Neo Vision’s Weekly Rant #42
Where we RANT about what goes exceptionally good and wildly bad when you turn your company digital
Each day, as we traverse the ever-evolving landscape of artificial intelligence, we find ourselves captivated by groundbreaking digital products and intriguing discoveries that hold the promise of shaping our future. Yet, amidst the bustling AI marketplace, few entities have sparked as much interest and admiration as OpenAI's GPT-3.5 and GPT-4, their latest large language model.
Google's Bard, another notable name in the large language model (LLM) realm, recently received an update and expanded its reach to the Romanian market, presenting an opportune moment for comparative analysis.
Being all too aware of the human tendency to fall prey to cognitive biases - we do have over 50 of them, after all - I decided to adopt a unique approach for this comparison. Instead of making the judgment myself, I turned the tables and let the AI models take the reins.
So, without further ado, let's dive into the findings of this intriguing head-to-head battle between GPT-4 and Bard, as outlined by none other than GPT-4 itself:
Based on the tasks given (link here), both GPT-4 and Bard have provided relevant and useful responses, but there are some differences:
Istanbul Itinerary: Both models provided a one-day schedule that included visiting Hagia Sophia and the Blue Mosque. GPT-4's plan was more detailed and structured, providing a timeline from morning to evening, which may be useful for a visitor looking for a comprehensive plan. Bard provided a more flexible itinerary, which might appeal to those who prefer a less structured plan.
AI Evolution Timeline: GPT-4's timeline was more detailed, considering various factors like transfer learning, common-sense reasoning, personalization, and the ethical and regulatory challenges related to AGI. Bard also provided a general timeline but with less specific detail about the stages of AI development.
AGI Development: Both models addressed the query about the possibility of LLMs autonomously developing into AGI. GPT-4 emphasized that current LLMs are not capable of autonomous learning or improvement, and AGI would require significant advancements in AI beyond current technologies. Bard suggested that AGI is possible, and LLMs could eventually develop autonomously, but also highlighted the importance of safety and ethical considerations.
Depending on what you value more in responses (detail and structure vs. flexibility and brevity), one model might seem to perform better than the other. However, both models successfully completed the tasks given to them.
🧠 How Smart is ChatGPT?
ChatGPT, OpenAI's language model, has been tested on various professional and academic exams, revealing notable performance improvements in GPT-4 compared to GPT-3.5. The new model has scored higher in exams like the Uniform Bar Exam, LSAT, and SAT, among others. However, it didn't show improvement in AP English and competitive programming.
🥀 RIP Metaverse
The Metaverse, a virtual-reality space introduced by Meta (formerly Facebook), has been abandoned by the business world due to a lack of clear vision, technological shortcomings, and the emergence of new trends like generative AI. Despite the initial hype and significant investments, the Metaverse failed to deliver on its promises of an immersive virtual experience, leading to financial and job losses across various companies. Critics argue that Meta CEO Mark Zuckerberg's focus was more on increasing share prices than genuinely improving human interaction.