How can I find deep fakes (images that look real but are generated by artificial intelligence)?
There are often subtle artifacts, or “teles,” such as strange lighting and shadows, strange poses, and blurring of complex features such as hands and hair. It’s a little easier to detect in image space than characters. Languages use discrete units (words) whether they are valid words or not, so currently, when AI is using meaningful sequences of words, human-generated and so-called There is no good way to detect what is not. That said, sometimes these chatbots have a certain style that they understand.
You are a fiction lover. Worried that these tools will spell the end for writers?
A lot of what I love about books and stories is personal to who the author is, and that’s what the author actually wants to say, so I’m actually not too worried about it. There is a difference between a written story and a story written by an author. I also think there are many ways the author can work with her AI tools. It doesn’t have to be an either/or situation.
In fact, thinking about things like coding made me sadder. Many people, myself included, are into computer science and coding. Because I really enjoy it as a daily activity. This is going to change a lot in terms of what humans actually need to program and what can be accomplished with an AI system that interprets the natural language we speak every day.
Are you worried about the emergence of Artificial General Intelligence (AGI), an AI that surpasses human intelligence?
I wouldn’t worry too much about this idea that AGI is overall very harmful, because you first need to make a lot of human decisions wrong before you get to the point where AGI is doing terrible things. Is not …
Regulation will be critical to how AI evolves over the next decade. But I think the pace of regulation and the people who are in a position to bring about change in that space are well aware of what’s going on, and can they take that information in quickly enough to make a difference? I am concerned about what
As someone who has worked in the AI field and who, as a PhD candidate, has taken a step back to think more broadly about the AI field, what would you like to leave readers with?
First, people should be aware that systems such as ChatGPT and powerful language models will continue to evolve rapidly, and these systems will advance by leaps and bounds and perform functions that are difficult to predict.
Second, it’s important to build an intuition about how these systems work. Because they often appear human just by briefly interacting with them, and it’s easy to attribute what that system is to human intentions and reasoning and thinking. Until you start noticing subtle quirks that indicate something a little different is going on. I think it’s important to think about that as we wait for social regulation and social adaptation to catch up. As a consumer of this technology, always be cautious and skeptical and understand that this is an AI statistical model and it is not. A human agent with whom you are interacting.
Artificial intelligence, ChatGPT, and other natural language processing techniques are obviously of interest to you. do they scare you?
Technology itself is not scary to me. I am excited for all that it can do. But what worries me is how humans will use these systems and how quickly society can adapt to such major technological changes. I am concerned about the economic implications of changing to , as is the fact that a significant amount of power is consolidated in the hands of just a few tech sector players.
You worked at OpenAI in 2019 and 2020. When OpenAI rolled out ChatGPT late last year, were you surprised at how much the technology had advanced since you last worked on it?
I was not surprised at how good ChatGPT was and how appealing it was to the world. Because I remember when he was working on GPT-3 (ChatGPT’s predecessor). something really different. Since then, it has incorporated several new technologies called Reinforcement Learning with Human Feedback (RLHF). This helps the model not only have these different features in the language, but also be more consistent with what humans want in their specific responses. It’s much easier to get meaningful responses now than when I was working on GPT-3. I had to prompt the system with many examples to get the kind of response I wanted.
