Large langue models are the rage. The challenge, right now, is monetization.
We are seeing billion dollar bets on generative AI. Unfortunately, it is mostly money going out and not coming back in. It is being reported that OpenAi will lose $7 billion this year. Whether that is accurate or not is immaterial. Anthropic is seeing Claude3 eat $3.7 billion in 2024.
There is a lot at stake.
The winner in this race could see enormous valuable. ChatGPT, based upon seed funding, is valued at $80 billion. That said, unlike many others in this field, they have to generate revenue. When you are battling Meta and Google, they generation hundreds of billions each year in revenue.
Against that, ChatGPT is bringing in nothing.
SearchGPT: Taking Over Search?
Google is long the dominant player in search. Will this continue is something everyone is wondering.
To me the answer is simple: it is no.
Google will not dominate search. Whether it is AI that alters this, we do not know. What we can forecast is that sites are already building in mechanisms that keep them on their site. For example, Walmart and Amazon both have built in search similar to what Google does for their products.
This is just one example of what is eating away at the market share.
OpenAI announced they are in beta testing of SearchGPT. This is a widget designed to use its technology for search. It is something that will be integrated, if all goes well, into ChatGPT.
Here is where we see another potential hit to Google search.
If OpenAI brings something like this out, we can presume the others will not be far behind. This is a copycat business with most ending up with the same models.
While Google might be able to fend off one company, such as OpenAI, can it do so if all bring out something like this?
Not likely.
Of course, we are a long way from this so Google isn't having to sweat it yet.
Generative AI: Not The Answer For Everything
Many believe that generative AI is going to be the answer for everything. Honestly, these people might be right, at some point.
As of this moment, however, this is not the case. In fact, generative AI is misunderstood by many.
Search is not the ideal use case. These models do not get fed information and simply mirror it back. That is not how they work.
Instead, they are trained on data and designed to generate information. The training has weighting mechanisms which continue during inference. This is what allows for the model to weigh the probabilities of what comes next.
It is also the core of hallucinations, an ever present problem with these.
That is not to say this is not going to be solved in the future. It might well be something that is addressed in forthcoming updates.
This means that generative AI is actually best for something that doesn't require high level precision. Search is exactly that.
For this reason, we see that the best use case is as a creative assistant. Here is where generative AI excels.
Will SearchGPT and other such endeavors change this? We will have to see. OpenAI certainly believes they are onto something and they certainly have a lot more knowledge about it.
The AI race continues.