A thread on Twitter wrote a shocking thing. Tweet from @giannis_daras states that there is an artificial intelligence (AI) that develops a secret language alias its own language.
"DALLE-2 has a secret language, 'Apoploe vesrreaitais' means bird. 'Contarra ccetnxniams luryca tanniounons' means insect or pest," reads the thread.
The researchers concluded that DALLE 2 considers Vicootes to mean 'vegetables', while Wa ch zod rea refers to 'sea creatures that whales can eat'.
DALLE-2 has a secret language.
"Apoploe vesrreaitais" means birds.
"Contarra ccetnxniams luryca tanniounons" means bugs or pests.
The prompt: "Apoploe vesrreaitais eating Contarra ccetnxniams luryca tanniounons" gives images of birds eating bugs.
A thread (1/n)🧵 pic.twitter.com/VzWfsCFnZo
— Giannis Daras (@giannis_daras) May 31, 2022
So what really happened? Is it true that AI DALLE-2 has its own language?
Science Alert reported as seen on Wednesday (8/6/2022) that they are more convinced that this is not a new language of AI. It is possible that the 'original' language used by AI is a word taken from the many languages he has learned. For example, 'Apoploe' which may be taken from the Latin Apodidae which is the family name of a bird species. Moreover, DALLE-2 is not only learning English.
A netizen also thought that what happened was the tokenizer effect. It is known that AI processes language unlike humans, they will break text into 'tokens' before processing.
DALL-E 2 uses an approach called byte-pair encoding (BPE). By studying the BPE representations for some of the 'secret' words, this could be an important clue in understanding the AI 'secret language'.
that "secret language" seems like mostly tokenizer effects. you can do the inverse too:
1) i picked two families of fish "Actinopterygii" and "Placodermi" from wikipedia
2) prompted dalle with "placoactin knunfidg"
3) dalle consistently generates fish images https://t.co/ndAe7MURyg pic.twitter.com/1kHk5NWJb3
— rapha gontijo lopes (@iraphas13) June 3, 2022