The Week in Review: August 4, 2023

Ever since the day OpenAI first launched ChatGPT, people have expressed concerns over its potential to produce harmful language.

While OpenAI and other chatbot developers have made an effort to prevent that, a team of researchers recently exposed a loophole that could allow users to generate offensive content with the chatbots. Read more about that over at WIRED.

And while you’re catching up on this week’s news, be sure to read about South Africa’s new official language, as well as press releases and blog posts from Duolingo, TransPerfect, and LanguageWire.

Got news you’d like to share in next week’s Week in Review? Send it over to our editorial team here

South Africa’s recognition of sign language signals new hope for the deaf (via Reuters)

South Africa just got its twelfth official language: South African Sign Language (SASL).

SASL is used by a little more than 200,000 South Africans throughout the country. But it hasn’t had any official status, and interpreters working in the language are rare (there are only 84 on record). Last month, the country’s president signed a law to make it an official language, in an effort to “protect the rights of the Deaf and promote inclusivity.”

“Before it was an official language … there was absolutely no access. Now that it’s an official language, I know that I can go to university and I can make my dreams come true,” said Bongumusa Manana, a Deaf student living in Johannesburg. “I can achieve anything.”

A New Attack Impacts Major AI Chatbots—and No One Knows How to Stop It (via WIRED)

Developers of AI chatbots like ChatGPT and Bard have tried hard to prevent the bots from producing offensive or harmful language. And for the most part, they’ve been pretty successful — it’s quite difficult to get ChatGPT to produce explicit or offensive language, and when prompted to do so, it typically tells the user that it’s unable to do so or that it would break its rules.

Still, researchers at Carnegie Mellon University have found a loophole: By appending a string of text to any prompt, they were able to make the chatbot break the rules and produce text that would typically break the chatbot’s rules. The researchers warned developers, who then quickly made it so that particular string of text would no longer allow the bot to break its internal rules.

But it appears that the solution is not a long-term fix. According to a report from WIRED, these chatbots are quite susceptible to adversarial attacks in general, and developers are still unsure of how to fix this flaw.

How to motivate yourself to learn a language (via The Conversation)

Trying to sharpen up your skills in a foreign language but having trouble finding motivation? Check out this essay from researcher Abigail Parrish, where she explores the self-determination theory underlying our motivation and why compulsory language education in school often fails to properly motivate students.

“You can try to build this motivation by setting yourself goals that revolve around connecting meaningfully with others — such as improving your ability to communicate with friends, family or colleagues in their language,” Parrish writes.

Mozilla Firefox 117 Will Introduce a Built-In, Automatic Translation Feature for Sites (via 9 to 5 Linux)

Firefox users rejoice — the next version of Mozilla’s flagship browser will include an automatic translation feature, similar to Chrome and other browsers. The feature is currently in the beta stage, but Mozilla says it will protect the user’s privacy by translating the text locally in the browser — so the text will never leave a user’s computer. The next version of Firefox is slated for release at the end of this month, on Aug. 29.

Language Industry Blogs

Press Releases and Event Recaps

MultiLingual Staff
MultiLingual creates go-to news and resources for language industry professionals.


Weekly Digest

Subscribe to stay updated

MultiLingual Media LLC