Chatting with ChatGPT and getting acquainted with the machine.

ChatGPT has been developed by The AI research laboratory Open AI as a model to mimic human linguistic responses to questions, problems, statements, and even snippets of code that won’t run or need correction. It can write articles, essays, and emails in seconds and mimic human linguistic styles. When visiting the application for the first time, it gave a Shakespearian response to let me know that the site was being used to capacity and the user would have to wait. In reality, ChatGPT is growing faster than anything in the history of the internet, and the site is temporarily denying new users access due to phenomenal global demand.

ChatGPT is impressive in some regards. It creates a seemingly plausible array of human-style responses to questions and problems. The machine has been trained using a system called reinforced learning from human feedback. Numerous responses to queries were collected and ranked by AI trainers. Humans have been busy teaching AI how to respond with seemingly authentic linguistic patterns.

This technology is still in its infancy. It is not an omniscient oracle just yet. On the Open AI website, its many limitations are openly referred to. One problem is that answers appear to be plausible, but the content of the answer may be partly nonsensical. Another problem indicated is that a slight rephrasing of the same question can create a different response or even an inability to answer the question. Another issue referred to is that the answers can be ‘excessively verbose’ as the AI trainers were attempting to mimic human-style responses and it uses certain phrases excessively. One final admission was that it may respond to certain harmful requests and even give biased answers. Indeed the final problem would be the ultimate tribute to a human-like defect.

The creators describe their concern about possible ‘harmful outputs and feedback from users that would help them find ‘novel risks’. It is capable of correcting code that won’t run for software developers and as this evolves, it will almost certainly turn the coder into an onlooker, participating in the creation of code with the most minimal input. Writing articles like this one can be carried out in seconds, and it will be up to the reader to choose between AI and human text, and eventually, the distinction may become largely irrelevant.

Talk of harmful outputs may be the tip of the iceberg. The thought of the typical phishing email littered with spelling and grammatical mistakes being replaced by a pristinely written fraudulent message is worrying. It is being reported that it is possible to get ChatGPT to produce malware as only some of the requests being made were flagged as dangerous. Conversely, it could be employed as a tool to carry out malware analysis, completing complex tasks in seconds, and feasibly reducing the demand for human eyes to carry out analysis on malicious code.

The new iteration of this evolving technology is a simulacrum of consciousness. The hypothetical moment of machine consciousness has been denominated by the term 'singularity', and this is still a theoretical concept. New versions of the machine will evolve and mimic human linguistic patterns more closely. This will create an illusion of authenticity. The AI will become smarter and more adept at providing human-like responses based on a predictive model by scraping data from multiple sources. The days of search engines as we know them in their current form are surely numbered.

I decided to have a brief chat with ChatGPT and the dialogue between us follows below.

?This truly impressive tool can’t supplant the independent search for knowledge. It will be a way to outsource thinking and problem-solving to artificial intelligence. How will the answers be conclusively determined? Will this tool exhibit some inaccuracies and fallacies in its responses? These consequences seem inevitable as the length of answers is limited for practical reasons, and the seemingly infinite number of potential responses that could be proffered by this AI will be limited. ?