DailyPost 2255
THE MODELS OF TODAY AND TOMORROW – GPT 4 AWAITED
While you would have seen models of a well-known kind, taking over the ramp in scintillating manner, others would also be talking about of role models to remodel their life. What we are talking about is something very different that is blowing off the IT world globally and with it the whole of humanity. These are the Artificial Intelligence Models, AI Models as well call them are redefining everything we do in technology. They are the models of today and tomorrow. The pace with these are being created would give us fair idea of the effort that has gone into it, for years and mostly for decades.
The launch of each of these models keeps on the completely changing the AI space. Open AI leads the way. The conviction to change mankind through their AI models is killing. In July 2022 it launched DALLE-2, a state-of-the-art text to image model. Just a few weeks later Stability.AI launched an open-source version of DALLE-2 called Stable Diffusion. A great feat that both these models have been showing promising results, both in terms of quality and ability to understand the prompt. They have been gaining popularity, but their penetration in India remains totally unknown.
Recently, OpenAI introduced an Automatic Speech Recognition (ASR) model called Whisper. This model has been able to outperform all other models in terms of robustness and accuracy. GPT has been the flagship AI model of Open AI. Large language models are in demand in the market. What is GPT in the first place? ”Generative Pre-trained Transformer (GPT) is a text generation deep learning model trained on the data available over the internet.” The popularity of GPT-3, the latest version / model is an indication of the fact that there is a clear-cut expectancy for better accuracy, compute optimization, lower biases, and improved safety from GPT-4. Now it is generally believed that GPT-4 is on the cards. In a recent Q&A session Sam Altman, CEO of OpenAI confirmed the rumors pertaining to its launch.
Undoubtedly, there are endless applications of GPT models, making them work for you is the name of the game, that would deliver you results. Have we been using them? If yes, have we fine tuning them to specific to create even better results. Or we would remain a laggard being in the plug and play mode. This does not deliver with technologies in the progress and yet to deliver at scale. People who growing and learning with this adoption would deliver even better with newer models that keep coming sooner than later. Right now with nothing much is being divulged by OpenAI about GTP-4, experts are rattling their brains to predict the model size, optimal parameter, and compute, multimodality, sparsity and performance.
LARGE AI MODELS ARE GIVING WAY TO MAKING SMALLER MODELS THAT PERFORM BETTER.
Sanjay Sahay