Is there really a Beers Law?

[Molandfreak]

Is there really a Beers Law?

Urban Dictionary answer: The darker the brew, the less light shines through!

Real answer: It explains why no matter how dense a forest is, the light will never be completely blocked by the leaves on the trees. The amount of light absorbed by each leaf is proportional to the leaves' concentration of chlorophyll and the length from the forest canopy to the ground.