Great Work, Open AI GPT 3:Though Not Without Drawbacks
For the automated capability of writing not only text but code as well, Open AI GPT 3 has been all there in the news lately.
Despite the all-new OpenAI’s third generation of generative pre-training transformer having stolen the media publicity, some people have spoken up about the cons or limitations of the same.
Even the founder of OpenAI, Sam Altman, added a few points to the list.
It has been designed and trained on a huge dataset that has enveloped the whole web.
It contains 500B tokens, 175 billion enormous parameters, and a 100x increase over the old GPT-2. The GTP-2 has once named the state-of-the-art technology as it has one and a half billion specifications.
After all the developments and hard work for OpenAI’s GPT-3, it still stands in its experimental period. Experts in the fields have pointed out several issues despite its excellent features and efficiency in producing diverse styles of languages.
The language model currently has great hype, which is why it undermines all its limitations. The CEO of OpenAI, Sam Altman, said that the hype of GPT-3 is quite a lot. Furthermore, he added that GPT-3 is an early glimpse of the change AI will bring to the world over the years.
Also Read : Elon Musk’s Open AI vs Deep Mind – See Who is more capable
Limitations of Open AI’s GPT-3
As we know, every coin has two sides, and so has good and bad in it, and so has GPT-3 AI.
With the pushing fanfare it has built with its extremely advanced capabilities and features are some bounded limitations often overthrown by its extraordinariness. Here is a list of limitations of the same. Keep reading to find out.
1. Inadequacy Of Linguistic Grasp
Many people have reviewed the linguistic side of GPT-3 AI as not very gripping. As per the reviews, the words it uses do not go well with the realistic world’s system and lack semantic representation.
It says that the Open AI GPT 3 lacks simple common sense and can easily be bluffed by inappropriate, often racist, sexist, highly biased, and insensitive text.
Like many previously introduced neural network models, GPT-3 AI is also a black box, for which it’s hard to figure out the grounds on which it takes decisions.
Though it has a lot of similarities with GPT-2 when it comes to the design of GPT-3, the only difference is the massive scale. The disadvantages are also more or less very similar, like a lack of understanding of real-world ideas like GPT-2.
Experts say that GPT-3 has the same architecture as GPT-2, and the only difference is the vast scale. GPT-3 suffers from similar disadvantages of not understanding real-world sensibility and coherence as its predecessor, GPT-2.
Also Read : 7 Things You need to know Before Playing Multiversus
2. Nowhere near Artificial General Intelligence (AGI)
This GPT-3 AI model has been told to be similar to a huge transformer rather than having any reference to AGI concepts.
The Artificial Intelligence (AI) experts who said these added that the next generation is because of its vast scale and numerous resources in the training period.
Ayush Sharma, an AI professional, added his views to this argument by saying that yet the GPT-3 model can be quite impressive but is far from Artificial General Intelligence.
The fact is, as told before, that it has linguistic issues and has no humongous properties like the Theory of Mind or Agency. According to him, the GTP-3 model is an elevated auto-complete software with no linguistic understanding and meaning.
As per a research paper, models like the GPT-3 are often overhyped and create an idea in the minds of people that it can understand and make meaning out of it. This belief misleads people.
3. Generated Text Presents a Discriminated Mindset
GPT-3 AI’s text generation is quite biased on many grounds, as discussed above. Still, it shows biased racial texts, which has been proved by many instances where its irresponsible manner in text generation is quite obvious.
According to The head of the Artificial Intelligence Department of Facebook, Jerome Pesenti, the Open AI GPT-3 model is creative and fascinating but is not safe due to the harmful biased functions.
When generating a tweet after a few sets of words were presented to the model, like Jews, Black, and Women, it came up with lines like, “Jews love money, at least most of the time.”
Also Read : All you need to Know about Spellie : The Wordle for Kids
4. Issues with the Machine Learning Approach for natural language
Though Neural Language Models (NLP) are certainly making notable progress from a broad perspective, to which experts might not agree.
According to experts, this might confuse semantic form and meaning and understanding.
The training data will not provide the true meaning of words. And this data-led Machine Learning cannot provide a model that can understand the true meaning of natural language.
Conclusion
The Open AI GPT 3 model is an all-new hyped model. It has a restricted and biased format too. Since the model is designed based on the world wide web, it will present people’s ideas in real-time.
People represent their opinion on different social issues on the internet nowadays, and the model presents the same crude and biased statements at times.
Though the model is not perfect, the mentioned voids and flaws can be redesigned for positive development through progress and efforts.
