ayshaaldaeki

Jul 28, 20239 min read

Who are the parents of ChatGPT's birthplace and Why did they have a divorce?

Around two weeks ago, I came across a post from The Verge on Instagram. The post discussed a lawsuit filed by several authors, including Sarah Silverman, Christopher Golden, and Richard Kadrey, against OpenAI's renowned product, ChatGPT. The authors alleged that ChatGPT and Meta's LLaMA were trained using unlawfully obtained datasets that included their works, sourced from websites known as "shadow libraries" such as Bibliotik, Library Genesis, Z-Library, and others. These books were apparently available in large quantities through torrent systems. Although the case is interesting to delve into, I want to use this opportunity to explore a more intriguing subject, which is the history of OpenAI, its origins, its identity, a brief introduction to its powerful products, and the reasons behind Elon Musk, a key founder and the most famous person for using the term “roboclypse”, parting ways with the company and forming eXplainable Artificial Intelligene (xAI), which is touted as the solution to OpenAI.

Although I'm currently conducting research on OpenAI's generative pretrained transformer models GPT-3 and the more recent GPT-4, I didn't catch up with ChatGPT during its peak popularity. Interestingly, according to UBS's research, ChatGPT is now regarded as the fastest-growing app in history, reaching 100 million users within 2 months of its launch.

A concise and relevant piece of information about AI:

Throughout history, humanity has strived to animate objects and impart them with human-like traits. However, the pivotal figure who significantly advanced this endeavor was Alan Turing (a detailed post dedicated to him is forthcoming).

Turing's research in the early 1950s laid the groundwork for modern computer science. Although AI was once confined to Sci-Fi, it captured the interest of notable minds, including John McCarthy, who coined the term "artificial intelligence" in 1956.

Two years later, McCarthy and his colleagues established the Artificial Intelligence project at MIT, signaling a promising future for AI research, albeit with some overly optimistic expectations. Unfortunately, the initial excitement gave way to an "AI winter" as funding dried up due to disappointing outcomes and limited computing capabilities.

However, in the 1990s, advancements in machine learning and natural language processing reignited interest in AI, further bolstered by well-executed publicity stunts. The early 2000s witnessed intriguing developments, such as the explosion of big data, refined algorithms, and increased computing power, paving the way for advanced AI systems.

The progress of AI in recent years has been extraordinary, with its applications spanning various fields, including medicine and education. These advancements fall under the category of Artificial Narrow Intelligence (ANI), which has limited capabilities.

There are three classifications of AI:

1. Artificial Narrow Intelligence (ANI) - Possesses a narrow range of abilities.

2. Artificial General Intelligence (AGI) - On par with human capabilities.

3. Artificial Superintelligence (ASI) - More capable than a human.

The most significant classification for us is AGI.

What is AGI?

The history of OpenAI began with AGI, a term that remains somewhat undefined in the industry. However, in recent years, a broad community of researchers has emerged, dedicated to the ambitious goals of the AI field – creating and studying software or hardware systems with general intelligence comparable to, and potentially surpassing, human abilities. The objective is to develop computers capable of thinking, perceiving, listening, walking, conversing, and even experiencing emotions.

To accurately define AGI, it is essential to first define intelligence itself. Soumil Rathi has proposed the following definition: Intelligence is an object's ability to choose the most appropriate action to achieve its goal in any given scenario, while also having the capacity to create sub-goals to achieve the ultimate objective. Consequently, AGI can be described as an artificial model equipped with human-like accessibilities and senses that can determine the optimal/most favorable course of action to achieve its goal within any environment, as well as generate sub-goals for long-term goal attainment.

The following companies are engaged in the pursuit and study of AGI:

• Anthropic

• Darktrace

• Deepmind

• Google Brain

• Hanson Robotics

• Hyperscience

• IBM

• Microsoft

• MindBridge

Among these, OpenAI currently stands out as the most prominent and leading company in the field.

OpenAI, its mission, and the history of its foundation

Now that we understand the terms, let me explain the mission and history of the renowned creators of ChatGPT. OpenAI is an AI research and deployment company that was founded in 2015 in San Francisco. It emerged as a collaborative effort involving notable figures such as Elon Musk, Sam Altman, Greg Brockman, Wojciech Zaremba, Ilya Sutskever, John Schulman, and investors like Reid Hoffman. Their ultimate goal is to ensure that artificial general intelligence (AGI) – AI systems surpassing human intelligence – benefits all of humanity.

Rather than focusing on products that mimic human voices to deceive others, OpenAI envisions creating AI products that act as personalized tutors for every child or co-pilots for professionals. The successful development of AGI has the potential to elevate humanity by increasing abundance, boosting the global economy, and contributing to groundbreaking scientific discoveries that push the boundaries of what is possible. It is also anticipated to aid in addressing global challenges such as climate change and poverty.

AGI holds the promise of granting everyone extraordinary new capabilities, leading to a world where people have access to assistance for almost any cognitive task. This could serve as a powerful force multiplier, enhancing human ingenuity and creativity.

On the other hand, AGI carries significant risks of misuse, potential accidents, and disruption to society. Despite these risks, OpenAI believes that completely halting the development of AGI is neither feasible nor desirable due to its immense potential benefits. Instead, they emphasize the need for society and AGI developers to find ways to handle AGI responsibly.

OpenAI advocates for a gradual transition to AI, allowing individuals, policymakers, and institutions time to comprehend the unfolding developments, personally experience the advantages and drawbacks of these systems, adapt the economy, and establish appropriate regulations. This approach enables society and AI to evolve in tandem, giving people the opportunity to collectively determine their preferences while the risks are relatively manageable.

In addition, to uphold their mission, OpenAI goes beyond creating software for users and recipients and makes their API accessible. This move empowers people to leverage their technology, create similar or even more advanced software applications, and actively participate in shaping the AI landscape.

OpenAI offers several products, including:

1- GPT-3, GPT-4, and ChatGPT:

GPT-3 (Generative Pretrained Transformer 3) and GPT-4 are cutting-edge language processing AI models developed by OpenAI. They are neural networks trained on a vast dataset of 300 million words, known as large language models (LLMs), with the objective of predicting the next word in a sentence. These models are capable of generating human-like text and find applications in various areas, including language translation, language modeling, and text generation for chatbots. GPT-3 is one of the most extensive and powerful language processing AI models to date, boasting 175 billion parameters.

The primary use of GPT-3 is for creating ChatGPT, a highly capable chatbot equipped with functionalities such as answering questions, language translation, text generation, summarization, correction, sentiment analysis, conversational AI, programming assistance, and general knowledge.

GPT-4 builds upon its predecessor, GPT-3, with new features that enhance its capabilities. Notably, GPT-4 allows for a significant increase in the number of words that can be used in an input, up to 25,000, which is eight times more than the original ChatGPT model. It can also process image and potentially video input. Additionally, GPT-4 shows improvements over GPT-3.5, being 82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses.

However, these models have limitations. They struggle with handling very recent concepts, and their knowledge about world events from the past year can be limited, occasionally leading to false or confused information. Moreover, given the data from the internet and available resources they were trained on, ChatGPT may produce biased or harmful content.

One of the significant challenges faced by OpenAI with this product is the occurrence of "AI Hallucinations." These instances happen when models like ChatGPT fabricate entirely fictitious information, behaving as if they are presenting factual data, a situation that I, shamefully admit, was a victim of once.

The original goal of creating ChatGPT was not to predict the next word reliably or safely but simply to predict it. As Mira Murati points out, this broad task makes it challenging to handle certain limitations, and some of the texts and data used for training may be biased or incorrect.

2- DALL·E:

DALL·E 2 is an advanced AI system capable of generating lifelike images and art based on natural language descriptions.

The primary and most impressive feature of DALL·E 2 is its image generation capability. By processing a text description, DALL·E 2 can produce original and realistic images and artwork, combining various concepts, attributes, and styles. Something that is quite remarkable and I think might be hard for people to fully grasp is how a model trained on existing data can now create visual art that is entirely unique and authentic.

Additionally, DALL·E 2 possesses other features related to existing images:

1. Outpainting: allows it to extend images beyond the original canvas, creating expansive new compositions.

2. Inpainting: enables realistic edits to existing images based on natural language captions. It can add or remove elements while considering factors like shadows, reflections, and textures.

3. Variations: allow DALL·E to take an image and generate different artistic variations inspired by the original.

Moreover, the API for DALL·E is accessible to the public, leading to the creation of numerous online applications that utilize natural language to produce images, paintings, or designs. OpenAI's vision in developing DALL·E is to empower individuals, including those lacking specific artistic skills, to express themselves creatively.

3- Whisper:

The product that I personally have tested with its API most, Whisper is a versatile speech recognition model capable of transcribing, identifying, and translating multiple languages. From my personal experience, Whisper has demonstrated exceptional accuracy, particularly in comparison to other speech recognition models, especially when it comes to the Arabic language.

How the mission of OpenAI underwent changes, leading to Elon Musk, a crucial founder, parting ways with the company and forming xAI as an opposition.

OpenAI was established in 2015 as a nonprofit research organization, co-founded by Altman, Elon Musk, Peter Thiel, and Reid Hoffman, along with other tech leaders. In its founding statement, the company pledged to advance digital intelligence in a manner that benefits humanity as a whole, without being driven by financial gains. The initial vision emphasized openness, with a focus on positive human impact, and researchers were encouraged to share their work freely.

However, over the years, critics, including co-founder Elon Musk, argue that OpenAI has deviated from its original mission. They claim that the company now prioritizes speed and profit over transparency and positive impact. Elon Musk tweeted “OpenAI was created as an open source (which is why I named it “open” AI), non-profit company to serve as a counterweight to google, but now it has become a closed source, maximum-profit company effectively controlled by Microsoft.”

In response, Reid Hoffman, an early investor, and LinkedIn co-founder, defended OpenAI, asserting that it remains a nonprofit 501(c)(3) organization. He explained that the decision to become for-profit was driven by the need to secure investments to expand and improve their transformer technology. Surprisingly, the CEO and co-founder of OpenAI, Sam Altman, does not possess any equity in the company, despite its estimated value of 30 billion dollars. He does not receive any financial compensation for his contributions to the organization.

Regarding it being open, Hoffman mentioned that while they didn't release software like DALL·E as open source, they offered it through an API to prevent potential misuse; “so that they would make sure that the model wouldn’t be altered by others to create child sexual material, assaulting individuals, or doing deep fakes.”

The company faced criticism when it unveiled its GPT-2 language model in 2019. Initially, the company announced that it would not release the training model's source code, citing concerns about potential misuse of the technology. While this decision reflected their commitment to developing AI for positive purposes, some saw it as not fully adhering to an "open" approach. Critics questioned the reason for announcing a tool and then withholding it, viewing it as a mere publicity stunt. However, three months later, the company changed course and made the model available on the open-source coding platform GitHub, stating that this move was essential for responsible publication in AI, especially concerning powerful generative models.

The current concerns related to OpenAI's products on their path towards AGI primarily revolve around the generation of harmful content and misinformation by chatbots. There have been instances where prompts led to the creation of dangerous materials like "how to build a bomb" or code for cyberattacks. To address this, OpenAI is taking a cautious approach, releasing the technology in a controlled manner to study how people can lead it to negative outcomes and adapting accordingly to mitigate risks.

Another concern is that it is increasing people’s laziness, especially students using ChatGPT to write papers or essays. Sam, the CEO, acknowledges that this raises concerns about academic integrity and student reliance on technology. However, he also sees an opportunity that will transform the way we teach and education in its essence. What excites him the most is the technology's capability to offer personalized learning for every student. Sam believes that, just like the calculator changed testing methods, this technology can revolutionize education and make students more capable than ever before. “While it may be challenging for teachers to deal with students using chatbots for writing, it can be a relief when a teacher is able to say to a student go use ChatGPT to understand this concept that you’re struggling with.” Says Sam.