translation: "Investors, please chill about ChatGPT already"
What if answering machine business is just a few billion dollars business and completely removed the need of search? What if it is a kind of business that makes hundreds of billions but makes it through selling the product itself? This is fundamentally different from what Google offers.
The idea that maybe Microsoft or Apple or Google or FB doesn't have enough good engineers to code a best-in-class language model is absurd.
They all have orders of magnitude more data than OpenAI to build a better model, though.
Apple released an iPhone, cannibalising their iPod market.
Google built a great search engine long before they worked out how to monetise it properly. If they can’t get AI working in their products someone else will, and google will go the way of zune.
This is a terrible example. Everyone knew cell phones were a bigger market than mobile MP3 players, Apple was using this mobile expertise gained in iPod to move to a larger market. We don't know if ChatGPT will be a larger market than search currently is. Digital cameras (include sensors used in webcams and cellphones) are currently a smaller market than selling film was at peak. If ChatGPT will destroy Search volume, and if it will bring in 1/20th the revenue than it might be best for Google to delay adaptation as long as they can.
> If they can’t get AI working in their products someone else will
Google is using 'AI' in all their major products. It's possible that soon AI will make their most profitable products obsolete.
Yet Meta's 175billion parameter blenderbot is absolute garbage. Despite claiming stats like 'being twice as knowledgeable' or ' Compared with GPT3, on topical questions it is found to be more up-to-date 82 percent of the time and more specific 76 percent of the time'
Feel free to try yourself.
At lot happens in the space in 3 months. See where OpenAI was a year ago compared to today...
> "See where OpenAI was a year ago compared to today..."
3 months versus 12 months - a 9 month difference. OpenAI wasn't actually far from today 3 months ago.
The complexity is in making it cost efficient to run, making it intuitive to use, etc. That's a problem that google is more than capable of executing on. They'll need to reorganize to pull it off though. They'll even sprinkle ad results in.
But the space is incredibly exciting as a user/searcher/average joe.
If your websites add no value to the LLM (hence to the searcher, to the world, how deep should we get?), well then thats a good side effect
If it’s like a featured snippet, should work great. “Here’s an answer, read more here”
If the LLM Hoovers up all the knowledge and doesn’t show where it got it, then there is no incentive to produce the knowledge
I imagine Google is acutely aware of this, they have a symbiotic relationship with site owners. But we’ll see.
There is some sort of issue here not well addressed by copyright
Google has failed to win decisively in the search result ranking arms race. Concrete example: sites that rip from stackoverflow have been ranking higher than stackoverflow in google search results. Other search engines fare worse.
If Google trained a discriminator that penalized these practices I think their search product would dramatically improve and the web as a whole might benefit - although tabloid article farms might sue.
They have 84% market share which seems like winning in all the ways businesses care about, regardless of how it ranks your stackoverflow searches.
The arms race of "fast trash" websites vs quality search results is not being won by Google or anyone else as of today. Google having a business model that doesn't even incentivize them to win that race =/= Google not losing that race. It's closer to them giving up in defeat, which is a loss for all of us (many times over).
Maybe it's more convenient for the user to pay the LLM for the info (paid via ad attention or whatever), but if I'm the one who made the measurements, then shouldn't I be paid too? If the user never "pays" me, then I'm going to stop publishing measurements, and the whole LLM falls apart.
Likewise for generative art. The people who made the training data are a critical part of the ecosystem. Cutting them out seems like cutting out some critical part of a food chain.
Google Search could pay some small percentage back when they use their work. But it would be a weird one so not sure that's really going to happen.