DAIRMar 4, 2023

Our fellow @asme was interviewed for this article by Mirtha Donastorg of the Plug.
https://tpinsights.com/potentially-useful-but-error-prone-chatgpt-on-the-black-tech-ecosystem/

He makes a number of important points in that article and his associated twitter thread: https://twitter.com/asmelashteka/status/1630966233217605632

"Let's start with a demonstration.

I asked #ChatGPT to give me examples of African countries in my native language Tigrinya. The output is not even in Tigrinya. There are words in Tigrinya. There are words in Amharic mixed with that. The output is complete gibberish."

Potentially Useful, but Error-Prone: ChatGPT on the Black Tech Ecosystem

KEY INSIGHTS: ChatGPT is a powerful tool that has spread like wildfire and seems like it is here to stay.But while it provides some good information, it is a

The Plug
Show threadDAIRMar 4, 2023

"This is the result of the approach that tech companies such as #OpenAI follow to build these systems. They rely on getting the most amount of data from the Web to build the biggest model possible instead of systematically curating data to solve a particular problem."

"Next time you read #OpenAI or another big tech telling you, our #AGI will "benefit all of humanity" remember this result. They burn billions of dollars on English and a handful of European languages."

Show threadDAIR

"#Tigrinya and the rest of African languages, and by extension the hundreds of millions of people that speak these languages, are an afterthought for them."

"When you compare #NLLB systems for African languages against those supported by Lesan (https://lesan.ai) or Ghana NLP (https://ghananlp.org), their systems have lower quality and are generally sub-optimal."

https://ghananlp.org

Lesan AI

Mar 4, 2023 at 8:44pmWeb
Show threadDAIRMar 4, 2023
"The #overhype from big tech is dangerous. Forget the uninitiated, well meaning people with descent expertise who don't know the details reading these headlines think tech companies have solved the machine-translation or speech-recognition problem for African languages."