Hacker News Reader

Comment by cmrdporcupine | Hacker News Reader

Google has its own ChatGPT-like/LLM thing that is at least competitive. There's almost for sure a bunch of very legitimate reasons they haven't tried to mate search with that.

#1 from my perspective is that those systems... lie.

View on HN

I suspect that the main problem with these systems is that Google doesn't know how it will affect Google revenue, i.e. if there's a possibility of destroying a lot of revenue without replacing it with larger revenue. I.e. it's what's known as an innovator's dilemma.

I don't think that's the main problem. They wouldn't invested money into Google Assistant, which directly hurts Google revenue. They also already give direct answers to questions on Google without ads.

Realistically, the main problem is scaling and costs/benefits. You have to remember Google is accessed by billions of people a day. Now imagine they serve a product like ChatGPT. You would be pissing off a lot of users with a product that is laggy, often crashes, requires a hourly rate limit just to protect your business from going bankrupt on computing costs. It's not really beneficial to Google to rush a model into their golden product that isn't ready for the scale it operates at.

But it is beneficial for a smaller competitor to slowly chip away. Meanwhile googles investors might rather the legacy company return value to shareholders with greater dividends/share repurchases. While the new competitor is given blank checks with investors not wanting to monetize too much just yet.

I would guess the bigger problem is that someone is going to get GoogleGPT to write Nazi propaganda and they don't want the reputational harm.

For a LLMs as search replacements there's a real big issue putting the 'Google' brand on something that could essentially replace wikipedia and be inaccurate yet trusted in unexpected ways. ( https://i.redd.it/40cnc2bluh6a1.png )

ChatGPT opened the box but there's a real reason to be hesitant

This holds for random blog posts too. Arguably, ChatGPT lies less in some cases because it aggregates multiple sources. (However, it doesn’t counter common misconceptions and blogs typically don’t either.)

Reply on news.ycombinator.com