January Roundup
Hi, and welcome to January’s monthly roundup!
This is the place that I share some of my current work and reading.
AI & Copyright
The New York Times is suing OpenAI and Microsoft, for use of its articles in training GPT models. While there are several ongoing lawsuits about copyright and AI training, this is one of the most high profile. The lawsuit provides examples where AI models provide text that is almost verbatim from paywalled NYT articles. Also, it provides examples of AI model hallucinations that were incorrectly attributed to the NYT, thus creating the potential for brand damage.
Nvidia ASR Model
Nvidia released a new ASR model - Parakeet - which is currently topping HuggingFace’s ASR leaderboard. Though, it does make sense to look in detail at the breakdown over the different test sets. Whisper still performs best on AMI and Earnings22 data, which are the hardest of the sets, and has a better RTF.
Mr Bates vs the Post Office
If you haven’t heard of it, the Post Office Scandal was the result of faulty accounting software leading the Post Office to wrongly accuse and convict several of its subpostmasters of theft and fraud. Computer Weekly first broke the story in 2009 and it’s since been described as the biggest miscarriage of justice in British history. This month ITV dramatised the events in a 4 part miniseries called Mr Bates vs the Post Office. The story is not just about faulty technology, but about how companies can use technology to entrench their power, when happens when technology is used far beyond what it was designed for, and the human impact. With ethics of AI being top of mind right now, it’s a must-watch for anyone who wants to build technology responsibly.
CES
Rabbit - a device you can talk to and have it interact with your apps - caused some buzz at CES. It’s a standalone device, separate from your smartphone, which has a push-to-talk button for voice control. The demo covered some relatively simple use cases (ordering pizza and booking a taxi home), so it remains to be seen how users get on once they get their hands on the device.