We can call it now: search is dead.
There is, as usual, a caveat. This is search as we know it, i.e. the search engine. Google, Duckduckgo, Bing, and others are cooked.
Naturally, that does not mean they will disappear overnight. In fact, some, like Google, will not disappear. That is something that will be transformed.
AI is affecting everything. This path to major disruption, even from LLMs, is underway. While they are limited is scope at the moment, they will end up spreading over time.
One of the key items missing is real world data. We are seeing the amount of digital data piling up. That said, real world understanding is still lacking.
Fortunately, it is not required for search.
AI Search Is Going To Take Over
OpenAi released SearchGPT. This is its answer to a search engine. We are going to see Meta, Grok, and Claude follow suit at some point. We already have Perplexity that promotes itself as an AI search, albeit with some obstacles.
The reason why search as we know it is dead is due to the fact that it deals with links. That is what it was designed for. Google scraped the Internet, indexing content to offer up websites to users.
Unfortunately, even leaving aside the manipulation Google engages upon, the approach is outdated.
Generative AI alters the Internet completely. We are moving towards a time when information will be fed through agents. Some of these will still be in the form of a website but it is really an agent. Basically, these will be tied to platforms.
In that realm, weblinks are not required.
This is going to disrupt a great deal online. What is the Internet without the webpage? We are going to find out.
People Want Information
How often has you done a search and got a bunch of results that had nothing to do with what you sought?
This is a common occurrence due to the technical limitations. Traditional search does not excel at providing information. It is designed to provide pages. The linking of pages is how Google (and others) try to establish context.
Of course, people simply want information.
Here is where AI search could excel. At the moment, we are seeing hallucinations as commonplace. Is it any worse that traditional search? Depends upon who you ask. Some feel it is inferior while others are proponents of it.
Nevertheless, as they say, patience grasshopper.
The next generation models are being trained. This is something that is going to jettison, in my opinion, things ahead a great deal. Both XAi and Meta are using 100K+ GPU clusters to train their next generation LLMs. XAI has stated that they will be adding another 100K H100 equivalents in 2025.
We are in the era of the 200K and 300K supercluster.
As a comparison, Llama3 was trained on 16K cluster using 16 trillion tokens worth of data.
It would not surprise me if Llama4 is trained on a 125K cluster.
This means the base model will be far superior to what is available at this moment. That translates into even more powerful, and accurate, applications. It will not eliminate hallucinations but should reduce them a great deal.
AI search is going to be one of them.
Of course, this is a race. The central premise is compute. That is the base unit everyone is looking at. Those who are constrained are going to see their rollouts stalled. This is what is happening to OpenAI.
Within a year, I would surmise that most of us will use AI search. Where we get this from is to be determined.
Search, as we know it, is on the way out. We will just await what the next big wave is.
Posted Using InLeo Alpha