Cutting Arabic errors and developing drugs faster

Cutting Arabic errors: Nvidia cut its speech-recognition model’s error rate on Saudi dialects to 30% from c. 55% after training it on more than 100 hours of Najdi and Hijazi speech, Arab News reports. The audio comes from a 667-hour dataset developed by Saudi Arabia’s national data and AI authority, SDAIA, and the Saudi Broadcasting Authority. The dataset spans more than 10 Saudi dialects.

Cracking down on model copying: OpenAI says it stopped people linked to Moonshot AI, the Chinese startup behind rival chatbot Kimi, from using its models to help improve competing AI systems. More via CNBC.

AI speeds up the drug hunt: Abu Dhabi-based Insilico Medicine says AI is speeding up its timeline for selecting cancer drugs for further testing to as little as 12-18 months from a previous 2.5-4.0 years, The National writes.

Drug development is, by all accounts, a very boring process. But it’s made for some gripping film scenes. In 1993, And the Band Played On made the tale of developing the first antiretrovirals a key component of the story of the first years of the AIDS crisis. Then there’s 2010’s Love and Other Drugs, a pretty decent romcom set in the morally dubious world of pharmaceutical sales. But, for our money, there’s nothing better than Harrison Ford as Dr. Richard Kimble in The Fugitive uncovering the pharmaceutical development scheme that led to his wife’s murder.

That’s it for today. We’ll be back in your inbox Thursday and then every Tuesday and Thursday after that.

We’re excited to have you along for the ride — feel free to invite some friends. Stay in touch with the team at [email protected].