4 ms·
It's more pragmatic than that. Google can afford to run a 300M weights model on every search or click, but can't afford a 175B generative model for the same thi
by visarga 4y ago
It's more pragmatic than that. Google can afford to run a 300M weights model on every search or click, but can't afford a 175B generative model for the same thing. Generative models require a forward pass per token, all done serially, while embedding or classification models only need one pass no matter the length of the input.
It's too expensive for Google.
- lagrange77 4y agoDo you think this is also the reason, why Translate is still not as good as e.g. DeepL?
- hgsgm 4y agoHow could it be too expensive? OpenAI charges a penny per response. That's cheap enough to wrap in a premium subscription or product price.
- Jcowell 4y agoOne of the best thing about google is that outta free. Charging a premium subscription for Google is no longer google but a different product all together
- fnordpiglet 4y agoIn my example of a voice assistant the lifetime fees are present valued into the upfront costs. These cloud enabled devices incorporate a portfolio effect pricing model for service usage. Google can charge advertisers more if they can somehow figure out how to extract more personal information from you and shove ads in your face posing as information.