I am sure most of you noticed the trend, but a lot of folks online, from influencers on twitter/instagram/tiktok to CEOs of some companies who were big on crypto, web3 (I never understood what that even means) now switched completely to AI/ML with the popularity of DALL-E/ChatGPT. Me and some of my colleagues are fairly active on the MachineLearning subreddit and we’ve been getting a lot of unsolicited dms to collaborate in a breakthrough ML/AI project. The conversations start something like this: https://imgur.com/a/z6GUTGc Yup, you guessed it, they have the idea and we have to implement it. They don’t even have the dataset. One guy proposed my friend that he has a startup idea to use GPT-model to let people talk to their pets and that it should be ‘fairly easy’ to finetune from an existing model. I am already fairly tired of seeing all the ChatGPT stuff on my socials, and I am not looking forward to another few years of more low effort/low quality stuff in peak of inflated expectations phase. I love GPT, I have many pipelines where I actively use it, but I also see the potential where people will abuse it, in every form from increased spam, personalized phishing, etc. Imagine scammers calling your grand parents not with an non-native accent anymore – heck maybe with your own voice (which in my head is fairly easy to do – get someone’s family tree, call the grandkid using a model fine-tuned on some local accent, perhaps of the opposite gender and engage them in a conversation – use the voice clips to finetune another model and then call their parents/grandparents to get money, heck even the transcript for the scam interaction can be auto-generated). I am a first-generation college student, and getting my parents to use a smartphone has itself been a challenge – there’s no way I can teach them to identify sophisticated scams. I bracing myself and not looking forward for all of this to come.
Story Published at: January 5, 2023 at 12:55AM
Story Published at: January 5, 2023 at 12:55AM