The autumn of the AIs: ChatGPT, GPT-4 and the global concern
Originally published in Linkedin at: 30 Mar, 2023.
March has been a very busy month in many respects. 300 million jobs could be affected by the latest wave of AI [24], world experts are calling for a halt to AI testing [25] by calling for a moratorium until new regulatory authorities establish security systems, among other things. The concerns that arise from this are misinformation [26], cyberattacks, and automated decision-making.
However, globally, several crises are happening simultaneously, such as the bankruptcy of Silicon Valley Bank [2] and Signature Bank (at the time of writing this article). These crises have affected both Latin America [3] and Europe [4]. Additionally, as a result of a domino effect, Swiss bank Credit Suisse nearly declared bankruptcy and had to be absorbed by UBS to try to curb the banking crisis [5].
In particular, in France, Macron is facing a situation similar to what Piñera experienced in October 2019 in Chile, by increasing the retirement age from 62 to 64 years through a decree without prior debate in parliament. This triggered a series of protests by workers and students against him [6][7][8].
Meanwhile, in the war between Ukraine and Russia, China tried to be a mediator for peace, but both the West and Kiev rejected this by putting “the aggressor and the aggrieved on the same plane” [9]. Although this can be interpreted in several ways, we must not forget the arrest warrant against Putin [10] and the current conflict between China and Taiwan, making the current military context very complex.
We are living in difficult times and it would be utopian to discuss technological issues without being aware of how the world is outside of that realm, including the economic, political, and environmental context. Regarding our main topic, we will continue with the same line as the previous ChatGPT article (which I recommend reading as a foundation for this piece) and delve into the latest version released on March 14th: GPT-4 [1]. Additionally, we will also analyze the various topics that authors have addressed in papers concerning both ChatGPT and the new model during March 2023.
Current virtues and sins of ChatGPT
In March, the news broke that Salesforce had announced Einstein GPT, a combination of the current Einstein and ChatGPT models, becoming the first generative AI for CRM [22].
Turning to current formal literature: in a recent paper [11], it is highlighted how ChatGPT can be utilized for environmental research, for example, by using Python to predict daily concentrations of organic carbon aerosols or to obtain general definitions such as direct extraction of lithium. However, the caveat is that the information provided by AI may be false, fabricated, or not up-to-date, leading to the conclusion that humans should remain the authors while AI should remain as assistants.
In [13], several practical examples are provided in C, C++, Python, MATLAB, Julia, and other programming languages for numerical algorithms involving problems such as Poisson’s equation, matrix eigenvalues, linear systems of equations, and sparse matrices. It was possible to successfully program numerical algorithms for different languages, but there are limitations and challenges to consider, such as the inability to generate singular matrices or stop execution when a very long code is being generated.
In the medical field, ChatGPT was evaluated in a study [16] to provide recommendations for the management of gliomas by a panel of experts in central nervous system (CNS) tumors. The conclusion was that ChatGPT performed poorly in classifying gliomas, but was effective in recommending treatments, making it a useful assistant for physicians treating brain tumors.
In [17], it is also used for medical questions, where it is noted that the answers can be misleading and limited. In [18], they address a rare syndrome of choriocarcinoma with intracerebral hemorrhage of brain metastases in a patient with mixed testicular cancer, where they use ChatGPT as support in their research. In [19], they evaluate whether ChatGPT can improve communication within hospitals, highlighting its diverse functionalities but noting that it does not replace humans and must be used with care.
In [14], the ethical aspects of ChatGPT are explored, highlighting concerns regarding privacy, bias, and ethical decision-making. Meanwhile, in [15], it is suggested that ChatGPT has the potential to violate academic ethics, and the limited access to technologies that can detect such violations is a significant challenge for academic writing. Finally, in [23], the role of AI in teacher training is studied, which has the potential to revolutionize this field but also has ethical considerations to be taken into account, such as privacy, bias, and cultural acceptability.
GPT-4
“GPT 4 is the most notable version yet, representing the greatest improvement since the initial release back in June 2018. Thanks to these enhancements, ChatGPT is now more creative and collaborative than ever before, capable of solving difficult problems in just a moment.” [20]
Capable of accepting both images and text as input and delivering text as output, the entire deep learning stack was rebuilt and, in collaboration with Azure, a supercomputer was co-designed from scratch for the workload. Furthermore, some errors were identified and corrected, and the theoretical foundations were improved.
How to use ChatGPT-4? I recommend checking out this article [21]. The state of the art bodes well for GPT-4: in [27], it is demonstrated that it can solve difficult tasks in math, code development, medicine, law, psychology, and more, but it concludes the same limitations in GPT-4 as other previous works, such as misinformation and bias. In [28], its versatility is perpetuated by successfully tackling medical issues, even mentioning that it was better calibrated than GPT-3.5.
Finally, in [29], a comparison is made with GPT-3.5, which highlights a notable improvement over its previous model, but still maintaining certain limitations: hallucinations, violent content (hate speech or incitement to violence), misinformation, and influence on opinions.
Limitations of GPT-4
According to OpenAI [1], it has similar limitations to early GPT models. It is not yet fully reliable (although it still has hallucinations, like in the GPT-3 model, they have been reduced), and there is still work to be done in addressing bias in the results. Care should also be taken with events where data is insufficient, and it does not learn from its experience.
It can even be gullible in accepting false statements from a user and make mistakes in difficult problems, such as introducing security vulnerabilities into production code.
Conclusions
Despite the global context, March proved to be a busy month for AI and ChatGPT in different areas, where their use expands into multidisciplinary fields such as environmental research, solving complex mathematical problems, medicine, and the ethical considerations of using them as tools or assistants. Additionally, the latest GPT-4 model was released, which is now capable of processing both images and text, and comes with various improvements at the system level, opening the doors to a greater variety of future uses while still retaining limitations and restrictions.
In general, most of the reviewed papers conclude that ChatGPT should be used as a tool or assistant for any task, given its current state. Emphasis is also placed on the need to use this tool in an ethical manner, avoiding plagiarism and other unethical behaviors that have been observed recently. Therefore, it is expected that a good professional will be aware of when and how to use this type of AI in an ethical way.
Bibliographic references
[1] OpenAI (14 Mar, 2023). GPT-4 is a large multimodal model that, while less capable than humans in many real-world scenarios, exhibits human level performance on various professional and academic benchmarks. https://openai.com/research/gpt-4 (Accessed 28 Mar, 2023).
[2] Tabby Kinder, Antoine Gara, Joshua Franklin, George Hammon (12 Mar, 2023). Silicon Valley Bank: La espectacular caída del banquero de la industria. Diario Financiero. https://www.df.cl/internacional/economia/silicon-valley-bank-la-espectacular-caida-del-banquero-de-la-industria (Accessed 28 Mar, 2023).
[3] Gwendolyn Ledger (15 Mar, 2023). Así se ve la quiebra del Silicon Valley Bank en Latinoamérica, una semana después. America Economia. https://www.americaeconomia.com/asi-se-ve-la-quiebra-del-silicon-valley-bank-en-latinoamerica-una-semana-despues (Accessed 28 Mar, 2023).
[4] Mathis Richtmann (15 Mar, 2023). El impacto en Europa de la quiebra del Silicon Valley Bank. DW. https://p.dw.com/p/4OjBc (Accessed 28 Mar, 2023).
[5] Mark Thompson (19 Mar, 2023). UBS comprará el Credit Suisse para frenar la crisis bancaria. CNN. https://cnnespanol.cnn.com/2023/03/19/ubs-comprara-el-credit-suisse-para-frenar-la-crisis-bancaria-trax/ (Accessed 28 Mar, 2023).
[6] Associated Press (23 Mar, 2023). Crece el descontento con Macron entre nuevas protestas. La Tercera. https://www.latercera.com/mundo/noticia/crece-el-descontento-con-macron-entre-nuevas-protestas/ONYTYYZJKBD3ZLO7UACG45EBZ4/ (Accessed 28 Mar, 2023).
[7] Bernd Riegert (23 Mar, 2023). Cómo enfrenta Macron la actual crisis y las protestas en Francia. DW. https://p.dw.com/p/4P8z8 (Accessed 28 Mar, 2023).
[8] Marc Bassets (23 Mar, 2023). Los incidentes violentos empañan el éxito de las protestas pacíficas contra la reforma de Macron. El País. https://elpais.com/internacional/2023-03-23/la-movilizacion-contra-macron-mide-sus-fuerzas-tras-la-adopcion-de-la-reforma-de-las-pensiones-en-francia.html (Accessed 28 Mar, 2023).
[9] DW Español. Visitas de China a Moscú y de Japón a Kiev reflejan el nuevo alineamiento global. Youtube, 22 Mar, 2023. https://www.youtube.com/watch?v=fYVxh2OGI10 (Accessed 28 Mar, 2023).
[10] CNN (17 Mar, 2023). La Corte Penal Internacional emite una orden de detención contra Putin y una funcionaria rusa por la supuesta deportación de niños ucranianos. CNN. https://cnnespanol.cnn.com/2023/03/17/orden-de-arresto-putin-trax/ (Accessed 28 Mar, 2023).
[11] Zhu, J. J., Jiang, J., Yang, M., & Ren, Z. J. (2023). ChatGPT and Environmental Research. Environmental Science & Technology.
[12] J. M. Sadurní (17 Feb, 2021). Alan turing, el arma secreta de los aliados. National Geographic. https://historia.nationalgeographic.com.es/a/alan-turing-arma-secreta-aliados_16352 (Accessed 28 Mar, 2023).
[13] Kashefi, A., & Mukerji, T. (2023). ChatGPT for Programming Numerical Methods. arXiv preprint arXiv:2303.12093.
[14] Akbar, M. A., & Khan, A. A (2023). Ethical Aspects of ChatGPT in Software Engineering Research.
[15] BALOĞLU, G., & ÇAKALI, K. R (2023). Is Artificial Intelligence a New Threat to the Academic Ethics?: Enron Scandal Revisited By ChatGPT.
[16] Haemmerli, J., Sveikata, L., Nouri, A., May, A., Egervari, K., Freyschlag, C., … & Bijlenga, P. (2023). ChatGPT in glioma patient adjuvant therapy decision making: ready to assume the role of a doctor in the tumour board?. medRxiv, 2023–03.
[17] Gravel, J., D’Amours-Gravel, M., & Osmanlliu, E. (2023). Learning to fake it: limited responses and fabricated references provided by ChatGPT for medical questions. medRxiv, 2023–03.
[18] Le, D. P., Hall, S. C., & Le, D. (2023). Medical Literature Writing With ChatGPT: A Rare Case of Choriocarcinoma Syndrome With Hemorrhagic Brain Metastases Due to Burned Out Metastatic Mixed Testicular Cancer. Cureus, 15(3).
[19] Santandreu-Calonge, D., Aguerrebere, P. M., Hultberg, P., & Shah, M. A. (2023). Can ChatGPT improve communication in hospitals?. Profesional de la información, 32(2).
[20] Xataka. CHAT GPT-4 en 4 MINUTOS. Youtube, 16 Mar, 2023. https://www.youtube.com/watch?v=9urHjGQoiTw (Accessed 28 Mar, 2023).
[21] Arjun Sha (28 Mar, 2023). How to Use ChatGPT 4 For Free. Beebom. https://beebom.com/how-use-chatgpt-4-free/ (Accessed 28 Mar, 2023).
[22] Salesforce (7 Mar, 2023). Salesforce Announces Einstein GPT, the World’s First Generative AI for CRM. https://www.salesforce.com/news/press-releases/2023/03/07/einstein-generative-ai/ (Accessed 28 Mar, 2023).
[23] Jamal, A. (2023). The Role Of Artificial Intelligence (AI) In Teacher Education: Opportunities & Challenges.
[24] Michelle Toh (29 Mar, 2023). ChatGPT and AI automation: 300 million jobs could be affected by latest wave of AI, says Goldman Sachs. CNN. https://edition.cnn.com/2023/03/29/tech/chatgpt-ai-automation-jobs-impact-intl-hnk/index.html (Accessed 30 Mar, 2023).
[25] DW (29 Mar, 2023). Cientos de expertos mundiales piden detener pruebas de inteligencia artificial. DW. https://p.dw.com/p/4PSDR (Accessed 30 Mar, 2023).
[26] Janosch Delcker (30 Mar, 2023). Inteligencia Artificial: Cuatro formas en que la IA transformará la sociedad. DW. https://p.dw.com/p/4PVmK (Accessed 30 Mar, 2023).
[27] Bubeck, S., Chandrasekaran, V., Eldan, R., Gehrke, J., Horvitz, E., Kamar, E., … & Zhang, Y. (2023). Sparks of Artificial General Intelligence: Early experiments with GPT-4. arXiv preprint arXiv:2303.12712.
[28] Nori, H., King, N., McKinney, S. M., Carignan, D., & Horvitz, E. (2023). Capabilities of GPT-4 on Medical Challenge Problems. arXiv preprint arXiv:2303.13375.
[29] Koubaa, A. (2023). GPT-4 vs. GPT-3.5: A Concise Showdown.
