1. One thing that helps
Scammers often target the elderly. Now, the tables have turned. Meet dAIsy, an AI bot pretending to be a grandma in order to scam the scammers.
This AI grandmother's number is put in a list of easy targets for scammers. When a scammer calls dAIsy's number, she responds like any grandmother would and carries on a conversation with the scammer - a very long conversation. She is designed to prolong conversations and give out fake bank numbers and names all in the name of wasting the scammer's time. More time with dAIsy means scammers have less time to target us.
If you've ever been a victim of a scam, just pray that next time they are too busy with dAIsy to hassle you.
2. One to be wary of
DARPA (US Defense Advanced Research Projects Agency) has been a regular in this newsletter. This time, they have announced project "Theory of Mind" seeking to develop an algorithm to predict the actions of adversaries.
All sounds very "Minority Report" to me, trying to predict what dodgy characters will do and stopping them before they do it. Can you arrest someone for a crime they are yet to commit?
This isn't the first time that DARPA has initiated such a project. After 9/11, the US has been very keen on predicting when the next crime or terrorist act will occur. A number of programs have been launched (and scrapped) with these goals. All they need is to grab everyone's data to feed the algorithm and predict and influence behaviour. Remind you of anything?
3. One to amaze
We've been using AI translators for quite some time now and they work OK. The holy grail though has been real-time language translation with high accuracy. Meta has now released their SEAMLESSM4T AI translator that can directly translate speech to speech without going through the slow speech-to-text-then-back-to-speech process. It can translate from 101 languages and speak into 36 languages. SEAMLESSM4T is nearly 23% more accurate than its AI peers and as fast as a human translator in speed.
This feat is largely thanks to "parallel data mining" where the researchers collected 443,000 hours of audio with matching subtitles. It also trained on 4.5 million hours of spoken audio in multiple languages.
The best part? Meta released their code for non-commercial use so anyone can use it for free. If you want your next trip to be a litle more convenient, you better start coding now.