We at Numorpho Cybernetic Systems (NUMO) are utilizing generative AI techniques for innovation, process automation and smart enablement by utilizing LLMs in conjunction with science and math toolkits to engineer products and solutions in the new and need to be mindful of the outcomes of AI as we progress our journey.
Artificial intelligence has made significant strides in recent years, and one of the most notable advancements is the GPT-4 language model. Developed by OpenAI, GPT-4 has the power to revolutionize the way to engage with information.
The key to unlocking this potential lies in prompt engineering, a method that allows users to communicate effectively with the model to obtain the desired output.
Background
Language has been the mode of communication for mankind. With the spread of humanity, language has evolved into regional variations each with its own nuances to express context verbally and/or via written text. Language has thus been central to human civilization and has adapted and diversified as societies have spread and evolved. The regional variations and nuances in languages reflect the rich tapestry of human culture, while the field of translation and the expertise of bilingual/multilingual professionals have emerged to facilitate effective communication between linguistic communities in our interconnected world.Computer-based translation services like AltaVista’s Babel Fish and Google Translate have played a significant role in bridging language barriers and facilitating communication in our interconnected world. They have evolved over time, thanks to advancements in machine learning and neural network models, although challenges in achieving perfect translations still exist, especially for complex and context-rich content.
I recall when computer-based translations were starting to evolve, web content management (WCM) softwares offered language translator plugins and some of them were funny:
- Mike (person’s name) got translated to microphone in German.
- The phrase “Though his spirit was strong, his flesh was weak” when translated from English to Russian and back to English got converted to “The vodka was strong, but the meat was rotten”.
Prompt Engineering
Prompt engineering is the art of crafting effective questions or inputs for a language model like GPT-4. It involves understanding the model’s capabilities and limitations, as well as tailoring the input to elicit the most accurate and helpful response.
A well-designed prompt can make all the difference in generating valuable insights, while a poorly designed one might result in irrelevant or unhelpful output.
How I stumped ChatGPT
Me: The trophy could not fit into the suitcase because it was too big. What is it?
ChatGPT: The “it” in the sentence refers to the trophy. The trophy could not fit into the suitcase because it was too big.
Me: The trophy could not fit into the suitcase because it was too small. What is it?
ChatGPT: In the sentence “The trophy could not fit into the suitcase because it was too small,” “it” still refers to the trophy, but the reason it couldn’t fit is that the trophy was too small to fill or occupy the suitcase properly.
Me: I think it in this second case refers to the suitcase. Fit is not about occupancy.
ChatGPT: You are correct, and I apologize for the misunderstanding. In the second sentence, “it” indeed refers to the suitcase. The sentence implies that the trophy could not fit into the suitcase because the suitcase was too small to accommodate the trophy. Thank you for the clarification.
Here is a similar thread in Bing that uses GPT-4
Me: The trophy could not fit into the suitcase because it was too big. What is it?
Bing: The sentence “The trophy could not fit into the suitcase because it was too big” implies that the trophy was too big to fit into the suitcase.
Me: The trophy could not fit into the suitcase because it was too small. What is it?
Bing: The sentence “The trophy could not fit into the suitcase because it was too small” implies that the trophy was too small to fit into the suitcase.
Me: That is not correct
Bing: I apologize for the confusion. Could you please clarify what you meant by “The trophy could not fit into the suitcase because it was too small”?
Me: Yes, it is subjective. When I said it was to large in the first case, it referred to the trophy. When I said it was too small in the second case, it refers to the suitcase. Fit is the key and it refers to the trophy in the first case and the suitcase in the second.
Bing: Thank you for clarifying that. I understand now that the sentence “The trophy could not fit into the suitcase because it was too big” implies that the trophy was too big to fit into the suitcase, while the sentence “The trophy could not fit into the suitcase because it was too small” implies that the suitcase was too small to fit the trophy. I apologize for any confusion caused earlier. Please let me know if you have any further questions.
Bing seems to have since fixed this issue:
Moving ahead to situations where LLM based generative AIs will include images and vide in their dissertions, wherein AIs are being trained to detect patterns in images to classify them as a dog or a cat for example. We talk about Large Language Models hallucinating in their prompt responses. When we add images and video to the referential generative engines they should also be mindful of apophenia.
What is Apophenia? (external reference)
Apophenia is the tendency to experience events as meaningful, even when they shouldn’t be. The word ‘apophenia’ comes from a German neurologist, Klaus Conrad, and his 1958 book about the symptoms of schizophrenia. While an epiphany is a sudden realisation of a true connection or meaning, an ‘apophany’ is the false realisation of one. Also called patternicity, it refers to essentially anytime that you are seeing patterns in the world that don’t exist. Extreme cases of it are what is called ‘referential mania’, where patients believe that everything that happens is a ‘veiled reference’ back to themselves.
The psychoanalyst Carl Jung used what is perhaps a better-known term, ‘synchronicity’, to describe instances of apparent connection between co-occurring events with no clear causal relationship. More than just coincidences, they take on powerful meaning in a person’s mind and appear to be a result of more than chance. In one example, Jung’s patient dreamt of a golden scarab and, as she described the beetle, Jung heard a tapping on the window of his office. There, he found a live scarab beetle, which he grabbed and handed to her.
For a character in Vladimir Nabokov’s short story ‘Symbols and Signs’ (1948), the clouds in the sky, stains and pebbles all communicate messages that he has to interpret. ‘Everything is a cipher and of everything he is the theme,’ wrote Nabokov. ‘He must be always on his guard and devote every minute and module of life to the decoding of the undulation of things.’
Apophenia in humans is reflected in pleasant and troubling experiences alike – from seeing faces in clouds to conspiracy beliefs. While apophenia is apparent in symptoms of psychosis – hallucinations and delusions – it is a quality we all share, to varying degrees. When someone lies on the grass and looks up at the clouds to find shapes, they are intentionally engaging in apophenia. One specific form of apophenia, based on visual perception, is pareidolia: seeing faces where there are none, in inanimate objects such as electrical outlets or bell peppers.
Apophenia is the cognitive tendency that helps explain Rorschach tests, sports superstitions, astrology and decisions like selecting birthday numbers for a lottery ticket. Understanding how universal it is – but also the fallacies it can lead to – can help us think more critically about the elements of the world around us and how they might (or might not) relate to each other.
Pattern Recognition and Generative AI
As we progress in AI to train machines in pattern recognition, we must be mindful of apophenia, which could manifest in AI systems in various ways, especially when integrating images and video into their generative processes. Here’s how apophenia could be relevant in this context:
- Overinterpretation of Visual Data: AI systems that incorporate image and video analysis may develop a tendency to overinterpret visual data. This means they might perceive patterns or connections in images or videos that don’t actually exist, similar to humans seeing shapes in clouds. For example, an AI trained to classify objects might mistakenly identify unrelated objects as part of the same category.
- False Positives: In the context of image recognition, apophenia could lead to false positives. AI systems might detect objects or features that aren’t there, potentially leading to incorrect conclusions or actions. For instance, in a security system using AI to identify threats, apophenia could result in the system flagging harmless objects as potential risks.
- Conspiracy Theories and Misinterpretation: Just as humans can sometimes misinterpret events and see conspiracies where there are none, AI systems might draw erroneous connections between unrelated visual elements in images or video. This could potentially contribute to misleading or incorrect analyses.
- Contextual Understanding: To avoid apophenia in AI, it’s crucial to train these systems not only to recognize patterns but also to understand context. AI should be able to differentiate between meaningful patterns and random noise. Developing algorithms that can discern genuine connections from coincidences is a significant challenge.
- Ethical and Societal Implications: If AI systems exhibit apophenic behavior, there could be ethical and societal implications. Misinterpretations or overinterpretations by AI systems could lead to misinformation, false alarms, or even biased decision-making.
In summary, apophenia is a concept that should be considered as we advance in training AI models to recognize patterns in images and videos. Ensuring that AI systems can distinguish between meaningful patterns and random noise is essential to prevent misinterpretations and their potential consequences. Additionally, addressing apophenia in AI is part of the broader challenge of making AI systems more context-aware and capable of providing accurate and reliable insights in a wide range of applications.
Summary
“Mindful prompting” refers to the practice of carefully crafting prompts or input instructions when interacting with language models and AI systems, such as Large Language Models (LLMs) like the one you are chatting with. The goal of mindful prompting is to:
- Guide Responses: Craft prompts in a way that guides the AI’s responses towards desired outcomes. This involves formulating clear and specific queries to elicit the information or responses you seek.
- Minimize Bias and Misinformation: Be aware of the potential for bias and misinformation in AI responses and construct prompts that reduce the chances of these issues occurring.
- Account for Context and Nuance: Take into account the context of the conversation or task at hand when composing prompts. Mindful prompting considers the nuances of language and the potential for misunderstandings.
- Enhance Ethical Use: Promote ethical and responsible use of AI by framing prompts in a way that respects privacy, avoids harm, and adheres to guidelines and ethical principles.
- Improve Efficiency: Craft prompts efficiently to maximize the AI’s utility. This includes using concise and relevant language to get accurate and relevant responses.
In essence, mindful prompting is a practice that recognizes the importance of how you interact with AI systems through well-considered input, with the aim of achieving accurate, ethical, and contextually appropriate results.
NI+IN UCHIL Founder, CEO & Technical Evangelist
nitin.uchil@numorpho.com
A discourse by ChatGPT
Language has played a crucial role in human civilization. It has evolved and diversified as humanity spread across the globe. Let’s break down this aspect into several key points:
- Language as the Mode of Communication: From the earliest days of human existence, language has been the primary means by which humans communicate with each other. It’s a complex system of symbols and rules that enables individuals to convey thoughts, ideas, emotions, and information.
- Evolution of Language: Over time, as human societies developed and expanded, languages evolved to meet the unique needs of their speakers. This led to the emergence of different languages and dialects. These changes were influenced by various factors, such as geography, culture, and contact with other linguistic groups.
- Regional Variations: As humans migrated and settled in different regions of the world, languages diverged, giving rise to regional variations and dialects. These regional variations often include differences in pronunciation, vocabulary, and grammar. For example, English has numerous dialects, including British English, American English, Australian English, and many others.
- Nuances in Expression: Regional variations in language often reflect the unique cultural and historical contexts of the communities that speak them. These nuances can be subtle but are essential for expressing specific ideas or emotions. For instance, some languages have multiple words for different types of snow or different shades of a color, highlighting the importance of certain concepts in those cultures.
- Cross-Language Communication: As societies have become more interconnected through trade, travel, and technology, the need for communication between speakers of different languages has grown. This has led to the development of translation and interpretation as fields of expertise.
- Translation and Bilingual/Multilingual Experts: Translators and interpreters play a vital role in bridging the language gap. They are individuals who are proficient in two or more languages and have the skills to accurately convey meaning from one language to another. Their expertise goes beyond simply substituting words; they must understand the nuances, idioms, and cultural context of both languages to ensure effective communication.
- Challenges in Translation: Translating between languages can be a complex task. Literal translation often doesn’t capture the full meaning or cultural connotations of a phrase or expression. Translators must make choices to ensure that the message is conveyed accurately and appropriately in the target language.
- Globalization and Multilingualism: In today’s globalized world, multilingualism is increasingly valuable. Many individuals and businesses need to communicate with people from diverse linguistic backgrounds. This has led to the growth of language learning and translation services to facilitate cross-cultural interactions.
In summary, language has been central to human civilization and has adapted and diversified as societies have spread and evolved. The regional variations and nuances in languages reflect the rich tapestry of human culture, while the field of translation and the expertise of bilingual/multilingual professionals have emerged to facilitate effective communication between linguistic communities in our interconnected world.
Computer-based translations and the evolution of translation tools provided by companies like AltaVista and Google. Let me provide some information on that:
- AltaVista’s Babel Fish Translation Service: Babel Fish was an online translation service provided by AltaVista, a popular search engine in the late 1990s and early 2000s. Babel Fish allowed users to enter text in one language and have it automatically translated into another language. While it was a pioneering service, the quality of translations was often limited, and it struggled with nuances and idiomatic expressions.
- Google Translate: Google Translate is a translation service launched by Google in 2006. It quickly became one of the most widely used machine translation tools in the world. Google Translate uses machine learning and neural network models to improve translation quality. Over the years, Google has continuously refined its translation algorithms, resulting in significant improvements in translation accuracy and the ability to handle a wide range of languages.
- Evolution of Machine Translation: The field of machine translation has evolved considerably since the early days of AltaVista’s Babel Fish. Modern machine translation systems, including Google Translate, employ sophisticated algorithms and deep learning techniques. They consider context and idiomatic expressions to provide more accurate translations.
- Challenges in Machine Translation: While machine translation has made significant progress, it still faces challenges in capturing the nuances and cultural context of languages. Translation quality can vary depending on the language pair, with more widely spoken languages often having better results. Complex or highly specialized content may also pose challenges for machine translation.
- Multilingual Communication: Tools like Google Translate have made it easier for people to communicate across language barriers, whether it’s for travel, business, or accessing information in different languages. These tools have become an essential part of the globalized world, but human translators remain crucial for high-stakes and nuanced translation tasks.
- Integration in Various Services: Google Translate and similar services have been integrated into various online platforms, making it convenient for users to translate text in real time. This integration extends to web browsers, mobile apps, and more.
In conclusion, computer-based translation services like AltaVista’s Babel Fish and Google Translate have played a significant role in bridging language barriers and facilitating communication in our interconnected world. They have evolved over time, thanks to advancements in machine learning and neural network models, although challenges in achieving perfect translations still exist, especially for complex and context-rich content.
REFERENCES