DailyPost 2263
APEX OUTPUT IN APEX TECHNOLOGY – OPEN AI
Today the accepted fact that need not be reiterated is the Artificial Intelligence (AI) is the apex technology where all technologies converge. It is the beginning of new system of life, business, research, finance and literally anything that mankind does. The development of AI since it was coined at Dartmouth in 1956 has gone through a cyclical run, alternating between AI winter and AI inflection point. The current inflection point is told be a permanent one – AI is here to stay and everything sooner than later would move on to it. The datafication of world, with storage and computing capabilities breaking all barriers and with algorithms which can literally power the world, transformative Machine Learning and Artificial Intelligence is here to stay.
There are business companies which have brought technologies to the ground, handing it over to people, to elevate their lives beyond what could have imagined. This time around there seems to be similar story. Instead of host tech business companies, which are also present and making their best efforts, the lead has been taken by a non-profit corporation. Undeniably, to make the apex technology happen on mass scale, an apex organization, unlike anything in human history needed to born. OpenAI is that trailblazer organization, scripting AI for the world. Open AI is an AI research lab which consists of for profit OpenAI LP and its parent, non-profit OpenAI Inc.
The galaxy of products brining AI to the ground has made the difference. While Demis Hassabis seemed to have taken the lead, founded in 2010 making international headlines through winning games like Go against human international grandmasters. With OpenAI being founded by Elon Musk and Sam Altman in 2015, with initial $1 billion funds made available. Microsoft invested $1 billion in 2019. Products make a difference and that differentiates it from the rest of pack. The results started showing up 2017 itself, now it gives a feeling of creating products on the fly. OpenAI research tends to focus on reinforcement learning. Gym a 2017 product is wider than ImageNet Large Scale Visual Recognition Challenge, used in supervised learning research, making published research more easily reproducible.
Other products followed. Rubosomo is a virtual humanoid metalearning robot. In 2018 OpenAI launched Debate Game. This could help in auditing AI decisions and in developing explainable AI. Dactyl the next creation used machine learning to train a Shadow Hand. Generative models, GPT, made OpenAI the gamechanger. These models are based on pre-training on the corpus of world knowledge and have the potential to deliver magic. Generative Pre-trained Transformer 2, GPT-2, established unsupervised transformer language model for good. In 2019 its complete version was released. GPT-2 was a general-purpose learner. A much-improved version GPT-3 was released in May 2020. OpenAI MuseNet (2019) could predict musical notes. Multipurpose API was released in June 2020. The now fame DALL-E creates images from textual descriptions whereas CLIP does the opposite. OpenAI Microscope is a visualisation-based interpretability tool. Github educated Codex is the AI powering the autocompletion tool GitHub CoPilot. OpenAI Five and GYM Retro are other products. All combined it is a new world in the making.
AI OPEN-SOURCE TOOLS ARE OUR FUTURE! INDIA STRUGGLES EVEN WITH THEIR ADOPTION.
Sanjay Sahay