ChatGPT For Bing Search Credit: thebrink.ai
Search-enhanced transformer models are likely to replace ChatGPT. The tech-giants position this new service as ‘Co-Pilot’ or ‘Electric Bike of our Minds’. Most people’s concerns seem to revolve around issues like trustworthiness and the potential spread of disinformation. Hence, Microsoft is releasing Bing-Search in combination with OpenAI’s GPT-transformer in a very controlled way. Millions of users have put themselves on a waiting list. Taking the lesson learned from the problems with ChatGPT, the first users are asked to add suggestions for improving the product and to provide recommendations how to prevent misuse. First feed-backs are rather positive, although fact-checking may still produce errors and challenging its dialogue capability with freaky questions might produce weird answers. What is obvious, however, is that mind-enhancement tools mark the beginning of a new era in the application of Artificial Intelligence (AI).
The Problem with the GPT-3 Transformer and ChatGPT
The transformer GPT-3 was originally trained as one huge language model to simply predict the next word in a given text. Performing this task, GPT-3 has gained capabilities that far exceed those that one would associate with next-word-predictions. ChatGPT as one of its applications, has proven to be proficient in writing stories, essays, emails, code, songs and more. However, the tool falls short if you are interested in learning about current events and news. ChatGPT only draws on information up to 2021. Despite its widespread deployment, we currently lack a clear understanding how Large Language Models (LLMs) work, when they fail and how they are incorporating new information. Moreover, any socially harmful activity that relies on generating text could be augmented based on deliberate modifications of the code. Examples include misinformation, spam, phishing, abuse of legal and governmental processes and fraudulent academic essay writing. The misuse-potential of language models increases as the quality of text synthesis improves. The ability of GPT-3 to generate synthetic content that people find difficult to distinguish from text written by humans, stipulates many ethical issues. The liability of platform-providers, application developers as well as the responsibility of governments to assess the risks needs to be resolved. Past experience shows that the tech-giants have always found loopholes to protect their market despite the efforts made by antitrust regulators.
The Battle for the Control of Search
Search has been the ‘Cash-Cow’ for Google for many years. Google owns more than 90 percent of the search engine market. It has been the undisputed leader, protected by its vast network of users and advertisers, accumulating huge revenues and spending power. This dominant position has allowed Google to maintain its market share without any urge for innovation. Google-Search has seen some improvements in the past few years, including better question-answering and deep-learning enhancements. But the core of the search-experience has not changed.
Following a scenario of corporate disruption with little to lose, Microsoft is suggesting a new search- experience. One will be able to use the classic search model powered by Bing combined with a conversational interface powered by ChatGPT. Google’s response to the formal presentation of Microsoft’s Bing-ChatGPT project was a hasty demo of its own LLM called Bard. During the on-line demo, Bard made an embarrassing mistake with a wrong answer to a question that temporarily resulted in a $100 billion drop in the company’s market cap. So far Google has not indicated when Bard will be released. Grabbing market share with a generative transformer-based AI is not like the early ‘move fast and break things’ days of Facebook. As the competition heats up, the tech giants will try to outmanoeuvre each other in various ways such as launching new products and adding AI capabilities to their existing markets. Another shortcut to innovation is acquiring Start-ups and research labs. The capacity of the largest AI-centers is doubling every six months, far outpacing Moore’s Law. Microsoft’s new service represents a major advancement in AI-technology. However, new business models are required to take advantage of its application potential and to adapt to a rapidly changing AI-landscape.
Adding Search to GPT-3
Microsoft unveiled the Bing-Chatbot in February and said it would run on a next-generation OpenAI large language model, customized specifically for search. Right now, this new service is only available to a select few, with multiple millions of people registered on Microsoft’s waiting list. According to a first user report, written by Sabrina Ortiz on Feb. 16, 2023, and published by CDNET – as with ChatGPT – one can start the service by chatting. The very first question ‘What can I use Bing’s ChatGPT for?’, provided the following answer:
‘You can use Bing ChatGPT for various purposes such as: Finding reliable and up-to-date information from across the Web; Asking complex or follow-up questions and getting complex answers; Getting summaries of search results instead of being overwhelmed by options; Learning more about topics that interest you by having a conversation with Bing; Having fun and being entertained by Bing’s witty and creative responses.’
One major difference to ChatGPT is that Bing’s chatbot includes sources for every answer it gives you, with footnotes that link back to the source. One example of using this feature of Bing ChatGPT, was demonstrated by the following question:
‘I saw on the news that the President is going to hold a conference today. What is he talking about?’
The system first produced a wrong answer as it was referring to a corporate CEO. This response provides a good example of how chatbots still cannot match humans’ conversational skills. Most people engaged in conversation would rightfully assume that the question was referring to the US- President. However, it was a positive surprise to experience how the links to the chatbot’s sources helped to quickly figure out what went wrong and how the question could be rephrased to get a better answer. Once the question was reworded, the answer provided the sources and news snippets, a feature which obviously will have huge consequences for individuals working as journalists. Providing the sources makes the user-experience much more trustworthy and eliminates misunderstandings or errors. However, as the above example shows, errors are still possible, and fact-checking needs to be improved. Moreover, Microsoft has started to put time-restrictions on the length of chats. Challenging its chat-capability with long dialogues and weird questions might produce insulting answers.
How to Create a Co-Pilot to our Mind
Positioning this new search capability as ‘Co-Pilot’ or ‘Electric Bikes of our Mind’, raises the question of how the mind is defined. Moreover, it opens up the question as to what further developments of this technology are to be expected. The mind can be defined as one’s capacity to be aware of the world, and to experience and feel with the faculty of consciousness and thought. Conducting a conversation with search focused tools will generate a behavioral profile of the user, mapping his individual interest to the system’s data base. Metaphors have often been used to explain the mind. John Locke described an infant’s mind as a blank slate and Freud compared the mind to hydraulic and electro-magnetic systems. Today’s favourite metaphor defines the mind as a computer. We tend to think that perceptual experiences tell us what the external world is all about, without being influenced by our own mind.
Recent empirical research indicates that this is not true. Our beliefs, expectations and other mental states can causally influence what we experience. The mind and brain are two very different but interconnected entities. The mind works through the brain but is separated from the brain. Hence, the brain changes with every experience you have, every moment of every day. Your mind is how you, as a unique individual, experience life. It is responsible for how you think, feel, and choose. Your physical brain merely responds to these unique experiences. One can conclude that advancing the capabilities of search based mind-tools will sooner or later introduce brain research as a driver of product development. In addition to just interacting with search, a new layer is added that analyses our search-prompting behaviour from a brain-research point-of-view. How this will be realized is difficult to predict. The mind of today’s AI is very different from ours and there is plenty of room for improvement. With thousands of researchers working in Neuroscience – adding new theories daily as to how to crack the human brain’s neural code – the momentum of AI-development will accelerate.
Creating value with AI requires humans who are knowledgeable in the technical domain of intelligent machines as well as humans who are knowledgeable in the domain of psychological behaviour. This ‘co-creative’ effort between man and machine has the potential of driving humanity to the next higher level of evolution. Sooner or later brain research will become a key-driver. To adapt and to learn how intelligent mind-enhancement tools can be used for solving problems is the key-challenge with which we are faced. It will set a new benchmark for success or failure applying AI-technology.
excellently long essay series such as this one, fully agree with high challenge conclusion, congratulations.
Pretty exciting times for you ;), No AI winter, we are in a disruptive period; giant companies heating-up competition and geopolitical messiness.
One may always ask why so much waste, damage and redundancy is produced in such periods of change, why positive change can’t go faster, leaner.
Probably main factors are legacy stickiness combined with social relations (Relational psychoanalysis).
Legacy stickiness bias (1) is fully inherited in the AI models (past data). Social relations bias (2) are mapped via ‘super cookies’ since many years evolving to highly complex AD systems by Google, Meta, etc. (opt-out possibilities got limited, e.g. TOR browser).
Bias 1 and 2 risks need to be understood, mitigated in governance and algorithms to eventually reach leaner evolution and keep the human mind free.
Bias 1: Stickiness. Can be the Giant Company Culture and collection of IT systems and documentation mess, some manipulated fanatic mindset or past biased public, science literature (‘the winner takes it all, has time to write and impose’ / ‘biased studies to promote medicals or hide risks’).
Bias 2: Relational. AD-PR individual grouping, manipulation (to brands, political lines is practiced since Edward Bernays).
Sure Bias 1 and 2 are strongly interwound.
AD, health tracking systems gain more and more control of individual’s immediate state (feeling, mindfulness) and context/social relations (Stephen Mitchell).
The geopolitical mess currently prevents from global AI governance, instead does rather protect each countries giant companies for some free rides (speeds-up but also further monopoly gains for winners).