skiing_crawling

I've been using Siri (via homekit) to turn all my lights on and off for about 3 years now. It's steadily getting worse and worse as somehow, Siri is becoming less accurate and Apple is failing to adopt this new technology in a timely fashion.

I would like to tell it to turn off certain light in a certain room, but unless I get the exact string name of those light correct when I speak, Siri doesn't know what I'm talking about. And it can't do multiple things in a command. I can't say "turn off all the lights in XYZ room" or turn of "this light and this light".

Meanwhile, I can vaguely tell a computer behind my tv to do very complicated things (build me an service that ...) and it can execute on it fairly well. But in apple's "product vision" which I am apparently too dumb to decide for myself what I want, I can't ask for two lights to be turned off at the same time.

AI didn't start this, journalist have been using wordplay to "technically tell the truth" forever.

what does it have to do with git?

How does git replace a vector db search exactly? They are orthogonal. Are you gonna burn a million tokens every time you wanna find some relevant files?

Its probably an antipattern on a car to need an OTA "safety" update in the first place.

I was surprised to learn that some skills don't even describe exactly what steps to take or what to do. They just kind of give a motivational speech which I guess primes the model to output better text for a certain task.

https://github.com/anthropics/claude-code/blob/main/plugins/...

This frontend design skill that claude uses basically just begs it to pick nice fonts and make the design coherent. No specifics about which fonts or how to make nice color schemes and layout.

I bought a 8gb m1 air just a few days ago to use as a travel laptop. The 8gb gives me memory anxiety coming from my 48gb m4, but it did force me to turn off some settings I never liked (siri, spotlight indexing) and I also discovered zed, ghostty, and orbstack to replace vscode, iterm, and docker desktop.

The memory limit is probably in my head now, it does pretty well as long as I'm not obsessing over activity monitor.

it looks fake. claude doesn't talk like that normally.

Same for me. I find through hole easy and can do it reliable. Soldering 2 wires together...I always mess it up. But its something I rarely do.

My favorite useless error message is in chrome/brave: `DNS_PROBE_POSSIBLE`.

If its possible, just do it?

I don't understand the water argument? I thought places like data centers would use something akin a closed loop and a radiator. Are they taking water from the town pipe, heating it once, and then launching it into the sun? Is there some other use for the water other than cooling?

How will they know what's encrypted? Maybe I just like sending random sequences of bytes across the wire

> At the same time, humans can move up the abstraction ladder faster than the LLMs can

This was kind of the point, its only true for now. I agree with you that this kind of stuff will take longer. I don't think there's probably good training data for it right now. Handling abstractions and course correcting is probably the job now, and it also happens to be exactly the data that we will be typing in our prompts. They'll train on it or something like it.

As software engineer, we should collectively realize that this is all cope. Every article or comment about how AI will never be smart enough, etc, etc will only be true until its not. One of our main valuable skill sets is now partially automated. Some of us are completely obsolete and its coming for the specialists and more experienced ones within a decade tops. You're not going to convince anyone that "um actually we're better because we bike shed more".

Stuff like this is ridiculous and comes off as frantically trying to save your ass. Its pretty obvious at this point that we will just throw more matmuls at it until it can do this or something equivalent.

> Agents cannot do osmosis. They do not get context by being in the room, by half-hearing the planning conversation, or by carrying the memory of the last incident.

I noticed they offer all the storage upgrades at or below retail cost. Nice.

The research is wrong because people are buying the iPad on dash cars. The data collection, sampling, or methodology of the research presents a different result than the market reality of what people actually bought with their money.

Ive been screaming this since touch screens in cars started catching on. We are surely beyond believing automation is done for convenience and betterment of our lifestyles at this point. It's all to create more time and a new surface for us to serve ads on. The average consumer is completely infatuated with the idea of their car being a doom scrolling device, so we get dragged further into hell.

they do the most obvious, sorely missing feature after over a decade of stubbornness and it goes straight to the top of HN

That actually sounds nice in theory. A model for getting work done and nothing else, which doesn't have to account or be trained for any type of user engagement. Like how chatgpt/claude are wasting their capacity on social niceties and glazing the user. I don't know if granite is one such but I'd bet that many of the other popular models can't be since they are consumer facing.

What the value of publishing purely LLM generated content? Anyone can prompt the same thing out of it.

I used to run qwen3.5 27b Q4_k_M on a single 3090 with these llama-server flags successfully: `-ngl 99 -c 262144 -fa on --cache-type-k q4_0 --cache-type-v q4_0`

I can't understand why this is a surprise to anyone. An iphone is still a computer, of course it can run any model that fits in storage albiet very slowly. The implementation is impressive I guess but I don't see how this is a novel capability. And for 0.6t/s, its not a cost efficient hardware for doing it. The iphone can also render pixar movies if you let it run long enough, mine bitcoin with a pathetic hashrate, and do weather simulations but not in time for the forecast to be relevant.

What would be really awesome is if it could use the server its hosted on's GPUs. I have a multi GPU server and it would be great to be able to edit videos from my table or couch without spinning up my laptop so hard.

An interesting point is, SoundTouch has always had a local API available (publicly documented) on a web server that runs on a speaker (at least for the last 10+ years).

A "single server" covers a pretty large range of scale, its more about how F-droid is used and perceived. Package repos are infrastructure, and reliability is important. A server behind someone's TV is much more susceptible to power outages, network issues, accidents, and tampering. Again, I don't know that's the case since they didn't really say anything specific.

> not hosted in just any data center where commodity hardware is managed by some unknown staff

I took this to mean it's not in a colo facility either, assumed it mean't someone's home, AKA residential power and internet.

I never questioned or thought twice about F-Droid's trustworthiness until I read that. It makes it sound like a very amateurish operation.

I had passively assumed something like this would be a Cloud VM + DB + buckets. The "hardware upgrade" they are talking about would have been a couple clicks to change the VM type, a total nothingburger. Now I can only imagine a janky setup in some random (to me) guy's closet.

In any case, I'm more curious to know exactly what kind hardware is required for F-Droid, they didn't mention any specifics about CPU, Memory, Storage etc.