A quick test, if I subtract out the mean of 70 samples, I get much more sensible results. the first 5 are chatGPT suggestions for dissimilar sentences, and the next 5 for similar:
-0.017663949682680865
-0.07352484277035345
-0.05597005318789076
-0.009429209531217298
-0.06919492370655664
0.6165518204173611
0.5964354661570286
0.7516415313500149
0.8033141561180126
0.6907252749720518
Sentences were:
"The cat sat on the mat" , "The number 42 is the answer to the ultimate question of life, the universe, and everything",
"The grass is green" , "The stock market has been very volatile recently",
"I like to play chess" , "The weather is hot today",
"She is a nurse" , "The company's profits have been increasing",
"The sun rises in the east" , "The movie was not as good as the book",
"The cat sat on the mat" , "The feline was perched on the rug",
"I like to play chess" , "I enjoy playing the strategy game",
"She is a nurse" , "She works in healthcare",
"The sun rises in the east" , "The morning star rises in the east",
"The apple fell from the tree" , "The fruit dropped from the branches"