A machine could replace you, or it could enhance your abilities. How many decades have we been hearing that? Now, the bots are moving in to new areas. The jury is out on the impact the new language-writing software is going to have on society, but it could be huge, depending on what you do for a living.

Shomron Jacobs, Head of Applied Machine Learning & Platform and Spokesperson at Iterate.ai, says not to hit the panic button. "ChatGPT's impact will undoubtedly revolutionize the way we interact with technology, but it will not be earth-shattering like email or cell phones. Professions that heavily rely on repetitive tasks are most vulnerable, but the tech will be able to assist rather than fully replace human workers. While the impact may not be noticeable in the next year, it will undoubtedly shape the future of work. At its core, AI technology is designed to make life easier for people and make things run smoother and more efficiently."

If it involves numbers, say accounting or insurance advisor, or journalist, you might be affected. If you’re not familiar with Chat GPT yet, you might be at risk. “No doubt technology takes the place of some things that humans do, but it frees people up to do other things that are more productive that we don’t have robotics to do,” says Al Thompkins, who has been doing news for 50 years.

The University of Pennsylvania did a study finding 80% of the workforce will find 10% of their work impacted. Sure, it’ll take a chatbot to make sense of that. Thompkins says before there were emails there was a pony express. Advancements means some things will be left in the dust. He likens the potential disruption to what photoshop did to pictures. Your lying eyes could deceive you to reality, and a search engine that speaks fluently in paragraphs may sound very personal though you’re conversing with a machine. “The hazard is people might become cynical, might not believe anything is human created.” Technology always moves faster than the ethics behind the decision making to use it or not use it.”

Thompkins says by nature we tend to default to the worst possible likelihoods, and it doesn’t always turn out to be all bad. There were sceptics about the introduction of the internet and emails, heresy in today’s thinking. “Sure, there will always be something lost, there will always be something gained. The trick is to make sure we gain more than we lose.”

We lost penmanship when we gained instant communication. Shakespeare gave way to random Twitter comments.

What is Chat GPT? Let’s ask Bing:

“GPT-3 is a large language model that can generate text for various tasks and domains. It has been shown to have impressive capabilities, such as writing essays, creating websites, and generating code. However, it also has limitations and challenges, such as being biased, inaccurate, unpredictable, and potentially harmful. The impact of GPT-3 on society and news is not clear yet, but some possible scenarios are:

These are just some of the possible impacts of GPT-3 on society and news. The actual outcomes will depend on how GPT-3 is used, regulated, and evaluated by humans.”

How do we know what is real and what’s not? Let’s ask Bing:

Evaluating the quality and reliability of GPT-3 generated text is a challenging and important task, especially as GPT-3 becomes more widely used and accessible. There are different methods and metrics that have been proposed and used for this purpose, such as:

N-gram based metrics, such as BLEU, ROUGE, and METEOR, which measure the overlap between the generated text and one or more human-written references 1 2. These metrics are commonly used for machine translation and summarization tasks, but they have limitations, such as being insensitive to meaning, style, and factual accuracy 3.
Reference-free metrics, such as BERTScore, which compare the generated text and the reference using contextual embeddings from a pre-trained language model like BERT 4. These metrics aim to capture semantic similarity and diversity better than n-gram based metrics, but they still rely on having a reference text to compare with.
Factuality metrics, such as FEQA and GLTR, which use question answering or language modeling techniques to assess whether the generated text is consistent with factual knowledge or common sense . These metrics can detect some types of errors or inconsistencies in the generated text, but they may not cover all aspects of factuality or reliability.
Human evaluation, which involves asking human judges to rate the generated text on various criteria, such as fluency, coherence, relevance, informativeness, etc. This is often considered the gold standard for evaluating text generation quality and reliability, but it is also costly, time-consuming, and subjective.

None of these methods or metrics are perfect or comprehensive. They may have different strengths and weaknesses depending on the task, domain, and purpose of the text generation. Therefore, it is advisable to use multiple methods and metrics to evaluate GPT-3 generated text from different perspectives and dimensions.”

You surely don’t remember before there were telephones and probably don’t remember or give much thought to before there was an internet of things. Your grandchildren will wonder that you ever actually wrote something for yourself.

photo: Getty Images