GPT-4, OpenAI’s new neural network model powered by artificial intelligence (AI), continues to make headlines. Unlike its predecessor, the new model can work with photos as well as text. GPT-4 also handles logical and other tasks better than its predecessor, such as writing code and passing exams.
GPT-4 recently created a simple game in JavaScript. It is based on a text description that was once a meme. He also took the bar exam and got better grades than most people who have been preparing for this exam for a long time.
2 year wait game
Denis Shiryaev, founder of the Neural Love startup, decided to test the new GPT-4 features and asked him to create a neural network game in JavaScript. He showed AI a letter that a certain “Kirill” had sent to his MiST Land South studio in Russia in the early 2000s. In his letter, Kirill asked for the release of his 3D action game in which he could play as a forest elf, a palace guard, a villain, and rob caravans. According to Kirill, he wanted such a game for two years.
The letter had a lot of grammatical and spelling mistakes, but the AI got the gist of it and created a 2D game in JavaScript. The game turned out to be simpler than Cyril had imagined, but the game included the caravan he mentioned and his four different colored game zones, as Cyril requested in his letter. was
Players can control the green dots with the arrow keys and shoot them with the spacebar on the keyboard. His job is to move his point to a random location on the map after attacking the caravan.
Game code and interactive demos can be viewed on the CodePen platform.
GPT-4 passed the bar exam
After all, GPT-4 is also getting the hang of jurisprudence. The company recently scored his 297 on the bar exam conducted in an experiment with his two professors and two of his employees at the judicial technology firm Casetext. According to researchers, in most states in the United States, a person with such results would be allowed to practice the law.
About four months ago, ChatGPT also took such an exam, but failed to achieve a passing score.
The bar exam includes multiple-choice questions and essays and tests designed to simulate legal practice. GPT-4 answered almost 76% of the multiple-choice questions correctly, putting him 7% above the average of test takers on this exam. In contrast, ChatGPT answered about 50% of the questions correctly.
AI was also able to write interesting essays. This was equally surprising for Professor Daniel Martin Katz of the University of Chicago-Kent School of Law.
Follow NEWS.am Tech on Facebook, twitter
