OpenAI : DRIVING AI TO A REALITY BEYOND SCIENCE FICTION

Share on facebook
Facebook
Share on twitter
Twitter
Share on pinterest
Pinterest
Share on linkedin
LinkedIn
In only five years, the company has become one of the biggest names in AI research, capturing attention and admiration from all over Silicon Valley

We are aware of the changes coming to the world of AI. One of the biggest companies at its core is OpenAI – a research and deployment-focused company pursuing the first to create a functional artificial general intelligence (AGI). For those who are wondering what AGI is, it is an AI capable of understanding and learning any intellectual task that can be carried out by human beings. With its launch only five years ago, the company has become one of the biggest names in AI research, capturing attention and admiration from all over Silicon Valley. It was founded by none other than Elon Musk and Sam Altman, with Musk later stepping off the board and remaining only in the capacity of a donor.

 

Generative Pre-trained Transformer 3 (GPT-3) is one of the most groundbreaking releases from OpenAI. It is an autoregressive language model which produces text content based on deep learning. It processes a huge data bank of sentences in English using a highly sophisticated neural network to identify patterns and determine the rules of language that are to be used to generate text based on its findings.

Before moving on to the AI from OpenAI, a foundational understanding needs to be created about how an AGI learns to carry out its tasks. Consider how a human, specifically an infant, learns in the first place. It looks at actions and behaviours that are carried out before it and mimics them until it gets it right. The exact process applies to AGIs as well. It is exposed to the behaviours or data that it is being built for, and then it is guided to execute those actions intelligently and adaptively, taking a vast array of variables into account. The company has lionised its goal to create a friendly AI that benefits its stakeholders and humanity as a whole.
OpenAI has been praised for its spirit of transparency and benevolence, because with the examples that the world has already seen in the great abuse of intelligence and power that has always plagued the world since the dawn of civilization, OpenAI wants to be a shepherd to ensure no catastrophes arise from the emergence of AGIs.

 

 

With the kind of sensitivity surrounding OpenAI’s whole field of work, the company follows several principles that apply to every aspect of its activities:

01. Broadly distributed benefits: The company believes that they have to play the role of a not-very-silent guardian and a watchful protector when it comes to the deployment of AGIs, keeping a strong leash over any form of undue concentration of power, and ensuring that no form of harm comes to humanity from the development of AIs or AGIs. OpenAI places this fiduciary duty over all its other priorities, including even the likes of employee or stakeholder conflicts.
02. Long-term safety: Not only is OpenAI extensively researching ways to make AGIs safe, but it is also trying to make sure that such practices are replicated across the highly competitive AI industry. In a delightful and noble display of openness and camaraderie, OpenAI has mentioned that their mission would be considered accomplished if their work comes to the aid of any other entity that can make AGIs a reality before they do and that it would actively refrain from engaging in any form of unhealthy competition that can work against such goals, choosing to assist the project instead.
03. Cooperation, not competition: OpenAI believes it is imperative that instead of competing, everyone working with AIs or AGIs should come together to create a community that places heavy emphasis on research and policy development. For now, their research publication is focused mainly on building the first AGI. Still, it is likely that once this massive objective has been attained, the direction of the collective will shift toward safety and standards to safeguard it from being abused.
OpenAI’s Most Significant Achievements

GPT-3 AND OPENAI CODEX
Generative Pre-trained Transformer 3 (GPT-3) is one of the most groundbreaking releases from OpenAI. It is an autoregressive language model which produces text content based on deep learning. It processes a huge data bank of sentences in English using a highly sophisticated neural network to identify patterns and determine the rules of language that are to be used to generate text based on its findings. There are around 175 billion learning parameters within the GPT-3 model to ensure that any task assigned to it can be carried out successfully. The quality of this model is so high that it is almost impossible to understand whether a human or a machine wrote the texts.
A newer and more improved version of OpenAI Codex has also come into play, an AI system that translates natural language to code. It can receive and interpret commands in natural language from users and translate them to code in over a dozen different programming languages. GPT-3 helps generate natural language in response to a natural language prompt, directly impacting the user’s mind. In contrast, OpenAI Codex allows computers to better understand their users’ intents, which enables people to take what they can do with a computer to much higher levels.

 

OpenAI believes it is imperative that instead of competing, everyone working with AIs or AGIs should come together to create a community that places heavy emphasis on research and policy development. For now, their research publication is focused mainly on building the first AGI.

 

CLIP: Connecting Text and Images
CLIP (Contrastive Language-Image Pertaining) is a new neural network architecture from OpenAI that learns transferable visual models from natural language supervision. When it comes to creating image datasets, there are a lot of limitations, such as the restriction of only having access to specific ranges of object categories, not to mention the long and painstaking process of creating said datasets. CLIP is one of the breakthroughs in providing prediction with captions on images based on simple pre-trained models in a more robust and scalable state-of-the-art method for image recognition that uses nearly 400 million images and texts assimilated from the internet. Once the model has gone through the necessary pre training, it uses natural language processing to match its user input with the visual concepts it previously learned. In short, an image entered into the CLIP model will use all the processing mentioned above (and more) to create the most relevant caption or summary for that image.

DALL·E: Creating Images from Text
DALL·E, named after both the famous artist Salvador Dali and Pixar’s WALL-E, is an artificial intelligence program that creates images from descriptions written in text, released by OpenAI at the beginning of 2021, using around 12 billion parameters of the GPT-3 model. Images created by DALL·E can appear realistic or completely fabricated. This model differs from its predecessors in its ability to generate ideas from natural language input, which it understands with a nearly negligible failure rate. According to MIT Technology Review, one of OpenAI’s objectives when creating DALL·E was to ‘give language models a better grasp of the everyday concepts that humans use to make sense of things.

With the moonshot founded in the benevolent spirit of transparency and for the greater good, OpenAI has been making headlines with its breakthroughs and exciting research subjects, along with more fun and whimsical projects like AIs playing DOTA2 and hide-and-seek using their neural networks. If they keep making progress, it might not be too far that AGIs become a reality. Still, the question remains – despite all the safety guidelines and policies, would it be safe and be the key to bringing about a better world for the future?

Share:

Share on facebook
Facebook
Share on twitter
Twitter
Share on pinterest
Pinterest
Share on linkedin
LinkedIn
On Key

Related Posts

COMPUTE, CREATE, CELEBRATE

An overview of a four-day computer science event organised by AIUB Computer Club. AIUB CS Fest 2024, presented by the Office of Student Affairs and

LOOKS SO REAL

The exciting possibilities of OpenAI’s text-to-video AI model, Sora. Just a year ago, an AI-generated video featuring a grotesquely deformed approximation of Will Smith enthusiastically

HUMBLE LEGACY

For over 170 years, the legacy firms that make up PricewaterhouseCoopers International Limited, a multinational professional service brand of firms, operating as partnerships under the

LEADERSHIP LESSONS FROM A CXO

As a CXO, you are at a level where you have tough contenders and practical challenges. What could separate you from the rest? Here are

Leave a Reply

Your email address will not be published.