3 ms·

This model is better by many other contenders, but still far from GPT4. "what famous brands are there which change one letter from a common word to make a non-

by vizzah 3y ago

This model is better by many other contenders, but still far from GPT4. "what famous brands are there which change one letter from a common word to make a non-existent, but a catchy name, such as "musiq" instead of "music".. etc?" There are several brands that have played with words by changing a letter or adding a letter to create a new and memorable name. Here are a few examples: Qatar Airways - This airline's name was inspired by the country of Qatar, but with the letter "Q" replaced by "Q" to create a unique and catchy name. Lush Cosmetics - This cosmetics company's name is a play on the word "lush," which means "profuse" or "luxuriant." Pinterest - This social media site's name is a combination of the words "pin" and "interest," but with the letter "e" replaced by "i" to create a more unique and memorable name. Tumblr - This social media platform's name is a play on the words "tumble" and "blogger," but with the letter "l" replaced by "r" to create a more unique and catchy name. --------- that is compared to hilarious Yi 34B Chat below: =) There are several famous brands that have used this strategy of changing one letter from a common word to create a unique and catchy name. Here are a few examples: eBay - Instead of "cBay," which would be a reference to "crab," eBay was named after the first three letters of the word "baY" and the "e" from the word "electronic," as in "electronic marketplace." Google - Originally named "BackRub," the company changed its name to Google, a play on the word "googol," which refers to the number one followed by one hundred zeros, expressing the vastness of the internet. Kodak - The name Kodak is derived from the word "cod" with an "k" added. It was chosen because it is short, easy to remember, and not suggestive of the photographic nature of the business.

brucethemoose2 3y ago
Here is my response from Yi, which is actually pretty good: One well-known example is the brand Apple. The company's name comes from a play on the phrase "Apple of My Eye." Another example is Google, which is derived from the mathematical term "googolplex," which refers to the number 1 followed by 100 zeroes. It was originally meant to be a play on the word "googol" (1 followed by 100 zeroes), but a spelling mistake led to the creation of the new word "google." Other examples include PepsiCo, which combines the words "pepsin" and "kola," and Amazon, which was inspired by the world's largest river, the Amazon River. In the case of "musiq," it seems like a made-up name without any specific reference or known origin. However, some possible variations could be created using existing words, such as "dramatize" instead of "dramatize" or "fashin" instead of "fashion."
- brucethemoose2 3y ago
  Actually I just realized I had the prompt template and sampling wrong. First shot once that's fixed: " Some examples of famous brands that use this strategy include: 1. KFC (Kentucky Fried Chicken) - The brand takes the first letter of each word in its full name (Kentucky Fried Chicken) and combines them to create a memorable acronym. 2. Pepsi - The brand's name is a play on the word "pepsin," an enzyme used in digestion. The name was chosen because it has a positive connotation and is easy to remember. 3. Adidas - The brand's name is derived from the founder's name, Adolf Dassler, with the 'Das' replaced by an 'a'. 4. Asics - The brand's name is an acronym for "Animal, Speed, Intelligence, and Spirit." 5. H&M (Hennes & Mauritz) - The brand's name is a combination of two Swedish words: "Hennes" means "her" in English, and "Mauritz" refers to the men's clothing store "Mauritz Widforss." 6. Nike - The brand's name comes from the Greek goddess of victory, "Nike." 7. Amazon - The brand's name was inspired by the South American river Amazon, which is one of the world's largest rivers. 8. Apple - The brand's name was chosen due to co-founder Steve Jobs' belief that "Apple" was simple, memorable, and had a positive connotation. 9. Google - The brand's name is a play on the mathematical term "googol," which represents the number 1 followed by 100 zeros. 10. Twitter - The brand's name is a shortened version of "tweet," which refers to short messages sent through the platform. These brands demonstrate how changing or combining letters from existing words can create unique and memorable names for businesses. " - This is not vanilla yi, but a merge of several finetunes.
mirekrusin 3y ago
LLMs work on tokens where characters are hidden away. They'd have to be explicitly trained on spelling each token out into single letter tokens and as they are bad at information symmetry - from single letter tokens back onto tokens as well. I don't think anybody does this so they're left with what's in training data only. Otherwise they don't have chance to reconstruct this information as tokens could map to any equivalent visual language ie. looking like hieroglyphs where questions you're asking would be nonsensical.
- jakderrida 3y ago
  I thought so, too. But then I asked it to define fake words that were portmanteaus I made up. Believe me, my understanding of BERT and discriminant models aligned perfectly with what you're saying. But testing out the theory that it can break down and make meaning of fake words with accurate depictions of what words I'm combining proved me wrong. Generative models must work differently than you and I thought.
  - mirekrusin 3y ago
    Now you're talking about words, above token abstraction - they're fine with that. They're bad at subtoken level.
nl 3y ago
But both are completely wrong! And technically the Google example is closer to correct than any others. The Yi 34B eBay and Kodak examples are both (wrong but) very interesting because it does seem to get the idea of changing one letter. Of GPT4 examples, the Qatar example (replacing "Q" with "Q" !?) is the only one that is internally consistent. The Pinterest and Tumblr examples are wrong in very odd ways in that the explanation doesn't match the spelling.
- miles 3y ago
  > Of GPT4 examples, the Qatar example (replacing "Q" with "Q" !?)... That appears to be from vizzah's testing of Mistral-8x7B-Chat rather than GPT4.
  - nl 3y ago
    Is it? I thought the first was GPT4 and the second was mislabeled "Yi 34B Chat" when they meant "Mistral-8x7B-Chat"? Otherwise why are they saying it is far from GPT4?
behnamoh 3y ago
Are you comparing a 8x7b model with GPT-4? Come on...
- brucethemoose2 3y ago
  A raw one with no wrappers, at that.

by vizzah 3y ago

brucethemoose2 3y ago

Here is my response from Yi, which is actually pretty good: One well-known example is the brand Apple. The company's name comes from a play on the phrase "Apple of My Eye." Another example is Google, which is derived from the mathematical term "googolplex," which refers to the number 1 followed by 100 zeroes. It was originally meant to be a play on the word "googol" (1 followed by 100 zeroes), but a spelling mistake led to the creation of the new word "google." Other examples include PepsiCo, which combines the words "pepsin" and "kola," and Amazon, which was inspired by the world's largest river, the Amazon River. In the case of "musiq," it seems like a made-up name without any specific reference or known origin. However, some possible variations could be created using existing words, such as "dramatize" instead of "dramatize" or "fashin" instead of "fashion."

brucethemoose2 3y ago
Actually I just realized I had the prompt template and sampling wrong. First shot once that's fixed: " Some examples of famous brands that use this strategy include: 1. KFC (Kentucky Fried Chicken) - The brand takes the first letter of each word in its full name (Kentucky Fried Chicken) and combines them to create a memorable acronym. 2. Pepsi - The brand's name is a play on the word "pepsin," an enzyme used in digestion. The name was chosen because it has a positive connotation and is easy to remember. 3. Adidas - The brand's name is derived from the founder's name, Adolf Dassler, with the 'Das' replaced by an 'a'. 4. Asics - The brand's name is an acronym for "Animal, Speed, Intelligence, and Spirit." 5. H&M (Hennes & Mauritz) - The brand's name is a combination of two Swedish words: "Hennes" means "her" in English, and "Mauritz" refers to the men's clothing store "Mauritz Widforss." 6. Nike - The brand's name comes from the Greek goddess of victory, "Nike." 7. Amazon - The brand's name was inspired by the South American river Amazon, which is one of the world's largest rivers. 8. Apple - The brand's name was chosen due to co-founder Steve Jobs' belief that "Apple" was simple, memorable, and had a positive connotation. 9. Google - The brand's name is a play on the mathematical term "googol," which represents the number 1 followed by 100 zeros. 10. Twitter - The brand's name is a shortened version of "tweet," which refers to short messages sent through the platform. These brands demonstrate how changing or combining letters from existing words can create unique and memorable names for businesses. " - This is not vanilla yi, but a merge of several finetunes.

mirekrusin 3y ago

LLMs work on tokens where characters are hidden away. They'd have to be explicitly trained on spelling each token out into single letter tokens and as they are bad at information symmetry - from single letter tokens back onto tokens as well. I don't think anybody does this so they're left with what's in training data only. Otherwise they don't have chance to reconstruct this information as tokens could map to any equivalent visual language ie. looking like hieroglyphs where questions you're asking would be nonsensical.

jakderrida 3y ago
I thought so, too. But then I asked it to define fake words that were portmanteaus I made up. Believe me, my understanding of BERT and discriminant models aligned perfectly with what you're saying. But testing out the theory that it can break down and make meaning of fake words with accurate depictions of what words I'm combining proved me wrong. Generative models must work differently than you and I thought.
- mirekrusin 3y ago
  Now you're talking about words, above token abstraction - they're fine with that. They're bad at subtoken level.

nl 3y ago

But both are completely wrong! And technically the Google example is closer to correct than any others. The Yi 34B eBay and Kodak examples are both (wrong but) very interesting because it does seem to get the idea of changing one letter. Of GPT4 examples, the Qatar example (replacing "Q" with "Q" !?) is the only one that is internally consistent. The Pinterest and Tumblr examples are wrong in very odd ways in that the explanation doesn't match the spelling.

miles 3y ago
> Of GPT4 examples, the Qatar example (replacing "Q" with "Q" !?)... That appears to be from vizzah's testing of Mistral-8x7B-Chat rather than GPT4.
- nl 3y ago
  Is it? I thought the first was GPT4 and the second was mislabeled "Yi 34B Chat" when they meant "Mistral-8x7B-Chat"? Otherwise why are they saying it is far from GPT4?

behnamoh 3y ago

Are you comparing a 8x7b model with GPT-4? Come on...

brucethemoose2 3y ago
A raw one with no wrappers, at that.