Techspace - OpenAI's brainchild ChatGPT has shook the world with its ability to answer meaningful questions on almost anything, is now bound to be in the spotlight once again. Its parent company, OpenAI, has just unveiled the newest iteration of ChatGPT, GPT-4. Users testing it have reported that this newest iteration is capable of so much more compared to its older iteration.
It seems that OpenAI intended ChatGPT and also GPT-4 to be the "The" AI when it comes to generating language data and conversations. The earliest iteration, InstructGPT, couldn't care less about the logical contexts of requests such as "tell me about when Christopher Columbus came to the US in 2015", would just answer the question as is.
This changed when its newer iteration, ChatGPT came to be. Not only did it recognize that something's wrong with said request, but it went so far as to entertain the idea of what Columbus would think when he sees the US in its current, modern state. With its successor, GPT-4, came forth newer abilities to generate more complex language data such as drafting lawsuits, working on standardized exams, and even building a working website from a hand-drawn sketch.
Newer abilities GPT-4 has that its predecessor doesn't
Trained on vast amounts of online language data, ChatGPT is sophisticated enough to answer requests/questions about coding and even philosophy. Its successor, GPT-4, is being equipped with more capabilities and could dwarf ChatGPT's, which could already understand natural human language and generates detailed and human-like texts.
GPT-4 has a new core capability which its earlier iteration doesn't: working off of an uploaded image. In a video demo from its parent company, OpenAI, GPT-4 showed how it could build a working website just based off of a drawing. This drawing was hand-drawn, uploaded to GPT-4, and then generated the code into a preview showing how this website could work.
Not only that, GPT-4 can also analyze a joke from a series of images. Furthermore, it could also when requested explain how the joke is funny. From a linguistic standpoint, this is a major breakthrough. Supposedly, an AI can now understand contexts so much that it can figure out why a joke is funny.
Years ago, there's an argument that AI could never replace a human because it is devoid of the ability to understand contexts. Now, it seems that AI is getting closer and closer to having that ability, too.
GPT-4 also has the ability to make coding easier for people who've had little to no know-hows of it. Some of its users requested that the AI guide them to recreate older classic games such as Tetris and Snake, with success. Other users have even made original games with help of the AI. According to its parent company, GPT-4 can write code in all major programming languages, as well.
OpenAI also simulated a law school exam for GPT-4, which it passed with a score in the top 10% percentile of all test takers. This is a major improvement compared to its predecessor who scored in the bottom 10%, reportedly. In other tests such as the LSAT, GRE, SATs, and many other AP exams, GPT-4 also performed strongly. The newest iteration of the AI can also generate longer language data up to 25,000 words.
Several companies have admitted that they're already using GPT-4 to perform various detailed tasks that are within its capabilities. Considering that its capabilities are varied and advanced enough, there's still a lot of potential the AI could bring to the digital ecosystem.
However, OpenAI themselves have admitted that GPT-4 does have its own limitations, namely in making simple reasoning errors or being so gullible. In spite of all that, this new development, as was the case with the advent of ChatGPT last year, has brought about newer worries on whether AI will make certain jobs irrelevant and push people out of work.
Read more:ChatGPT Creator Shocked Everyone With A New Tool To Detect AI Generated Text
ChatGPT Creator Shocked Everyone With A New Tool To Detect AI Generated Text
Leave a comment