100
The year 2022 saw a host of brilliant inventions, ranging from the James Webb Space Telescope, the HIV vaccine, and DALL-E, to transparent solar windows. But just when we thought the year had seen its fair share of excellence and innovation, ChatGPT caught us by surprise.
Developed by the OpenAI foundation, an independent research body founded by Elon Musk, the Artificial Intelligence-powered chatbot released on November 30, is everything mind-blowing, bizarre, and daunting. The San Francisco-based company is also responsible for the breakthrough image generator DALL-E 2.
The real-world applications of ChatGPT are more than one can fathom
It has even been termed as a potential 'Google slayer' and a 'Code Red' for the internet search engine. Why? Well, there is pretty much nothing that ChatGPT can't do. Want to compose music out of thin air and impress your friend? ChatGPT is there for you. Maybe you need a helping hand with that code? ChatGPT to the rescue. It can summarize topics, give out recommendations, generate keywords, translate languages, solve crossword puzzles, and explain scientific concepts...need we say more? It also does look like your brain cells might be out of a job.
This is number 4 in Interesting Engineering's series showcasing the best innovations of 2022. Check back to discover more about groundbreaking AI, unique solar panels, new 3D printing methods, and much more.
Potential applications can extend beyond coding and summarizing. A new study from Drexel University's School of Biomedical Engineering, Science, and Health Systems, revealed that the GPT-3 algorithm can identify clues from spontaneous speech that are 80 percent accurate in the prediction of early stages of dementia.
ChatGPT/OpenAI
ChatGPt evolved from text-generating AI
How does the algorithm work? ChatGPT evolved from the Generative Pre-Trained Transformer(GPT) family of text-generating AIs and is OpenAi's fix to GPT-3. It is trained by AI and machine learning and is a "sibling model" to InstructGPT, which, as the name suggests, follows the instruction to respond.
According to OpenAI, the model was trained using Reinforcement Learning from Human Feedback (RLHF), with the same methods as InstructGPT. An initial model was trained in which human AI trainers provided both-side (the user and an AI assistant) conversations. As per a statement from the company, they mixed the resultant dialogue dataset with the Instruct GPT dataset. It was then transformed into a dialogue format.
"The dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests," the research body said in a statement.
ChatGPT/OpenAI
Can ChatGPT steal our jobs? Some think so
In the days after ChatGPT was released, there was growing speculation that several professions could be outmoded, including programmers and content writers. Fortunately, that won't be the case. ChatGPT, 'like humans', can give out wrong answers and repurpose misinformation. Fixing such incorrect answers is challenging as during the RL training, there was no source of truth. Training the model to be more careful can also result in it declining questions that could be answered correctly.
ChatGPT's knowledge is also pretty much restricted to things it learned before 2021. However, it is undeniable that the AI Chatbot is the smartest one ever seen by the public. And it isn't even OpenAI's best AI model, which would be GPT-4, releasing in the next year or so.
This is number 4 in Interesting Engineering's series showcasing the best innovations of 2022. Check back to discover more about groundbreaking AI, unique solar panels, new 3D printing methods, and much more.
