In November 2022 the artificial intelligence startup OpenAI launched ChatGPT, a viral online large language machine that has the tech world and marketers giddy with excitement…and full of questions. ChatGPT was free to use, but now there is a fee, a necessity to triage capacity issues brought on by the free access that drove huge user numbers.
It seems everyone has an opinion about ChatGPT. Engineers and entrepreneurs tout it as a welcomed new frontier. It’s predicted to fuel new products, services and solutions, but not everyone is on the bandwagon. Social scientists and journalists are worried because they see the potential this technology may have to wreak havoc. Don’t worry. According to experts a full AI takeover isn’t imminent. ChatGPT doesn’t speak with sentience and doesn’t “think” like a human. There is an intrinsic difference between how humans produce language versus large language models.
What puts this chatbot in the spotlight is how close it gets to replicating human-like language abilities thanks to generative AI, which uses statistics, reinforcement learning, and supervised learning to index words, phrases, and sentences. It’s a revolutionary technology that has been trained to learn what humans mean when they ask a question. It can also produce essays, solve coding problems, provide customer service, translate, and more. This is possible because it’s powered by large amounts of data and billions of words scraped from the web and other sources along with computing techniques to mimic writing styles, avoid certain types of conversations, and learn from questions. ChatGPT can access a vast vocabulary and database of information and understand words in context, helping it mimic human speech patterns to dispatch its encyclopedic knowledge.
OpenAI is not alone in its use of large language model tools. Other well-known tech companies like Google and Meta use the large language model tools to develop programs that respond to human prompts and devise sophisticated responses.
The technology is spreading to virtually every field – health care, marketing, law, you name it. But, it’s not a magic bullet. Open AI has acknowledged as much, sharing on its website that “ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers.” As with any technology, there are deficiencies and limitations. Here are a few in the spotlight now:
Plagiarism and Cheating
Plagiarism and student cheating are a valid concern. Thankfully, there is software that can be used to detect OpenAI text. However, due to the learning nature of the AI chatbot, it will keep learning and getting smarter, which may make it harder to identify plagiarism using the prescribed software.
Racism, Sexism, and Bias
ChatGPT fell under scrutiny when racial and gender biases were associated with the chatbot. Implicit bias built into technology is not a new discovery. UC Berkeley psychology and neuroscience professor Steven Piantadosi stated on Twitter in early December 2022 his concerns, “No, OpenAI has not come close to addressing the problem of bias. Filters appear to be bypassed with simple tricks, and superficially masked. And what is lurking inside is egregious.” He used queries based on race and gender, and ChatGPT produced results that favored white males over women and people of color.
Despite the ongoing hype over ChatGPT, it also has ongoing issues with accuracy that have been well-documented since its inception. OpenAI has conceded that its chatbot possesses “limited knowledge of world events after 2021.” If there is insufficient data on a particular subject, the chatbot gives replies that contain incorrect data. This has started a much-needed discourse on the use of AI as a reliable source to create content – especially among news sources.
Clearly, some refined, deep content is necessary to address this problem. While it’s one thing for the chatbot to work “just ok” to relieve writer’s block, let’s say, this level of performance isn’t acceptable when using the chatbot to source valid, deep, and expansive domain data. If the information programmed is inherently flawed and the algorithms aren’t constantly regulating for reliability, then this new technology is just a big disinformation machine. Other issues with inaccuracies include:
Phrasing Sensitivity – ChatGPT is sensitive to tweaks to input phrasing or multiple attempts with the same prompt. It can claim to not know an answer, but when rephrased can answer the questions correctly.
Excessively Verbose – The model can be unnecessarily wordy, overusing certain phrases and being repetitive due to biases in the training data that favor long answers that are believed to be more comprehensive.
Inaccurate Guessing – Ambiguous queries should be questioned; however, the current models usually guess at the user’s intention.
In the tech world, more data is always better. That philosophy is at play here as well, with OpenAI suggesting the system will “learn” from invalid data as long as the data set gets bigger. Fixing issues with inaccuracies is a challenging assignment that will garner more attention as the technology matures.
Not Original Content
A limitation is if more than one user asks the chatbot to write on the same topic, everyone making that request will get the same response. Therefore, the content created isn’t unique or original – all users will end up with the same copy. Thus, personalization is an area that still requires a lot of development.
No Mobile App
Currently, ChatGPT is strictly a browser-based platform with no mobile app feature available as of yet; however, I’m certain one is coming, but there has been no confirmation of this by the brand. This lack of cross platform availability has given rise to copycat apps charging exorbitant prices. Beware of ChatGPT fakes not associated with or supported by OpenAI. A popular workaround for this is to load the chatbot on your smartphone browser.
Looking to the Future
As with most new technologies, pioneers usually take most of the heat for early misses. However, the market will grow and advance, fueled by eager VC firm funding. Financial investors predict ChatGPT will spawn new companies and products moving forward. Some also foresee a new form of Internet search coming from this technology.
Competition is certainly on the way, as all of the players will enter the ring, perhaps introducing industry-specific and domain-specific offerings, new products, and other creative solutions. Just yesterday OpenAI announced version three of ChatGPT is out, unveiling “an entirely new user interface” complete with:
- Five new creative sizes for those who do display or programmatic ads
- A new version of AI that can generate more and better creatives even faster
- An OpenAI-supported startup with several copywriting methodologies
The opportunities this technology presents is staggering. Rather than replacing vast amounts of workers, the technology will be used to enhance and upgrade the systems we use to do our work so we can do that work more effectively and efficiently.