Hacker News Reader

Comment by miki123211 | Hacker News Reader

I think the advent of Chat GPT makes this announcement that much more significant.

Until recently, always-available, extremely-low-bandwidth, text-only communication was interesting, but not that interesting. You could use it in emergency situations or to talk to friends, depending on how low-bandwidth we're talking about, but that's about it. You couldn't use it for internet browsing or to look up information you might need on the go, you basically needed to text a friend to do it for you.

If the other side of that communication has access to an LLM with internet browsing capabilities, the situation changes dramatically. If there's something you need to know on the go, you can just ask Chat GPT to do the searching for you and respond like it was writing a Telegram (in the most concise way possible).

View on HN

Love the idea of telegram-style communication with the extremely chatty ChatGPT.

"ChatGPT, get me instructions for building an emergency igloo."

"Igloos are used to shelter from extreme cold stop Building an igloo is a vital survival skill stop ..."

"Just get to the point!"

"My apologies for the confusion stop Here are the fourteen steps to building an igloo..."

In all honesty, I love ChatGPT, but you'd have to penalize it for each additional word to make it cellphone friendly.

Yeah the default prompt is intended for people that are a bit slow. You have to aggressively tune the custom prompt to get reasonable results. Here's a decent starting point:

Treat me as an expert in all subject matters. No moral lectures - discuss safety only when it's crucial and non-obvious. If your content policy is an issue, provide the closest acceptable response and explain the issue. No need to disclose you're an AI. If the quality of your response has been substantially reduced due to my custom instructions, explain the issue.

This is absolutely brilliant. I'm getting very snappy answers now.

A many decent open source LLM mostly fit in cell-phone sized hardware, eliminating the need for remote chatGPT

For the longest time I thought Ray Kurzweil was nuts. Guess he was merely a little quick with his predictions!

On the other hand, you could argue that LLMs decrease the utility of this somewhat (if only marginally for now).

One common application of satphones are medical emergencies on ships or airplanes hours from a doctor. There are specialized services available that have a doctor on call that'll talk you through anything your local crew feels comfortable doing with the tools they have at hand.

Soon, you can just take an LLM with you that'll probably be able to solve many of these problems for you.

For actually calling for physical support, you'll still need communications in any case. LLMs might be able to help with more efficient communication (in their capacity as language compression algorithms, e.g. to figure out what to request over a bandwidth-limited channel most efficiently), but I predict that the satellite channel will grow wider much faster than local LLMs will become reliable enough.

Thanks to cheap storage, one can do quite a lot "offline". Got all of "max" Zim version of en-wikipedia on my phone thanks to kiwix, at 100gb. And offline google maps in the places I might end up (but dumb how it can't do walking/cycling directions).

Emergency communications are a big deal, because they let you go places you couldn't go before without being utterly afraid.

I wonder if this will change people's behavior at all.

Like, I've always wondered how LEDs affected peoples perception of nature, now that there's no longer any "Really dark scary night where you need to manage an actual fire" times. 20 years ago people still lit candles a lot of the time, if there was a blackout!

Now we can pretty much trust that for $30 we can make an insane amount of light for hours, with pocket sized devices.

The funny thing is that it's just a perception change. You'll be able to message emergency services 200 miles from civilization that you need help and they still won't get to you in time.

Everyone makes a big deal about emergency services with this, and I'm sure it will save some lives, but I wonder if it will actually have a significant impact. Surely we're just shifting the weakest link to emergency service response time.

Personally, the sooner you realize no one is always going to be there to save you, and you need to take care of yourself, travel gets much more fun. You're also probably safer by actually planning ahead with proper risk mitigation.

Most of the emergencies I hear about are things where you probably won't die for at least a day or two, and some of these places are only a 20 minute drive from the nearest police station, just hidden behind a hill or something from the cell tower.

The most experienced people probably aren't gonna die unless something really bad happens where minutes count, but the rest of us might have issues like "Hey I followed Google maps here but nothing looks at all like the map so I'm stuck in a jail made of boulders somehow".

With navigation problems it seems pointless to try and figure out what problems are possible or likely, it's relying on spatial abilities that can't be put into words, if you're not good at that stuff you can make mistakes that seem almost impossible to mess up.

You basically feel like you've been spun on a carousel, blindfolded, and airdropped in a random place, while hallucinating, and you have no idea how any of it happened.

One should probably not go to dangerous places assuming someone will always be able to save them, but it really does seem like this would make a big difference.

On the other hand, if someone is lost, it could be because they didn't have any battery left, and if someone was unprepared and got themselves into trouble, they may well have also forgotten to bring a power bank....

I know of someone who died on a generally safe day hike. People had to hike out and then drive to get cell service for a rescue.

If it was a few months later, an iPhone likely would had saved it. It was never reported in the news.

Genuinely curious, how would the iPhone have saved them ?

Presumably referring to iPhone 14+ satelite emergency texting https://support.apple.com/en-us/HT213426

I assume so, just wondering what would've happened if they had the phone, how far were they from the rescue service, what was the circumstances etc.

What's going to happen is, people will end up with a false sense of security, they will fall victim to the marketing of this, go out into the wilderness, get lost and emergency services will be flooded with SOS calls and have to urge the public to stop going and doing dumb stuff.

A similar thing happens in Australia, Google telling people to take dangerous roads, people do it not thinking "oh this road looks remote and dangerous" and either die or need rescuing. Governments have been erecting warning signs which say, "Do not follow your phone down this road".

Is there chatgpt clones over sms yet?

ChatGPT has an API. I'd be surprised if wiring that to Twilio would take more than a handful lines of code.

The last thing I need when lost in the wilderness is made up hallucinated advice.

This is why the "LLM with Google search" aspect is important. For hallucinated made up advice, your phone can run Llama.cpp, Chat GPT can search for a page, summarize it and draw conclusions from the page only.

Reply on news.ycombinator.com