As a caveat I’m not an expert on Generative AI and have never used ChatGPT. However I do know more than 99.99% of the population and 100% than anyone in the media on this matter, as a self taught A.I. coder of 7 years.
The most important part of the name is the “G” == Generative, this means it creates something as opposed identifying a result, giving a prediction or ultimately making a decision.
Your probably very familiar with CGI Computer-Generated Imagery, that creates the backgrounds or realistic visuals in movies or games. It is very cost effected, by not requiring humans to undertake intensive stop go animation and, allows visuals that would be near impossible to create eg a tiger talking.
Below is a composite of what venture capital is funding in the space including: visuals, interfaces, text, speech, audio and even coding.
It claimed that Generative AI can create 50% of code for new projects, providing massive amounts of productivity. However as someone without any formal training my personal experience is, the first 80% of coding only takes a few hours but its the last 20% that takes the weeks. Perhaps the most scary part of this is, Generative AI coding ultimately birthing its own offspring.
Today the biggest problem the industry is facing is copyright infringement. The internet permits instant access to content that could be utilised by the Generative AI e.g. a Picasso painting, an article by a leading lawyer, an engineering paper by a student. This content could be appropriated for use against the wishes of the originator.
However with proliferation of information on the internet for me the biggest problem is to avoid, in your area of concern, the equivalent of what someone had for lunch or their miming and dancing in a TikTok video, which appears the majority of new human content. In-fact I personal check a sample of most input data that I use.
My second area of concern is that the AI needs classification or labelling of initial training data to give it a reward mechanism for its actions on subsequent data.
Most people are aware that Silicon Valley is where the largest proliferation of new technology industry is based. You may also be aware that Silicon Valley is in California, which is famous for its “progressive” views. It could be a lazy assumption that, the labellers of the training data could have a high propensity to this political bias, including using a rare event as the representation result.
If you have no knowledge of a subject using ChatGPT is probably an easy way to give you content. If you are a successful specialist you are probably better labelling your own data. If you have no coding skills its probably better to engage your own boutique to assist you in this labelling. Or, you could go to one of the specialist companies below, to give you the accuracy, speed, productivity and access to legal specialist information already available from your industry.
Source: Sequoia Capital