AI’s Great Leap Forward:
The year 2023 has seen AI evolve and expand like never before, with investment in AI hitting record levels. Major tech industry giants such as OpenAI, Google, Meta, and their smaller rivals, have become significantly better in a matter of months. More improvements are expected by the end of 2023, especially coming from chatbots and other AI-powered solutions.
A striking way to gauge progress is by comparing how chatbots behaved in 2019, 2021, and 2023 or by examining the evolution of AI-generated images from their pixelated beginnings to their current sharpness.
Assessing this progress is straightforward:
1. Improved Models: AI now benefits from better mathematical models.
2. More Computing Power: Enhanced computing power fuels AI’s growth.
3. Robust Processors: AI systems leverage more potent processors.
4. Quality Data: Access to comprehensive, high-quality datasets is driving AI advancements.
It’s evident when you compare how chatbots differed in their responses in 2019, 2021, and 2023. Or how images created by AI software have transformed from being pixelated three years ago to being crystal sharp today.
Several factors have led to the swift improvement of AI software over the past year including superior mathematical modeling, increased computing power, more sturdy processors, and access to complete and high-quality datasets.
Training AI: The New Model
In the last two years, the progress made by AI software has been overwhelmingly significant. This is largely due to a new model designed to “train” AI, known as Transformers. This model allows for the rapid processing of colossal blocks of language while simultaneously allowing for testing and validation of the outcomes.
Transformers was first introduced by Google researchers in 2017. These are natural language models, which, just five years ago, were majorly used to solve natural language processing issues like machine translation, generating texts, or responding to queries.
Image-generating AI tools like Dall-E and Midjourney have implemented a training method referred to as “diffusion”. This involves uploading billions of images, each accompanied by descriptions, allowing the computer to identify the correlation between images and words. Following this, the system also learns to recognize and eradicate imperfections from pictures, resulting in images that are as accurate and lifelike as possible.
The Incredible Rise of GPUs
Significant investments in Graphics Processing Units (GPUs) have been a driving force behind the rapid evolution of AI systems. GPUs can process multiple operations simultaneously, a crucial requirement for training and optimizing large datasets.
Nvidia, a prominent GPU manufacturer, has played a pivotal role in advancing AI technology. Initially designed for gaming, Nvidia GPUs found a new market in AI, enabling the rapid advancement of AI systems. Furthermore, virtual linking of multiple GPUs in the cloud has democratized access to immense computing power, even for smaller companies.
Data Quality, Expert Talent, and Operating Costs
The progress witnessed in the last two years was only possible due to the availability of larger, better-labeled, and more comprehensive datasets. Unlike earlier times, today’s datasets encompass everyday conversations, books, and news articles, reflecting how people naturally communicate. Training AI models on such diverse datasets enables them to provide responses that closely resemble human interactions, making AI more relatable and accessible.
Additionally, top talent in AI development, often referred to as “star researchers,” joined ambitious companies with substantial budgets for salaries. The influx of these experts played a critical role in pushing AI technology to its current state. These experts often command salaries comparable to performance athletes, underscoring the importance of talent acquisition in AI development.
Given the multitude of factors involved, including hardware, software, data, and personnel, it’s not surprising that developing and perfecting chatbots comes with substantial operating costs. Some reports indicate that operating costs for chatbots like ChatGPT are significant and even result in losses for the companies behind them.
As a result, companies offering generative AI products must find revenue streams, often reserving the best-performing software for paying subscribers. Free versions, while available to the public, may not offer the same level of performance due to high computing costs.
For instance, with GPT-4, a subscription of $20 per month allows users to send a maximum of 25 messages every three hours, reflecting the challenges of balancing performance and costs. Estimates suggest that ChatGPT’s daily operating costs could reach $700,000, highlighting the substantial financial burden associated with maintaining advanced AI systems.
The Future of AI and Concerns
OpenAI has confirmed that it has not yet initiated training for GPT-5, the successor to GPT-4, and it’s not planning to do so in the near future. This deliberate approach is characterized by thorough testing and auditing, ensuring that AI advancements align with ethical standards.
The concerns of AI creators and experts regarding the future of AI stem from several factors:
- Misuse of Technology: There’s apprehension that AI technology could fall into the wrong hands, such as terrorists, hackers, or authoritarian governments. The increasing unpredictability of AI raises concerns about its potential misuse.
- Loss of Human Control: AI systems, as they become more advanced, might surpass human control. The fear is that autonomous AIs could make unpredictable or unintentional decisions that may harm individuals or society.
- Bias and Discrimination: AI systems, trained on existing data, can inherit and amplify biases and discriminations present in that data. This can lead to unfair or discriminatory decisions, particularly in sensitive areas like recruitment and criminal justice.
- Impact on the Labor Market: As AI becomes capable of handling complex tasks, concerns arise about job displacement, affecting both manual and intellectual labor.
- Ethical and Liability Issues: Decisions made by AI systems can have tangible consequences, especially in fields like health, security, and data privacy. Ethical and liability concerns regarding AI decisions are becoming increasingly important.
- Data Privacy and Surveillance: The extensive use of AI involves the collection and analysis of massive amounts of personal data, raising concerns about data protection, surveillance, and its implications on individual privacy and freedom.
