Mine as well. 2 years ago my mind was blown that I could code in a language I didn’t know (scala) while on a log train ride with no internet (Amtrak) using a local model on a laptop. Couldn’t believe it.
The staggeringly effective compression of LLMs is still under appreciated, I think.
2 years ago you had downloaded onto your laptop an effective and useful summary of all of the information on the Internet, that could be used to generate computer programs in an arbitrarily selected programming language.
Yes! Continuing on thoughts of LLM compression, I'm now convinced and amazed that economics will dictate that all devices contain a copy of all information on the Internet.
I wrote a post about it: Your toaster will know mesopotamian history because it’s more expensive not too.
And as other commenter pointed out, a smart toaster with ads or data collection can be subsidized and thus be more profitable. (Oh what a world we're headed for!)
In any case, I think the LLM-everywhere thesis holds even strong for even moderate-complexity devices like power plugs, microwaves, and mobile phones.
I got excited about that, until I actually tried to download a model and run it locally and ask it questions. A current gen local LLM which is small enough to live on disk and fit in my laptop's RAM is very prone to hallucination of facts. Which makes it kind of useless.
Ask your local model a verifiable question - for example a list of tallest buildings in Europe. I did it with Gemma on my laptop, and after the top 3 they were all fake. I just tried that again with Gemma-4 on my iphone, and it did even worse - the 3 tallest buildings in Europe are apparently the Burj Khalifa, the Torre Glories and the Shanghai Tower.
I wouldn't call that effective compression of information.
Yea, it's not an encyclopedia of facts. Language models store the FEELS of the data in vectors (or angles in Gemma4's case, it's a cool thing) not the exact string.
But what you can do with local models is give them actual data and tools to search it. Download a copy of Wikipedia locally, give the agent a way to search it and BOOM accurate information without an internet connection.
Also "small enough to live on disk" is a bit vague, especially when models get super stupid super fast when you get to the smaller size. At that point they're just basically 40k servitors that can use tools and nothing much.
I don't think any LLMs are good at accurately regurgitating arbitrary facts, unless they happen to be very common in their training, and certainly not good at making novel comparisons between them.
My go-to technique has been to offer the offender a pair of headphones, saying something to insinuate that they must forgotten theirs or be too poor to afford them. Most of the time they say “oh I have headphones!” and then realize that they’ve outed themselves. (I stockpile the free headphones from gyms or airplanes, or get the $2 ones from AliExpress)
But on a plane they'll have already been asked by a flight attendant either by the time the plane takes off, or as soon as it stops climbing. So clearly this isn't working on the people this rule targets. In fact I think this rule is actually the ideal response from the airline and should be adopted everywhere, as anyone who is so unconcerned with the wellbeing of others as to play audio on their device without headphones shouldn't be allowed to fly, as they're obviously happy to fuck up everyone's day, and won't follow instructions.
This is actually a really good response though. Because the act of having a device blaring demonstrates contempt for everyone one around them. It's hard to act in a hateful way to someone who just offered you something for free.
I concur. And would just add two points:
(1) Make it that you’re not asking for anything / don’t open with something that could be perceived as a setup to asking for money, or pushing a religion. :)
2) be sensitive to social cues or that they want to be left alone, like terse answers or shifting their attention away from you
I’ve found it can be helpful to shift your own attention after someone answers you, but not to a phone (which just makes you look like you’re communicating with someone else).
Look at a flyer on the wall, or your beverage if you’re in a bar, and they’ll follow up if they want to talk and appreciate the reduced pressure either way.
And yeah, never open a conversation with something like “can I ask you a question?” which is usually a trick of a salesperson or beggar to make you acknowledge them and start saying yes.
This actually jives with my personal experience living in NYC.
New Yorkers have a reputation for being stone cold with strangers, but the truth is that anytime somebody approaches you out of the blue, there's an assumption that they're about to ask for money or try to get in your pants. Once you demonstrate you're not looking for either (or, if the second I suppose, that you're at least smooth enough for it not to be immediately evident), people are generally really kind. With some exceptions, I've usually found that the coldest looking person will stop to give a lost tourist directions if it's clear they're in need.
What motivation would the Supreme Court have to get “back on Trumps good side?” If anything, after his recent name calling of them I’d think they’d be less inclined to appease him. The can’t be fired.
> Just what the process was in the scam that snared Mr Williams remains unclear but it appears the scammers were able to access his card details and add them to a new Google Pay wallet on their own phone.
> "We're never going to truly know how Ian's card was breached. What we do know is that he wasn't in that Coles … and that he had the proof," Ms Brooks says.
This is more akin to a race car driver give a review of, for example, a new type of electric car. It doesn’t matter that the driver is not a domain expert in electric motors and regenerative braking; what matters is he knows how to operate these machines in their use case at the limits.
Hearing a programming legend weigh in on the latest programming tool seems entirely completely reasonable.
reply