Hacker News Reader

Comment by kickaha | Hacker News Reader

Ten years ago you could say to your iPhone, “Hey, Siri. What’s the ratio of the distance to the sun to the distance to the moon?” And it would tell you the answer and give a pop up with the parameters and calculations from Wolfram Alpha. Solidly reliable.

This was such a fun game in the car with little kids who didn’t know any better that (1) it’s magic I couldn’t have imagined ten years before and (2) it was nerdy as hell.

Today it will, at best, give you the “Here’s what I found on the web” uselessness. So frustrating.

View on HN

Siri had a WolframAlpha integration and removed it?

I don’t know if it’s removed, but it seems to be. I just asked “how many calories in 500lb of beef” which would certainly provide a WA answer in 2011. Today it just gives useless web results.

Siri f-ing sucks. I guess Apple didn’t want to pay anymore.

You used to be able to say "Ask Wolfram ..." for a while to get around it, but now even that seemingly is gone

Yes, and it apparently was removed.

That's a great comparison because I believe the same trajectory of voice recognition will be followed by these chatbots.

Everyone is pissing in their pants over this new natural language interface, but in a few years we're gonna collectively realize this is just search but worse.

Why would it be the same as voice? Siri/Alexa etc didn't do anything different than convert speech to text and text to speech. You get the same response typing it into a search bar. I never really used voice assistants because I'm fine just typing my query.

This is a clearly different paradigm.

No it's not. You're just searching into the training data space and getting opaque results. It's a downgrade.

Some happy path examples are getting hyped, and the investors are falling in line, as in every classic tech bubble. But to think that we can go from processing and generating natural language to AGI if we do it harder is preposterous.

ChatGPT is a downgrade to Siri?

I'm not really sure how to respond to that assertion. I audibly giggled, so there's that.

A downgrade from plain search. But I'm glad you're amused, not that I find that a very respectful response in conversation.

What made it so much fun was that I could do it while driving the car. Or that the kids could do it before they could type, or even read.

It can answer questions that have never been asked before. Search can't.

You're thinking of older chatbots. Search can absolutely return valid results to queries that have never been made before.

Search can return valid results that you will have to synthesize yourself into the answer. LLMs integrated with tools can return the answer directly. This is a significant advantage over search.

You still have to not only synthesize the answers these models return, but now also first analyze the answers for the relevant parts, and then perform searches to validate them.

LLMs can show their work in their output, making the validation easier. Search cannot answer the question to begin with.

Hook up the chatgpt shortcut to siri and boom problem solved.

Link?

There are a billion GPT-3 Apple Shortcuts if you search for them on the Shortcuts app on your iPhone/iPad/iWatch, but I preferred to just make the shortcut myself so I could easily modify it and be sure I wasn’t leaking my API key to a 3rd party. I just followed this simple guide: https://www.innoq.com/en/blog/openai-gpt-3-on-macos-ios-appl...

Can it still do stuff like set timers on my phone for me?

Reply on news.ycombinator.com