ChatGPT Artificial intelligence or machine learning?
OpenAI created ChatGPT a chatbot, built using the OpenAI language model to create human-like text responses. A prototype that is fine-tuned quickly generated attention over its detailed responses. Chatbots have existed for many years at this point. Remember the dozen or more that existed in the early days of instant messaging programs? I can think of a couple that companies use to help provide support. What makes this one different is the detailed responses and wide knowledge pool. A bonus here is the human-sounding responses. The bot uses old data sets with rules to limit responses, no hate speech is allowed. With a little bit of prompt engineering, you can bend the rules or break them. Sometimes it can be uneven factual accuracy or tricked.
It is a huge cybersecurity risk, capable to write software. Therefore, creating a massive issue for open-source software, writing phishing emails and more. Loads of journalists have incorrectly called this artificial intelligence, just advanced machine learning and not yet created an artificial general intelligence, that remains far off. On the surface, some of its responses look impressive but they can be flat-out wrong. Far from perfect and needs more work.
Google uses an algorithm which draws on machine learning. Using common patterns from users to direct you towards the search results. Not just one model or algorithm at play here but loads. Many Google engineers will tell you; it is alive. Similar comments from people who work at Facebook. Big tech firms have benefited from a light touch but slow, the framework around their business. I’m already concerned about facial recognition being widely used with limited laws suited for it. I do hope that this style of software doesn’t get slow or has no oversight.
OpenAI has two arms one for-profit and one not-for-profit. You could put it to use on various websites providing customer support or a useful chatbot directing them to products. Remains unclear the monetization strategy for the software. The worry it ends up in hands of bad actors for a quick buck writing code or attacking companies.
Another problem here that is the datasets being used are free. Already problems around machine learning tools that produce art without references. Therefore, putting artists out of business. I don’t think that going to happen but going to cause conflict. As chatGPT shows, you need up-to-date datasets. Content creators that feed the system for free, Google uses that input to create an advertising output suited to you. I can see Google looking closely at harvesting that for its business. The company already has an eye on TikTok to index its own service. TikTok has regulatory problems around how it handles copyright and user data. I suspect more companies are going to face that problem soon enough. The age of big tech companies in a wild west with light touch could be over. That would mean revisiting copyright laws and more.
ChatGPT could operate in a world with three major bodies shaping the digital space. A growing divide between China, EU and USA regulatory orbit. Light touch low regulation that helped create ChatGPT could find itself limited in what European data it can harvest.
High-risk gamble to create a chatbox for websites of the future. I could see companies upgrading to it and teaching it the knowledge it needs. Not sure if investors are going to get a good return here. I remain sceptical about it being AI. On the bright side does have some healthcare uses, a helpful tool to record how people feel. Investors better are willing to take the long-term view, going to take a while before OpenAI makes a return.
In conclusion, I think ChatGPT is machine learning with a human-sounding output. Imperfect with sometimes out-of-date information. Exciting watching things happen slowly. Maybe another use is a debug tool helping programmers write code.