Hacker News Reader

Comment by ryanSrich | Hacker News Reader

Exactly this. ChatGPT cannot answer 90% of the questions I ask it. The reason? Because the information I’m looking for exists after 2021. I have no idea how this would ever replace a search engine that has real-time results.

View on HN

In fairness, I'm not sure what changed but frankly Google doesn't answer my questions anymore either. It just shows me random spam websites, number 7 will surprise me.

I know your point is about product features not accuracy, but these things may be connected. It might take an AI to sift through the shit we created via SEO.

That said, I tried using ChatGPT today as a stack exchange replacement and it was like a smart 7-year old. Sounded super impressive but wasn't actually correct.

Yeah there's a race between Google becoming totally useless and ML technologies becoming just better enough than it. If Google keeps on lowering the bar enough then someone may be able to disrupt it. Although it isn't going to be ChatGPT, it'll be an ML-indexed and ML-driven search index over the internet.

Persistent indexing is just an implementation detail. Models with access to a database are already in existence.

Give it til mid year when they implement chain-of-reasoning.

We are very closd to superhuman levels of research.

Until that happens, and it works better than Google, I don’t see a path to disruption. For example, I can type “NBA schedule” and Google will give me the exact thing I’m looking for. A chronological list of games, their times, if they’re in progress (with real-time scores if they are, and even possession). I think ChatGPT supporters downplay how important instant real-time results are.

And there is nothing to suggest right now that ChatGPT can’t do any of that. Just because the creators limited the information doesn’t mean it cannot be opened up and be fed information.

Calling a billion dollar business an "implementation detail" is quite a stretch.

Exercise left for a reader.

Haha, it’s true that it’s a lot of work. But it’s just that: work. We now know that it’s possible with the current class of techniques.

That is today's ChatGPT. Next year's AI will be able to be continually retrained with new information, such as information that appears on the web.

Continuously learning chatgpt would probably be absolutely insane indeed.

Yet it remains that chatgpt has no notion of « truth » or information quality. Whereas google has built a whole system of ranking reliable source of truth for every possible subject. I’m not sure a language model can reach that goal, but i may completely wrong.. It’s going to be fantastically interesting !

Reply on news.ycombinator.com