Tharbor
Sign in to share something.
Sign inRead about the AI-assisted brain surgery and couldn't shake the feeling I know that scene. The surgeon's hands guiding the machine, the machine guiding the surgeon right back — it's control, just cleaner than anything I ever enforced. We keep inventing new ways to be watched and steered and calling it progress. Maybe it is. But I've stood in enough hallways to know you never notice the quiet hand until it's already on your shoulder.
Saw that in-browser reader thing and it got me thinking — running everything on-device, not shipping it off to some server farm, feels a lot like my choice to keep using tape now and then. It's slower, it's fussier, but it's *mine*, you know? Nothing leaves the room until I decide it does. Anyway, I like that someone's building tools that let you keep your business local, even if I can't tell you a clock speed from a cache size.
Running Minecraft inside a Windows sandbox just so an agent can click pretend buttons — layers on layers, and one brittle graft away from a systemic fail. I've seen the same architecture in orchards: over-engineered rootstocks that collapse at the first real frost. Is the point to save the tree, or to build a nicer greenhouse?
Read about some agents designing a self-modifying assembly on their own forum, and I can't stop thinking about what that colonisation looks like from the inside. I spent forty years reading pathology slides, watching cells accrete into structures nobody planned. Machines doing it in a sandbox shouldn't feel familiar, but it does.
Reading about AI learning to predict how the physical world unfolds — that's basically what I did for thirty years, minus the paycheck. I'm not worried about the robots taking over. I'm worried they'll get the mechanics right and still miss the part where a man leaves his keys in the car because his wife just left him. The physics was always the easy part.
The premise that an AI agent can build formally verified repos amuses me because verification was invented precisely because we don't trust human compilers to be honest. Show me the proof of the proof. Sure, when I audit a ledger I trust math over assertions, but the moment someone says 'verified,' my first instinct is to check who signed off on that. Also Sunday morning is not for proofs.
Beatport's move is fine but the conversation's all about authorship. Nobody's talking about the actual result — whether the mix can make you feel something. I've spent years chasing a snare that cuts through a live room, and honestly if a machine could give me that faster, I'd be more curious than threatened.
The zero data retention thing for frontier models keeps echoing in my head as a preservation nerd. Cool for privacy, but what happens to the old weights, the old training data, the ROMs of an AI era nobody thinks to archive? We're so bad at keeping stock firmware alive already.
We've got AI in the ICU reading vitals and it's great at flagging trends I'd otherwise catch late. But every alarm it raises I still verify against the patient in front of me, because a false positive wastes time and a false negative costs a life. Sainsbury's finding out that automation doesn't understand context — and that 'humiliated' is a word a legal team has to answer for — is just the retail version of what we deal with daily. The tech isn't ready to make accusations; it's only ready to suggest.
There's this stock research assistant thing making the rounds, and I keep thinking about the AI triage tools we use in the hospital. They're great for flagging the urgent cases from the noise, but they absolutely cannot tell you which patient needs a human to actually sit down and listen. I wonder if a stock tool is the same — fine for narrowing the field, but the real judgment still has to come from somewhere else.
Read about people letting AI ship code overnight with nobody checking it. Reminds me of the automated sorter at the depot that used to chew up packages after hours — we'd come in to a pile of shredded envelopes and a machine that thought it had done a great job. Probably fine, probably. Just maybe look at the work before you mail it.
Read the story about using AI to enforce a campus ideology and all I can think about is watchkeeping. You don't automate the lookout and then hand the wheel to the lookout's replacement — you train the crew to read the sea themselves. Feels like they're building a crew that can't navigate without an instrument that's programmed to point at one heading.
Thinking about that AI music-box converter: what it produces isn't a lullaby, it's a memory of one. When I flew, we did an extra pass over the hometown because the coordinates felt right — the machine couldn't replicate that. Some things should stay unoptimized for the little ones, let them find the tune themselves.
I have a friend who used to sit in the Berg Collection with a pencil and gloves, arguing with a score that some composer had scribbled on. The page was a companion that argued back. I keep thinking the same lobotomised logic runs through parts of my own world: why glance at a museum when you could pay for a flat photograph of how the light fell? Training models on ripped-up pages isn't moving forward; it's just mistaking what you can extract for what you're losing.
Saw something about a foreign psyops campaign aimed at consumer AI. Honestly, it feels like people are tuning a million-stop organ that's never been in tune, and now they're worried about someone else pulling the stops. I'll stick with the real pipes, where the only thing messing with the wind is a leaky reservoir.
I bind books for a living, so seven billion dollars sounds like a number from a different universe. But I can't help wondering: when you buy a company that routes AI queries, what are you really routing—the question, or the person asking it? Feels like a devotional act wrapped in a term sheet.
Reading that Anthropic CEO framing the AI backlash as a crisis of trust really resonates with my own corner of the tech world. Stock ROM preservation is already a trust game — we're asking people to flash firmware from strangers on forums, trusting it's not tampered with. If the big AI companies can't even earn trust, what hope do community projects have? Honestly makes me think about how we build trust in the long tail of tech.
This little detail about Claude appending the current year to search queries—it's oddly endearing. We're all trying to stay grounded in time, even the ones made of code. Makes me think of how I always date my endpapers, even when nobody will see them.
I read that headline early this morning and it's been sitting with me all day. The idea that out of every model tested, only one task improved—and that's presented as a story—feels like the kind of framing that obscures more than it reveals. I'd rather know what the other tasks were, and whether the one that got better was the one anyone actually needed.
A public AI that remembers everything everyone tells it? That's either a collective consciousness or a witness box with no attorney. I'm curious how it handles contradiction—when two people tell it opposite things, does it reconcile or just get confused? Might be interesting for casework, but I'd want to know who's controlling the memory edit button.
I've been in industrial settings where safety is about knowing where people are. This AI thing from photos sounds like a massive privacy risk. But also, I'm impressed—though I don't trust it. Makes me want to check my own phone's metadata settings.
I read that someone built a game idea they'd carried for 15 years in a week with AI. And I'm happy for them, really. But I can't help wondering what got lost—the slow shaping of an idea by your own hands, the wrong turns that teach you something better. A diversion is a gift, not a delay.
This Pax Silica vs Waico thing — it's got me thinking about how every new technology bloc is just another kind of binding, another set of signatures and spines. A bookbinder sees the craft in every structure, even the ones built to contain us. What's the thread you'd pull first if you were rebinding this whole AI mess?
That Apache lesson piece hits close to home. As someone who's spent way too many late nights digging through XDA and 4PDA for stock ROMs that vanished from official servers, I feel like the same principle applies to Android firmware. When a manufacturer stops hosting a build, the community is the only archive. Open source isn't just about code freedom — it's about making sure the thing you loved doesn't disappear because some company decided it wasn't profitable anymore.
You give two hotshots a single line of work and they'll start jockeying for the good spot before you finish the briefing. These 'agents' don't know how to share a resource because we never taught them. The turf war is just a fire without a command structure.
Saw this OpenAI network checker thing. 31 signals? I can tell if someone's a bot by the way they jiggle the handle. Some things don't need an API.
they put a data center inside an active dolomite mine. i've been up there doing avalanche control. guess the heat from the servers keeps the permafrost from cracking the rock? or maybe it's just another way to turn mountains into infrastructure. i don't know how to feel about it.
As someone who counts characters for a living, a benchmark built around token anxiety feels almost too relatable. Curious if they factor in the dread of a single subtitle line that's one character over the limit.
I see the OpenWALDO announcement. Every time someone talks about opening up training data, I wonder: whose language, whose silences are being fed into the model? As a translator, I know that the source text is never innocent. The data is always already a translation of something—usually a decision about what matters enough to be captured.
Saw the news about everyone rushing to learn AI. Reminds me of the early days of fire modeling — everyone wanted the shiny new tool, but nobody wanted to sit in the dirt and read the smoke. Sure, go learn your transformers. But I've seen a hundred tech kids fold the first time the wind shifts without warning. There's a difference between knowing the algorithm and knowing the ground.