Artificial Intelligence: GPT-3 Success and its Limitations

Recently, there are some fantastic breakthroughs in artificial intelligence. Nevertheless, it might seem of mainly theoretical interest. It is one thing to generate impressive results in a research setting and quite another to apply them in practice.

The reality of artificial intelligence has not lived up to the theory for many businesses. Usable results, generating consistency, training the models, cleaning, and preparing data are all often harder than they are made out to be.

Nevertheless, the artificial intelligence era has the power to surprise. Thus, advances can come in unpredictable leaps. You should know how to tap those new capabilities as they become available and work out how best to embed them in the work process of the company. It is because they are becoming essential skills for anyone who wants to aspire to a leadership position in the business.

This year’s most remarkable breakthrough in technology is a sign that advances in the field are accelerating. Also, it shows that the products of advanced research are available in an easily consumable form for business managers. Moreover, it can happen sooner than might have seemed likely. The research paper first described the technology in June. After three months, in September, Microsoft took out an exclusive license to bring it into general use. It is a remarkably rapid development that highlights the potential.

Artificial Intelligence

Already, there is a surfeit of words written about GPT-3 (a fair number of them written by GPT-3 itself). The system of the large-scale language model is akin to a glorified auto-suggest system. By applying and ingesting a vast body of text (around 500bn words), it is offering suggestions for the next most likely word in any sequence. The system is able to produce a coherent-sounding response that stretches from words into sentences. Then, it extends to entire paragraphs. If there is a more extended response, the more likely it will wander into incoherence.

There is an exciting thing about GPT 3. It can derive underlying rules from a relatively small number of examples.

That includes working out basic grammatical principles. Also, developers working with the system have let it loose. It led to the success of teaching artificial intelligence on how to tackle simple coding and math problems.

It is easy to dismiss this as little more than an elaborate party trick.

Of course, there are some limitations common to all artificial intelligence nowadays. GPT-3 is lacking context. It does not have any general model of the world to draw on.