3 ms·
Two important points this doesn’t address: 1. Money. How cheaply can this be run per query, and how much value add does this have over the current product? 2.
by janalsncm 4y ago
Two important points this doesn’t address:
1. Money. How cheaply can this be run per query, and how much value add does this have over the current product?
2. Modalities. GPT is only just ok at factual responses. It hallucinates a lot but let’s assume it didn’t. It’s damn near useless for location based and recency based queries. The day after an election, you’re gonna want to update it on who the president is. It’s not feasible to retrain. And don’t even think about answering “best restaurants near me” based on model parameters alone.
What may be possible however is synthesizing the top N results into one coherent answer.
- AnimalMuppet 4y ago> What may be possible however is synthesizing the top N results into one coherent answer. And there's an organization that excels at selecting the top N results: Google.
- moistly 4y agoLately they excel at selecting the top N SEO blog spammers, which IMO is unlikely to result in the synthesis of coherent, or at least truthful, answers.
- janalsncm 4y agoThinking out loud here, is it possible that Google is doing the same thing as always, but the results have gotten more spammy? Were there ever good results for “best headphones for running” or “how to make salsa”? I checked Neeva and it’s not much better. Ditto on Bing. Maybe Google has gotten worse because the internet has gotten worse.