What's is the least related thing to MIT?

results.txt

=== MIT ===
'Ġthe' : 0.27957552671432495
'Ġand' : 0.324958860874176
'Ġin' : 0.3969202935695648
'Ġa' : 0.6269370913505554
'Ġ(' : 0.6961137056350708
=== mit ===
'Ġscoop' : 1.1742193698883057
'Ġtext' : 1.2469427585601807
'Ġon' : 1.2686882019042969
'Ġthe' : 1.2753552198410034
'Ġand' : 1.286698818206787

words.py

"""
https://github.com/huggingface/transformers/issues/1458
"""

from transformers import GPT2LMHeadModel, GPT2Tokenizer
from torch import dot, abs

tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
model = GPT2LMHeadModel.from_pretrained('gpt2')  # or any other checkpoint
word_embeddings = model.transformer.wte.weight  # Word Token Embeddings 

vocab: dict = tokenizer.get_vocab()
reverse_vocab = { v: k for k, v in vocab.items() }

def most_orthogonal(token: str):
	token_of_interest = vocab[token]
	embedding_of_interest = word_embeddings[token_of_interest]

	dot_products = (
		(reverse_vocab[token], abs(dot(embedding_of_interest, word_embeddings[token])))
		for token
		in vocab.values()
		if token != token_of_interest
	)

	sorted_dproducts = sorted(dot_products, key=lambda x: x[1])

	return sorted_dproducts

for token in ["MIT", "mit",]:
	print("===", token, "===",)
	for token, abs_dot in most_orthogonal(token)[:5]:
		print(repr(token), ":", abs_dot.item())