Remember seeing something about GPT-4 doing well on standardized tests? It turns out it may have memorized the answers.
https://aisnakeoil.substack.com/p/gpt-4-and-professional-benchmarks
#gpt4 #AIHype #ThisIsWhyWeDontTestOnTheTrainingData
This profile might be incomplete.
Open on wandering.shop Janelle Shane
Wall 10 posts
Janelle Shane
Remember seeing something about GPT-4 doing well on standardized tests? It turns out it may have memorized the answers.
Janelle Shane
For your daily December dose of surreal:
Janelle Shane
I have used #dalle2 to faithfully recreate product photos of several popular brands of Halloween candy. This is a great way to fool people into believing I have not already eaten all the Sunger bars.
Janelle Shane
Does GPT-3 know what it's like to be human? Yes, it turns out an internet-trained AI can indeed cite the universal human experiences of:
Mans R
@janellecshane Here's an attempt by DALL-E to illustrate one of those. You can probably guess which one.
Janelle Shane
Felt like using #dalle2 to mess up some brands. Here's breakfast cereals! https://www.aiweirdness.com/ai-recreates-classic-cereals/
Janelle Shane
Testing a theory that dalle2 does worse at generating things the smaller they get, I discovered that it makes really mean eye charts.
Show previous comments
teledyn ð“‚€
@janellecshane this is likely an important finding for #GPT3 and #Copilot too, that obscure things (especially to a WIMPish web culture) will be less well defined. So cat portraits are safe...
Janelle Shane
I assumed that if GPT-3 generated pickup lines, they would be terrible because they would be something like human pickup lines. Instead, they are terrible for more delightful reasons.
Janelle Shane
I've been previewing the new @OpenAI API, and honestly despite myself I was a bit nervous when I typed "go ahead and try it." It's that good.
Janelle Shane
Trained StyleGAN2 on pictures of my cat. Not sure whether to laugh or scream |
@janellecshane poke @Agar
@janellecshane not surprised in the least, but it's nice to see it confirmed