OpenAI releases GPT-4 the world's most advanced language model
OpenAI announced yesterday, March 14, the launch of GPT-4, the world's most advanced language model that can generate videos and understand photos, according to an article published on his research blog. GPT-4 is the fourth generation in a series of natural language processing systems that have revolutionized the field of artificial intelligence. GPT-4 has a capacity of 1.5 trillion parameters, which allows it to generate coherent and fluent texts on any topic. In addition, GPT-4 has multimodal models that allow you to create audiovisual content from text or images. For example, you can write a description of a scene and GPT-4 will generate a video that realistically represents it. You can also enter a photo and GPT-4 will explain what's in it in detail.
New arrivals
Then a list of the ten most outstanding novelties Regarding previous versions:
- You can accept text and images as input and reply in text.
- You can generate, edit, and iterate with users on creative and technical writing tasks, such as composing songs, writing screenplays, or learning a user's writing style.
- You can solve difficult problems more accurately, thanks to your extensive general knowledge and problem-solving skills.
- Can process 8 times more words than ChatGPT: According to OpenAI, the GPT-4 model can reply using up to 25.000 words, instead of the 3.000 word limit for the free version.
- You can generate video from text using another multimodal model that creates it from the text generated by GPT-4.
- You can create more intelligent and coherent human texts than GPT-3. You can give it a hint and it will create a piece of text that continues.
- It is more creative and collaborative. You can generate more relevant and personalized responses based on the user's context and purpose.
- It is safer and more ethical than GPT-3. It has mechanisms to prevent the generation of offensive, false or harmful content.
- It is faster and more efficient than GPT-3. It has an optimized architecture that reduces the consumption of computational and energy resources.
- It significantly exceeds GPT-3 in the scores obtained in tests of various subjects.
Regarding the academic and scientific field:
- You can write academic essays or scientific articles on any subject, citing relevant sources and following the style guidelines.
- You can generate source code or pseudocode from a natural language description or from an input and output example.
- You can solve mathematical or logical problems, showing the necessary steps and explanations.
- You can translate texts between different languages, maintaining the fidelity and fluency of the original content.
Some references in networks
Yusuf Mehdi Corporate Vice President and Head of Consumer Marketing, Microsoft confirms that the new Bing works with GPT-4
We're happy to confirm the new Bing is running on GPT-4, which we've customized for search.https://t.co/RVj26gJVQG
- Yusuf Mehdi (@yusuf_i_mehdi) March 14, 2023
Try it out by joining the new Bing preview at https://t.co/tRrsjYDpfr. pic.twitter.com/zUCUBHa05U
New apps developed with GPT-4, such as Duolingo Max:
Duolingo partners with OpenAI
— AI Breakfast (@AiBreakfast) March 15, 2023
New AI powered features: Role Play, an AI conversation partner, and Explain my Answer, which breaks down the rules when you make a mistake pic.twitter.com/20WTUcfT96
Create HTML and CSS from a sketch
From a hand drawn, paper and scanned drawing, GPT-4 is capable of creating the HTML structure for a web page.
— 💀 WILD 💀 (@Sdesalvaje) March 15, 2023
This just started and it's already crazypic.twitter.com/qcqWVGZesT
Creation of classic games in different languages
I don't care that it's not AGI, GPT-4 is an incredible and transformative technology.
—Pietro Schirano (@skirano) March 14, 2023
I recreated the game of Pong in under 60 seconds.
It was my first try.
Things will never be the same. #gpt4 pic.twitter.com/8YMUK0UQmd
Carlos Santana demonstrates that in addition to being able to request the implementation of a video game, you can also request the implementation of an AI trained to play that game:
OK… This is cool!
— Carlos Santana (@DotCSV) March 15, 2023
I've no longer just asked GPT-4 to implement a working Flappy Bird, but I've also asked it to train an AI to learn how to play.
In a minute he has implemented a DQN algorithm that has started training the first time 🔥
(The AI playing) pic.twitter.com/g71J0KM3w6