By Ismail Ismail Tijjani
Large Language model (LLM) is a subset of Generative AI that focuses on generating human-like text based on the input it receives. Evidence shows how good it is in generating creative text formats, like poems, code, scripts, musical pieces, emails, letters, etc. Chatgpt plays a significant role in bringing LLMs to wider public attention, though it wasn’t the first one. I will use Chatgpt throughout this article because of its popularity, though there are other popular ones like Gemini, Bard.AI, LaMDA and much more.
Let me provide a very simple description of how ChatGPT works. Just imagine you enter a library and ask the librarian a question. The librarian will first try to understand your question and then scan the shelves, looking for books they think might contain your specific answer. Using their records and expertise, they connect related stories from different books and give you the best possible answer. This is what ChatGPT does in a few seconds.
ChatGPT underwent training using an extensive and diverse internet dataset, which covered a wide spectrum of content ranging from different subjects, styles, and perspectives. Its core skill lies in tranformer architecture, a neural network which is primarily designed for language processing to encode an input text, analyze its structure and meaning, and decode it to produce an output by predicting the most likely next word in a sequence.
Certainly, the introduction of new technologies often sparks heated debates. Critics often strive to oppose and even reverse these advancements. However, their efforts typically falter in the end. Some critics may genuinely misunderstand the technology, while others, perhaps a majority, are driven by the pursuit of publicity rather than accurate assessment.
ChatGPT wasn’t an exception. When it was launched in October 2022, some people argued that it would make students lazy, lead to job loss for editors, result in high plagiarism, breach copyrights, steal people’s data, exhibit sentiment, intentional bias, spread disinformation, create deep fakes, and much more. We will discuss some of these concerns below. Some of the allegations are true and have already been addressed, while others are false.
ChatGPT lacks the ability to discern whether information is biased, disinformation or false. It operates based on its programmed structure and produces results accordingly. OpenAI, the creator of ChatGPT, has taken measures to enhance the model. Although the technical details haven’t been disclosed, it likely involves implementing guardrails and filtering mechanisms to address accusations of misinformation, bias, falsehood and more.
For students and researchers, ChatGPT will serve just like an advanced internet search engine that will generate output after going into multiple webpages, saving time and stress hopping between sites. It will in no way make students lazy, However, some concerns related to students’ use will be discussed in a later article.
Its ability to remember previous prompts, though impressive, doesn’t imply sentience. It’s merely a clever technique within its architecture. This raises separate questions about the path to achieving true artificial general intelligence (AGI), discussed in my previous article. It is only a technique in tranformer that makes it capture it.
The impact of ChatGPT on jobs is a complex and nuanced issue with both potential downsides and upsides. While some job losses are inevitable, they will likely be offset by the creation of new ones. Adapting to this changing landscape through education, reskilling and responsible policymaking is key to ensuring a future where AI benefits everyone.
Despite the evident improvements seen since ChatGPT’s initial release, OpenAI must continue to dedicate significant resources to refining its model. This is crucial not only to mitigate legal risks but also to enhance its accuracy and reliability for responsible public use. While striving for absolute perfection is unrealistic, reaching a consistently high level of trustworthiness should be a priority. Additionally, users must be mindful of the model’s limitations and exercise critical judgment, fact-checking, and verification before relying on its output.
AI is here for good. Innovation often sparks a variety of perspectives, and AI is not an exception. Some people believe that AI has the potential to solve some of the world’s most pressing problems, such as poverty, hunger, climate change, corruption and disease. Others are concerned about the potential for misuse, such as the development of autonomous weapons systems or the use of AI to manipulate people.
AI is not like any other innovation we have seen before in the history of humankind. It is among the most powerful of all, and it is likely to be among the last innovation that we ever need.
AI is already making significant positive impacts in various industries, such as healthcare, finance, retail, manufacturing and many others. Of course, like any industry, there may be individuals with malicious intent in the AI sector who are willing to exploit it for negative purposes. For example, I recently came across a women-uncovering app with high precision that raised concerns. However, the actions of these individuals should not lead to the shutdown of the entire industry.
We don’t shut down the arms and weapons industry because of terrorists, the financial industry because of fraud, the biotechnology industry because of bioweapons or social media because of misinformation and hate speech. Instead, we regulate them by establishing governing bodies to oversee their operations and foster collaboration between top companies, stakeholders, researchers and the government to develop effective solutions. This approach will also be applied to AI.
Our primary focus should be on humanity. It is crucial for everyone to actively participate and collaborate in order to develop effective solutions that will propel us and the industry forward as a unified whole.
Path to Artificial General Intelligence (AGI) 1: the year 2023 stood out as an exceptional period in the AI industry, marking a significant moment when the masses truly connect with the essence of AI.
AI has been around for years, primarily utilized in backend functions like relevance ranking, personalization, spam detection, and more. ChatGPT was the remarkable innovation that astonished the world, revealing the true potential of AI. While it may not have surprised researchers who had already witnessed AI capabilities in the lab, its impact on a broader audience is undeniable to the extent that many non technical individuals use the words AI and ChatGPT interchangeably, thinking its same thing.
Other notable innovations in 2023 include Hugging Face, Google Bard, Capcutand many others, all these are great innovations we have seen in 2023.
Are these innovations clearing a path to Artificial General Intelligence (AGI), which informally means machines reaching human-level intelligence? This question remains unclear to researchers, as there are two camps with differing opinions on the matter. Some believe AGI is imminent, while others hold a more skeptical view.
Yann LeCun the chief AI scientist in Meta is among those with skeptical perspective, he believes that Language models like ChatGPT that people are using as an evidence that AGI is imminent are not as smart as a cat which is the truth and believes it will happen in decades or even a century — a point of view that I share.
For machines to achieve true intelligence, they must possess both cognitive and metacognitive abilities. While significant advancements have been made in cognitive AI, bridging the gap to metacognitive intelligence remains the key barrier. Researchers are diligently seeking solutions to overcome this challenge. For machines to be metacognitive intelligent, which necessitates the ability to sense the environment and effectively process and interpret sensory signals. Our discussion was focused on the process of it being intelligent alone, not as intelligent as a human being, which is the AGI. This clearly shows that we are nowhere near AGI.
The timeline of AGI is not only a matter of time but rather depends on the speedy research and innovation advancement. Improvement in advanced neural networks, symbolic reasoning, embodied cognition, unsupervised, and reinforcement learning will make a great impact in clearing the path to AGI.
The path to AGI is not a solitary trek for AI researchers. It demands a symphony of minds, where scientists, physiologists, engineers and other researchers from diverse fields join hands in a grand collaborative effort. Only through their combined expertise and tireless dedication can we hope to unlock the secrets of true machine intelligence.