<<< 2 years of ChatGPT >>>
A brief history of ChatGPT LLM progress and versions
2018: OpenAI releases GPT-1, a language model with 117 million parameters.
2019: in February GPT-2 was initially released and fully released in November, with 1.5 billion parameters, showcasing significant improvements in generating coherent and informative text.
2020: in June, GPT-3 was released with a massive model that has 175 billion parameters, showing remarkable capabilities in a wide range of tasks.
2022: In March, ChatGPT3.5 had a middle version of the model and in November, a fine-tuned version of GPT-3.5 was launched, specifically designed for conversational interactions. At this time, it was released to the public.
2023: ChatGPT4 was released in March. A subscription-based version, ChatGPT Plus, with faster responses and priority access.
In November, GPT-4 Turbo was launched, as a faster and cheaper variant of GPT-4 to power their ChatGPT products and other API services.
DALL·E 3, an image model, was included in the Chat GPT4 Plus and Enterprise, to “create unique images from a simple conversation”.
Copilot: a Microsoft partnership was introduced, firstly called Bing Chat.
2024: GPT-4o was released in August, providing JSON Structured Outputs that allows developers to define output formats more easily. Input costs reduced by 50% and output costs by 33% compared to previous models.
The GPT4-o had several models, released in September, like: o1-preview and o1-mini that are limited weekly to be 30 messages for o1-preview and 50 for o1-mini.
Additionally, Canvas was integrated into ChatGPT, facilitating real-time code collaboration and revisions, further establishing ChatGPT as a more effective creative and technical partner.
Microsoft Copilot was released in January in some countries.
Some problems and limitations that need to be mentioned
Hallucinations: It has the tendency to “hallucinate,” which means that it sometimes provides information that sounds plausible but is factually incorrect or fabricated. This continues to be a major issue.
Factual Accuracy: ChatGPT is trained on a massive dataset, as it can sometimes generate incorrect or misleading information. This can be due to biases in the training data or limitations in the model’s understanding of the world.
Bias: Despite efforts to mitigate, it exhibits certain biases, mainly political, social, and cultural, due to biases present in its training data.
Privacy: There were concerns about the security of information, especially because sensitive data that is shared during conversations with the AI. OpenAI told us that they implemented stricter privacy policies and transparency in data usage, but there is still doubt about it all alongside its users and mainstream public.
Creativity: ChatGPT can generate creative text but its creativity is limited by the data it was trained on. It may struggle to come up with truly original or groundbreaking ideas.
Contextual Understanding: While ChatGPT can understand the context to some extent, it may struggle with complex or ambiguous queries. It can sometimes generate responses that are off-topic or completely irrelevant.
Misinformation: Some advocates have expressed concern about the potential of ChatGPT to spread misinformation. This can be highly problematic in areas like politics, healthcare, and finance.
Elimination of Jobs: There is a fear that the massive adoption of AI language models like ChatGPT could lead to job losses, especially in fields that involve writing and content creation.
Ethical Concerns: The development and use of AI language models generates ethical questions, such as the potential for misuse, data privacy concerns, and the impact on society.
A broad view of the environmental impact of the LLMs like ChatGPT
Energy Consumption: Training and running large language models requires thousands of GPUs or TPUs running for extended periods, meaning large computational power, which depends on significant quantities of energy.
Data Center Cooling: Data centers often require significant energy for cooling, which can also contribute to environmental impacts. This energy consumption can contribute to greenhouse gas emissions, especially if the data centers powering these models are not powered by renewable energy sources.
Increasing Demand and Growth: As AI adoption expands, so does the enthusiasm for continuous inference of the process that is running AI models in real-time for users. This increases the energy consumption of AI operations. Every interaction with AI systems, whether to answer a question or generate text, requires computational resources.
Hardware Production: The production of the hardware used to train and run AI models, such as GPUs and servers, can also have environmental impacts, including resource extraction, manufacturing processes, and waste disposal.
Mitigation Strategies that can be adopted
Energy Efficiency: Are efforts being made to improve energy efficiency of AI models and the hardware used to run them? This includes optimizing algorithms, using more efficient hardware, and exploring alternative training methods.
Renewable Energy: Increasing the use of renewable energy sources to power data centers can help reduce the environmental impact of AI.
Hardware Recycling: Promoting responsible hardware recycling and disposal can help to minimize the environmental impact of the production and end-of-life of AI hardware.
As we reflect on the journey of ChatGPT over the past two years, we see an AI technology that has reinvented the way we interact with machines and the world around us. From its humble beginnings to growing into a global phenomenon, it has impacted the economy, education, creativity, and everyday life. But with these breakthroughs, there are substantial challenges that we must face as a society.
The future of AI like ChatGPT holds tremendous possibilities—AI can be a powerful tool for innovation, problem-solving, and creativity, empowering human beings to accomplish things we never expected to be achievable. Yet, with great power comes great responsibility (no, It is not a reference to Spider-Man).
We must be extremely careful about the ethical aspects of AI, ensuring that it is developed and deployed in ways that respect privacy, prevent bias, and protect the values of transparency and fairness. Environmental sustainability will also be crucial as we aim to balance AI’s growing computational needs along with a commitment to energy efficiency and a smaller carbon footprint.
After all, the story of ChatGPT—and AI in general—is not just about machines, models, and algorithms. It’s about how humanity chooses to handle these aspects of technology, exploiting its strengths and realizing its limitations. The road ahead is demanding for improvement, adaptation, and learning. As AI continues to evolve, so should our approach to its conscious and controlled incorporation into society.
The future is not about AI replacing the human ability to think or judge, but improving it—where machines and humans collaborate to shape a better, more informed, and livable world.
