juan M

In a smilar nature of What will be the impact of chatGPT on society (a very slim reply), someone in a whatsapp group of my former MBA classmates asked the following

has anyone done an analysis on the risks/opportunities to Google from OpenAI. Would love to get your thoughts. Seems like we’re living through a new iPhone moment.

There were a few answers, including a bit of banter targetted at management consulting firms, at Google itself, this answer I particularly liked coming from someone that seemed to have been getting exposure with the tool (as I have been)

OK, my take. ChatGPT knowledge is quite outdated (2021) and it’s answers lack refs to sources of information. On the other hand, support of context is chat is great (based on my experiments) and summary answer is also better than search results. It’s an open question who builds a search that has all four features first. I find myself switching between google/chatgpt depending on the task at hand and it never occured to me to go to chat with “where to buy” question. That’s because I’m aware about outdated results.

And then I added my take on the matter:

I would say this is a bigger threat to Google:

Google is already quite deep in the world of ML/AI, with the biggest “moats” being talent, funding and training data, so with the proper focus and steering Google should be able to put a good fight (Google Brain and Deepmind are already in the forefront of many advances in the area).

I don’t know if GPT is better than Google’s own models, but I imagine google has been using it in the backend of many of its services, they didn’t need to create a splash by opening it to the general public like OpenAI did.

I didn’t want to add much more, as the forum is 300+ people in a whatsapp group, and I felt that if everyone started to give their long takes it could get annoying, but I also think that there are several things to discuss around this topic, including in particular an analysis of the threat to Google as a search engine coming from OpenAI’s chatGPT, or more generally, the GPT family, other Learning Language Models (not just from OpenAI).

There was another interesting take a classmate made reinforcing the fact that Google (Alphabet actually?) should actually be in a great position to have great results in LLMs:

The organisations that own the datasets to train language models like GPT will have an edge over Open AI — the likes of Facebook, Google, Quora, Yelp, Reditt, Apple etc have a formidable advantage in this regard.

Not sure that OpenAI has the rights to all the training data for GPT, hence commercializing it will be challenging. The notion here is that copyright protection of training data would apply to the model output. ChatGPT output being of a derived nature, is very different from that of a google search result that simply links to a web page containing the content.

I think the more valid points to discuss is to understand if this old-time UI (a chat experience) powered by a Machine Learning model (ok, I’ll bite and call it “an AI” to avoid being too pedantic) will indeed divert people from going to Google to search for answers.

A few things to consider here are:

• Searches vary by monetisable intent, so if I want to buy something, I will likely still go to Google to find who is selling that something (ok, I could also go to Amazon)
• Indirect intent searches (“how to do X”) that end up with ads pointing me to sellers/products, etc., may be partially affected
• Non monetisable intent searches (“what is the distance between earth and sun”) will likely be less and less done on Google (specially once chatGPT-like experiences can provide a higher degree of reliability in terms of truthiness), but it’s likely less worrying

However, even taking all of that into account, there is still a question of what will be the way to pay for the computing efforts of these LLMs. Will there be subscription models (I doubt people will be willing to pay for a search provider, even less to have several, as currenly many do with streaming providers -of course is not comparable because of content ownership and licensing-)? Will there be ads? Will the AI reply with paid-for recommendations or ads, or will they lie on the side of the chat UI? I guess it will be quite awkward, but there will be people already researching this (I bet we’ll see the word anthropomorfization in articles in the coming months and years).

Finally, what is true as a threat to Google, is the fact that Microsoft is partnering with OpenAI, and if LLMs (and LLM-based applications) start to divert search and browsing time and attention, if Bing becomes better than Google at the application of this technology, then Google will sure feel the pain. But again, as I mentioned in my reply to my classmates, it’s not like Google has been just doing other stuff and not doing its own research and its own applications of these models and technology. They even fired a researcher that thought their version of LLM (LaMDA) was sentient! And they have a separate division inside Alphabet (DeepMind) that has shown progress after progress in Machine Learning models across industries and activities.

All in all, I think it will be amazing to watch. There will be lows for sure, but I remain optimistic that the highs will be great, and hopeful that it will have a positive impact across society.