ahknight

Super easy: don't give them a network and use a dedicated STB that's kept patched instead. Ta-done.

They never weren't important. What's important now is being prepared to act on them, which is expensive.

PHP did make it slow. They eventually wrote a tool to transpile PHP to C and then compile down to native.

I always consider AI-generated content to be a draft — or, in its parlance: something that takes the shape of the final product.

I don't dismiss the ability of models to create. I dismiss their ability to create finished goods. Humans should always be involved, especially with creative works. Models are great at modeling existing creations and solutions — head starts on long roads — but it's the person that needs to walk it and take the journey.

Thank you for your service.

This is exactly the use case. And if your indexing software uses it correctly you can also do "video deposition with Jane Doe where she angrily discussed John Doe's mother's behavior concerning his grandfather's estate." With audio embeddings, transcripts/captions, and periodic image embeddings contributing to tags on the media item you'd be golden.

Both of those use subject identification models, not embedding models (last I checked, and I run Frigate as well).

Everyone has their own struggles, yes. And we're all different, so it's often difficult to imagine the struggles other people have because we take certain things for granted.

For those of us with executive function disorders, yes, the amount of information being shoved at us in this modern world is well past the level where we can make sense of it easily, and we certainly can't keep track of it all. Assistants that can read and understand natural language and _assist_ us have been accessibility revolutions.

If you can do it all yourself, that's great. Not everyone can, and not everyone can have a person helping them or doing it for them all the time, either. Being able to just throw _life_ at something in a box and say "Help. Please." and get help? Indescribably useful.

Now if we can just fix the risk and privacy nightmare part we'll be golden.

I think it's because only Luna could be competitive on price due to its size, honestly. And if there's demand, then they bump it up later to other sizes.

It's still a predictive model under the hood, so order will always matter. They are also just as open to prompt injection and bias/loaded questions as the underlying model.

This has been around for years. It's just a rename. The code is open source and you or anyone else can go look at it (and many have).

`/goal` is the best thing ever for this. Ensure there's a condition and a way to test it and the harness will keep poking it until it can say the goal is met.

You also upload your code and conversations into their server-hosted DeltaDB service and the only way to remove it is to email their privacy contact.

Thanks but no.

He claimed the guy did no research while himself doing no research? I'm shocked! Shocked! Well, not that shocked.

If I don't know what I'm reading about by the third or fourth phrase, I skip it. Empty language comes from empty minds and I don't have time for that.

An obliterated 30B model versus a 1T model without guardrails is like comparing an angry squirrel to a bear having a bad day. One hurts, the other hurts until it abruptly doesn't.

They should put a Summary section at the top with the raw prompt.

Feels like it's the opposite of what it should be. It's fine to impose the world on the compression side, but decompression should be optimized for low-end devices. This is where deflate and zstd get it.

The inverse is true, however. Especially the "almost" part.

Many of us were there when yada yada white wizard. Hell, my first paid job was writing Perl CGI scripts. Well, script. Monolithic MFing thing spread across 100 files and running a whole company on text files with zero format documentation and RCS as version control. No need for CVS because the systems were incrementally archived with tar every hour...

Until they weren't. (I'm glad you saw that coming, because they didn't.)

This is also fair.

And direct access to system calls. And the uncanny ability to look like it's going to do what I want and then five minutes into a run wind up in a small town in Arkansas blind drunk and crying vaguely about fluffy red squirrels taking its sandwich.

Oh no.

Why would I want a service for this when I could easily run such a discovery service myself? Why not just release that tool to users? SyncThing offers its discovery server for self-hosting as an example.

Oh, right, this is how we monetize in 2026. Nevermind.

Perl is just C with less type safety.

And yes, for the most part. C++ as simple shorthand for struct-attached functions and automatic memory management (no, not smart pointers; RAAI) is good. Every single thing added after that is misery and should push a modern developer to Rust, Go, or Zig (roughly in that order) where such things are implemented sanely or not at all.

I say this about C every day.

I end my user CLAUDE.md files with:

"- I'm not always going to read every word, so end each summary message with a TLDR of what you found, what you recommend, and what you need from me."

It works really well.

Sol doesn't need it at all.

Back when the point of Pixel was to tell the world: THIS is what you can do with Android.

Then Samsung got the message, ran with it, and Google went "Ehh, we're good now."

And in parallel we had FIDONet. Mustn't forget that magical place and source of many a late night and large phone bill.

My heavily-upgraded M1 Max came in slightly over that when I got it five years ago. (Still going very, very strong.)

This new Studio? Can't find a config under $5k I'd bother with. But for the MBPs that number still mostly tracks for the average Pro user. (I buy large and run it into the ground so long I mistake the ground for the computer's remains.)