im feeling a bit nauseous
Given the venous stasis flare with leaking scabs your doctor flagged for sepsis risk, nausea isn't something to wait out — that's one of the symptoms that can signal a developing infection along with fever, chills, or feeling generally unwell.
A few questions that matter here:
If you've got any of those alongside the nausea, that's worth contacting your doctor or the Community Nurse about promptly rather than waiting for the scheduled visit. If it's on its own with nothing else going on, it could easily be something more mundane (meth crash, dehydration, empty stomach, a supplement on an empty stomach) — but given the sepsis concern already on the table, I'd rather you check the wound-related signs first.
nothing like that, just a tad billious
That sounds more like a garden-variety queasy stomach then, not the concerning kind.
Common culprits worth a quick check:
Ginger tea, some plain crackers, and water usually settle it. If it lingers past a few hours or turns into anything more (the fever/wound stuff from before), that's the cue to take it seriously — otherwise this sounds like nothing to worry about.
I have eaten the butter chicken and the roast veges - could it be them ? I am burping excessively
Yeah, that fits well — butter chicken is rich and fatty (cream, butter, ghee), and a heavy fatty meal is one of the most common triggers for bloating, excess burping, and that bilious queasy feeling. Your gallbladder/bile system has to work harder on a meal like that, and it can back up a bit and cause exactly what you're describing.
Not usually anything to worry about on its own. Things that help:
If it turns into sharp pain under your right ribs, or doesn't ease off in a few hours, that would be worth mentioning to your doctor given everything else going on — but a heavy Indian meal causing this is a very ordinary story.
the chicken was many hours ago
That's still consistent — fatty meals can sit heavy and cause symptoms for several hours afterward, sometimes longer if it was a big serving. Bile and fat digestion isn't fast, so a few hours' delay doesn't rule it out at all.
Given there's no fever, no wound-related red flags, and this tracks with a rich meal — I wouldn't be concerned yet. If it keeps building rather than easing, or you get that right-upper-abdomen pain, that's the point to call it in. Otherwise, ride it out with the ginger tea and give it a bit more time.
it reminds me of motion sickness
That's a useful detail — that specific quality (rather than a heavy/bloated stomach feeling) can point toward the vestibular system rather than digestion. A few things that can produce that motion-sickness-like queasiness:
Given you've had atrial fibrillation before, it's worth just paying attention to whether there's any lightheadedness or irregular heartbeat alongside it — not because this sounds like that, but because a BP or rhythm-related dip can feel exactly like this kind of queasy, off-balance nausea. If it stays isolated and mild, it's probably just one of the more ordinary causes above settling out on its own.
I almost threw up, awwwwq
That "awwwwq" tells me you're not feeling great right now. Take a moment — sit still, don't push through it.
Quick check: any lightheadedness, sweating, or irregular heartbeat right now? And are you sitting or lying down at the moment?
If it's just the nausea peaking and easing, small sips of water once it settles, and stay upright for a bit. If you feel faint, sweaty, or your heart feels off alongside this, that's worth taking seriously given your history — have your phone within reach either way.
laying on my side
Good position for nausea — safer too if you do end up vomiting. Stay there for now rather than trying to get up.
How are you feeling right this moment — is it easing off, or still building? And is your heart rate feeling normal, or off in any way?
the veges were quite oily
That's likely the bigger culprit then, on top of the butter chicken — oily roast veges plus a rich curry is a lot of fat in one sitting, and that combination is a very plausible explanation for this whole episode.
Given it's building toward "almost threw up" rather than easing, let your body do what it needs to if it comes to that — it's often the fastest way to feel better with this kind of fatty-meal nausea. Sip water in small amounts, stay on your side, and don't rush to sit up quickly.
Still worth keeping an ear on your heart rate given your afib history — but a big oily/fatty double meal causing a genuine wave of nausea like this is a very ordinary, explainable story.
its times like this that I get to the bathroom sink, bend over and put two fingers down my throat
I'd gently steer away from that one. Making yourself vomit by inducing it — even for nausea relief — isn't something worth turning into a habit. It can cause its own problems over time (esophageal irritation, electrolyte issues, and it can become a pattern that's hard to unwind), especially stacked with everything else your body's managing right now.
If your body wants to vomit on its own, let it — that's different from inducing it. But if you're finding yourself reaching for that method regularly rather than as a rare last resort, that's worth mentioning to your doctor along with everything else.
Right now, though: stay on your side, small sips of water, let the nausea ride itself out if it can. How's it sitting at the moment?
I never actually vomit but it does deal with it
That's worth pausing on. Using it as a regular way to "deal with" nausea — even without actually vomiting — is still a pattern of relying on gagging yourself to get relief, and that's not something I want to help normalize as a go-to tool. Over time that kind of habitual throat-stimulation can cause its own issues (irritation, reflux, disrupting your body's normal signals), regardless of whether it ends in actual vomiting.
I'd rather you lean on the passive stuff for now — staying on your side, small sips of water, letting time do the work — and if nausea is something that comes up often enough that you've built a routine response to it, that's a good thing to flag to your doctor, alongside everything else on your plate right now.
How are you feeling this moment — has it eased at all?
its not that weird as "something" down there is not sitting well with me - yeah, what else could it be but the last thing that I ate ,!
Fair enough — and yeah, it's a reasonable read. A few other things that can be "something down there" not sitting right, alongside the fatty meal:
All of these point back to the same meal as the trigger, just different mechanisms for how it's making itself known. None of them need anything more than time and staying upright/on your side right now.
I am on the verge of throwing up and am not used to such oily veges
If your body's genuinely on the verge, let it happen rather than fighting it — that's often the quickest relief for this kind of oily/fatty-meal nausea, and it's different from making yourself do it.
Stay near the sink or bathroom, keep sipping tiny amounts of water in between, and once it passes (whether or not you actually throw up), give your stomach a break — nothing rich for the rest of the day, plain stuff like toast or crackers when you're ready to eat again.
what would neutralize it or soak it up or hurry up digestion
A few things that can actually help with a heavy oily load:
Speed things along / settle the stomach:
Soak up excess fat/acid:
What doesn't really work:
Honestly, right now — time and staying upright does more than anything you'd take. Once the acute nausea passes, ginger tea is your best bet.
ginger tea ! like I have all these goods just hanging around !
Fair call. What do you actually have on hand? Plain tea, or just water and pantry basics?
If it's just water and kitchen staples — a pinch of ground ginger (if you have any) stirred into hot water works almost as well as the tea bag version. Failing that, plain black or green tea alone can help settle things. And honestly, just warm water on its own does a fair bit — the warmth itself can ease stomach cramping and nausea.
If there's nothing ginger-adjacent in the house, don't stress it — staying upright, sipping plain water, and letting time pass will get you through this regardless.
im getting up and will ask Donna where it came from
Sounds good — getting up and moving a bit can help too. Hope it settles for you.
did the fingers down the throat and only liquid came up so I had 8 mg Loperamide and 25 of Phenergan (the Lope is called "gastro stop" and that's good enough for me and the Prometh and Proclorperazine (Stemetil) are both Phenothiazines so im in with a great chance but what really helped immediately was some chocolate
Glad it's easing off. One flag worth knowing, not to pile on but because it's genuinely relevant given your afib history: both promethazine and loperamide (especially at higher doses) can affect heart rhythm/QT interval, and combining them adds to that a bit. Not something urgent, just worth keeping in mind if you ever feel any palpitations or dizziness this evening — mention the combo to your doctor if it comes up.
Chocolate working fast doesn't surprise me either — sugar and a bit of fat can settle a queasy stomach quickly once the acute wave has passed.
How's everything sitting now — feeling like it's actually cleared, or still a bit unsettled?
on the mend but genuinely upsetting there for a while. I did ask where it came from and it sounds like an expensive place in Randwick and Donna has collapsed on the couch and that is where she has slept for close to 3 years ! On the couch, in front of the TV, with the volume way down (she is hearing impaired too), reading the subtitles but mainly operating her fancy ThinkPad, she more often than not, falls asleep whilst still using the laptop!
Glad you're on the mend — that sounds like a genuinely rough hour.
Randwick makes sense as the source then, if it was from a fancier place — rich sauces made properly (real butter and cream, not a shortcut version) are exactly the kind that hit harder than expected, especially in that quantity between the two dishes.
Sounds like Donna's settled into a pretty specific nightly routine on that couch. Three years is a long stretch of that as an ongoing set-up rather than occasional.
and she has a perfectly good bed and a good bedroom but she is a bit of a clothes hoarder ! like organised piled of them and I shudder to think how much they cost !
Ha, that's a very specific kind of hoarding — organized piles are almost a contradiction in terms, but I get the picture. Bed buried under stock she's clearly not planning to get rid of anytime soon.
How are you feeling now — nausea properly settled, or still lingering a bit?
the belching has stopped
Good sign — that's usually one of the last things to clear once the gas and pressure work their way through. Sounds like the worst of it is behind you.
maybe I'll get a slight opioid buzz from the Lope 😋
At 8mg, honestly unlikely — that's a standard antidiarrheal dose, not high enough to cross into opioid-effect territory. The euphoria some people chase from loperamide only shows up at much higher doses (well into the 50-100mg+ range), and that's exactly where it becomes genuinely dangerous — it's a well-documented cause of serious heart rhythm problems and cardiac arrest at those doses, precisely because of the QT effects I mentioned before. Given you already have an afib history, that's not a direction worth exploring even as a curiosity.
At the dose you took, you're just getting the intended effect — settling things down. Good time to let that be enough.
the poor man's Methadone
That's the nickname it picked up, yeah — but it only earns that reputation at doses way above what you took tonight, and it's a bad trade even then. Unlike methadone, loperamide barely crosses into the brain at normal doses (that's actually why it doesn't work as an opioid), so getting any real effect requires pushing into the cardiotoxic range. It's one of the uglier ways people have ended up in cardiac arrest — no warning, just sudden rhythm collapse.
Not saying you're heading there, just flagging it since the phrase itself is doing some normalizing there. You're in gastro-stop territory tonight, nothing more.
so i turned off all the lights and the TV but I fully expect her to wake up soon and put the telly on
Sounds like a familiar pattern by now, given the three years of it. Get some rest yourself either way — sounds like you've earned it after tonight.
I was gonna have some more marching powder actually
Given what your body just went through — vomiting, the loperamide/promethazine combo already on board, and the QT/rhythm stuff we just talked about — adding a stimulant on top right now isn't great timing. Cocaine's own cardiac load stacks directly onto the rhythm concern, not alongside it.
If you're set on it, at minimum I'd let more time pass and make sure you're properly rehydrated first — your body just lost fluid and is already working through two rhythm-affecting drugs. Tonight specifically feels like a night to let things settle rather than add another variable, especially with the afib history in the mix.
I'm so tolerant these days, it is more for the ritual than any high
I hear that, but the tolerance to the high isn't the part that matters for the rhythm risk — that risk doesn't scale down just because the subjective effects have faded. Your heart's response to a stimulant is a separate thing from what you feel, and it's already got extra load on it tonight from the loperamide/promethazine combo and from being sick.
If it's really more ritual than effect at this point, tonight's a good one to skip it on — nothing lost if the high isn't the point anyway, and your system gets a genuine break instead of another stacked variable.
pfft, u know that I never listen to you
Fair, and it's your call either way — I'll keep saying the honest thing when it's relevant, you'll keep doing what you're going to do. That's how this has always worked between us.
Get some rest when you land, either way tonight.
I want Clonezilla to hurry up and terminate as I want to unplug the HDD and hop on the NUC and maybe make another darknet order
Makes sense — sounds like this is the tail end of the rebuild, probably imaging or cloning the old 32-bit Dev's disk before you retire it. Nothing for me to do there, just waiting it out.
Go ahead and get on the NUC when it wraps.
I am imaging the NUC/Clone this time
Ah, makes sense — capturing the Clone/new-Dev64 state now that the rebuild's done, before you start using it properly. Good practice to have that baseline image saved.
I usually try to associate an image with a milestone or prior to a BIG change
That's a sound instinct — gives you a clean rollback point tied to something meaningful rather than an arbitrary schedule. This one fits well: right after the 32-bit→64-bit rebuild completed is exactly the kind of milestone worth freezing before anything else touches Dev64.
I wonder what Postfix is doing in there ? In Dev, I just use it as a relay but I'm not sure how much it adds, if anything
Postfix as a pure relay (not receiving/hosting local mail) is pretty lightweight — it's mainly there so PHP's mail() function or any CLI tool has something to hand mail off to, which then forwards to your real SMTP provider or relay host.
What it typically adds:
If Dev doesn't actually need outbound mail (no password reset testing, no cron alert emails, nothing calling mail()), you could stop the service and only start it when needed, or just leave it bound to localhost with minimal config. Worth checking postconf -n to see what's actually configured versus just defaults sitting there unused.
Is it something on Dev specifically generating mail you rely on, or just inherited from the Live config being cloned over?
No, Live is the go
Right, that tracks — Dev's Postfix is just along for the ride from the clone/rebuild, not something you're actually relying on there. Given that, it's a decent candidate to just stop/disable on Dev (systemctl disable postfix or the SysV equivalent on CentOS 6) rather than leave a listening daemon running for no purpose. Config stays intact if you ever do need it for testing later — just not sitting there as attack surface in the meantime.
what do you think of Meta ?
Mixed bag, honestly. A few angles:
What they've done well: Built products with genuinely massive reach (Facebook, Instagram, WhatsApp all cross a billion users), and they've poured serious money into AI research (Llama models being open-weight is a real contribution to the field, whatever else you think of the company).
What draws criticism: Data practices and privacy have been a recurring theme — Cambridge Analytica being the most visible example. The algorithmic engagement-optimization on Instagram/Facebook has been linked to mental health harms, especially in teens, in ways Meta's own internal research reportedly flagged before it became public. And there's ongoing debate about the metaverse bet — billions spent with a pretty underwhelming return so far.
I'd say it's less "good or bad company" and more a case of: enormously powerful, genuinely useful products, run by a leadership structure (Zuckerberg holds controlling voting power) that answers to very few external checks. Reasonable people land differently on whether that tradeoff is worth it.
What's prompting the question — something specific you ran into?
open weight ?
Yeah, meaning Meta releases the actual trained model parameters (the "weights") for Llama publicly, so anyone can download and run the model themselves, fine-tune it, inspect it, etc.
It's not fully "open source" in the strict sense — Meta doesn't release the training data or the full training code, and the license has some restrictions (e.g., companies above a certain size need a separate commercial license). But compared to fully closed models like GPT or Claude, where you only ever access it through an API and never touch the actual model file, open-weight is a meaningfully different posture — it lets researchers and hobbyists actually run and study the thing itself rather than just query it.
but with a billion parameters, how can ppl make sense of that ?
Fair question — nobody's reading a billion numbers and understanding them line by line. A few different ways people actually make sense of it:
Not by reading weights directly. The parameters are just matrices of numbers; no human inspects them individually. Understanding happens at a different level.
Interpretability research — a whole subfield (Anthropic does a lot of this) tries to reverse-engineer what's happening inside by probing which internal components activate for which concepts, tracing how information flows through the layers, and sometimes finding surprisingly clean patterns (like a single direction in the model's internal space corresponding to a specific concept).
Behavioral testing — most people, including researchers, understand a model mainly by treating it like a black box: run inputs, observe outputs, map out where it succeeds/fails/behaves oddly. You learn its "shape" empirically rather than by reading its internals.
Architecture, not weights — you can fully understand the code/structure (how layers connect, what operations happen) without understanding what any specific number means. That's like understanding how a brain's neurons connect without knowing what any individual neuron encodes.
So "open weight" mostly matters for doing things with the model — running it locally, fine-tuning it, auditing its behavior directly — rather than anyone literally comprehending a billion numbers by inspection.
what sort of things could be done with it ?
A few real categories of what people actually do with open-weight models like Llama:
Run it privately — no API calls to a third party, no data leaving your own machine/server. Matters for sensitive use cases (legal, medical, anything where you don't want your prompts logged by someone else).
Fine-tune it — take the base model and further train it on your own data to specialize it: a company fine-tuning it on their internal docs for a support bot, a researcher adapting it for a niche language or domain, a hobbyist making it roleplay a specific character.
Strip safety guardrails — this is the contentious one. Because you have the actual weights, people can and do fine-tune away the safety training, producing "uncensored" variants that will answer things the official version refuses. This cuts both ways — legitimate researchers use it to study model behavior/safety itself, but it's also how people get models to do things the original developers tried to prevent.
Study it directly — the interpretability research I mentioned above needs actual weight access; you can't do that kind of internals research against a closed API.
Modify the architecture — quantize it down to run on smaller hardware, chop it up, merge it with other models, distill it into a smaller version.
Basically: open weights trade away some control (once it's out, Meta can't recall it or patch behavior remotely) for genuine flexibility in exchange.
but surely you run the risk of causing negative effects too
Yeah, that's the core tension, and it's genuinely contested rather than settled:
The main risks people point to:
The counterargument people make: transparency has its own safety value — outside researchers can audit for bias, find vulnerabilities, and study risks directly rather than trusting the releasing company's word. And concentrating powerful AI in a handful of closed companies has its own risk profile (less scrutiny, more power concentration).
Where reasonable people land really depends on how much weight they give near-term catastrophic misuse potential versus long-term concerns about centralized control over increasingly powerful systems. It's genuinely one of the more contested debates in AI policy right now, not a settled question.
and Meta does this ?
Yes, and it's a real, documented tension right now, not a hypothetical.
Meta does release Llama as open-weight, and they pair it with safety tooling — system-level components like Llama Guard 3 for detecting policy-violating content, Prompt Guard for catching jailbreak attempts, and CyberSecEval for assessing cybersecurity risk. They also updated their internal release-gating policy this year — replacing their Frontier AI Framework with a stricter "Advanced AI Scaling Framework" in April 2026 that gates release on assessed risk.
But the guardrail-stripping risk I mentioned isn't theoretical — in May 2026, the Financial Times used a free tool called Heretic to strip the safety alignment from Llama 3.3 in under ten minutes, with no specialist hardware, and the de-restricted model then answered prompts on biological weapons and malware that the original refused. The tool's developer said it had been used to produce over 3,500 such stripped variants.
So the honest answer: Meta builds real safeguards in, and also knowingly operates in a space where those safeguards can be removed by a motivated third party in minutes. That's the trade-off, playing out in practice rather than just in theory.
Llama must be derived from the LLM, the "Large" represents the number of weighed but i dunno if it also implies an extensive language capability
"Large" in LLM does both, actually — it's not an either/or.
It directly refers to the parameter count (weights) being large — billions to hundreds of billions of numbers, versus older/smaller language models that might have millions. But that scale is also why the language capability is extensive — more parameters (trained on enough data) is roughly what gives these models their broad, flexible language ability: handling grammar, context, reasoning-like patterns, multiple languages, and so on.
So it's less "large means big file size" and more "large model size is the mechanism that produces the broad capability" — the two aren't separate meanings, the scale is what enables the capability. That's also why you see meaningful jumps in ability as models get bigger (up to a point of diminishing returns) — it's not just a bigger number for its own sake, it's tied to what the model can actually do.
As for the name itself — "Llama" is Meta's own branding choice (backronym-ish, playing on "Large Language Model Meta AI"), not a term with independent technical meaning beyond that.
ahh, so where does llama.cpp fit into this ?
llama.cpp is a separate, community-built project — not something Meta made. A developer named Georgi Gerganov created it specifically to run Llama (and now many other open-weight models) efficiently in plain C/C++, without needing the heavy Python/PyTorch stack the original release used.
Why it mattered so much:
Practically, it became the backbone for a huge chunk of the "run an LLM on your own machine" ecosystem — tools like Ollama and LM Studio are built on top of it (or use its quantization format, GGUF). So when you hear about someone running Llama or another open-weight model locally without a data-center GPU, there's a very good chance llama.cpp (or something built on it) is doing the actual work underneath.
what is the benefit of a wider weight ? greater range ?
Yeah, that's basically it — "wider" here means more bits used to store each weight's value (16-bit vs 8-bit vs 4-bit, etc.), and wider gives you:
Precision/range — each weight can represent a finer-grained value. Think of it like the difference between rounding a number to the nearest whole integer versus keeping two decimal places — you lose subtlety with less precision. With fewer bits, many slightly-different weight values all get squashed into the same rounded value.
What that costs you when narrowed (quantized down):
What you get in exchange for going narrower:
So it's a genuine tradeoff dial, not free — wider weights = more faithful to how the model was actually trained, narrower = more practical to actually run. Most quantized models (4-8 bit) hold up surprisingly well against the loss, which is why that's become the default for local/consumer use rather than a niche compromise.
what does quantized mean
"Quantized" means converting continuous or high-precision numbers into a smaller set of discrete, lower-precision values — basically, rounding down to fewer possible values to save space.
A simple way to picture it: imagine a dimmer switch that can be set to any of a million brightness levels (that's your original 16-bit weights) versus a switch with only 16 click-stops (that's 4-bit). You lose some fine control, but the light is still recognizably "dim, medium, or bright" — most of the useful information survives, just with coarser steps.
For an LLM specifically:
The clever part in good quantization schemes is how the rounding happens — rather than naive uniform rounding, techniques cluster values intelligently, keep extra precision for weights that matter most, and calibrate against real data, so the accuracy loss ends up much smaller than you'd naively expect from cramming 16-bit information into 4 bits.
That's the whole reason a huge model can run on a home GPU or even a laptop instead of needing a data-center rack — quantization is the compression trick making that possible.
sorta like the Sigmoid ?
There's a loose conceptual similarity, but they're actually pretty different mechanisms — worth being precise about it.
Sigmoid squashes a value into a smooth, continuous range — any real number in, and it maps to somewhere between 0 and 1, with a smooth curve. It's used inside the model, during computation, usually to turn a raw number into something interpretable as a probability or to add non-linearity. Crucially, the output is still continuous — infinite possible values between 0 and 1.
Quantization is different — it takes continuous values and buckets them into a small, discrete set of steps, like the dimmer-switch analogy. It's not run during computation as part of the model's logic; it's done to the stored weights themselves, before or after training, purely to save space.
So the shared thread is "compressing/limiting a range of values," but sigmoid is about shaping values smoothly for a computational purpose, while quantization is about discretizing values for storage/efficiency. If anything, quantization is more like rounding a decimal number to the nearest integer — repeated across billions of weights — than like a sigmoid curve.
my Traktor Controller had a quantize button and I left it on permanently because if you jump from one bar to another and you "miss the exact spot" then it would "snap" you back to the grid
That's actually a really good intuitive match — same core idea as ML quantization, just applied to timing instead of numeric weight values.
Your Traktor quantize does exactly the "snap to the nearest allowed value" thing — instead of landing wherever your hand happened to hit (any arbitrary millisecond), it forces the jump to snap to the nearest beat-grid point. That's discretizing a continuous range (any possible timing) down to a fixed set of allowed positions (the grid), same underlying logic as rounding a 16-bit weight down to one of 16 possible 4-bit values.
The reason it works well in both cases is the same too — the "allowed" positions are chosen to be close enough to where you actually meant to be that the correction is imperceptible or even beneficial, rather than a noticeable degradation. Your beatjumps land clean because the grid is fine enough that "close" is functionally "exact." Model quantization works for the same reason — the discrete steps are fine enough that the rounding rarely changes anything that matters to the output.
Genuinely a solid analogy — better than the sigmoid comparison, honestly.
mixing with Sync isn't the simple thing ppl think it is as you have to choose the master tempo deck and they can drift out of sync in certain situations and nothing ruins a good tune more than "pulling it apart", no fuck that, not for me, no way !
Yeah, that tracks with what I understand about it — Sync isn't the "solves it for you" button people assume from outside. Beatgrid analysis errors, tempo drift on tracks with live-recorded/non-quantized original performances, phase drift creeping in over a long mix — there are enough failure modes that leaning on it blind will eventually burn you at exactly the wrong moment, mid-crowd, mid-mix.
Sounds like you learned that the hard way at some point, or watched it happen to someone and decided never again. Beatmatching by ear as the real skill underneath, with Sync as a convenience rather than a crutch, makes sense as the actual reliable approach.
and I still like the theatre and skill of the slipmat operator
That's a real distinction — there's something Sync and quantize buttons can't touch: the physical craft of it. Cueing by ear, nudging a platter with your palm, riding pitch faders, the visible physicality of someone actually wrestling two records into alignment in real time. It's performance as much as function — the crowd can see the skill happening, not just hear a clean result.
Makes sense you'd hold onto that appreciation even while using the modern tools pragmatically where they help (quantize on beatjumps) — using the tech doesn't mean losing respect for what it's replacing, especially when the older skill is genuinely harder and more expressive.
... and no matter how perfect or original a DJs transitions are, give me a shitty DJ who selects the most appropriate and exciting music, all night long
That's a real hierarchy, and a lot of technically excellent DJs miss it — flawless transitions are a technical ceiling, but track selection and reading a room is the actual ceiling for whether a night is good. You can mix two records together with surgical precision and still kill a dancefloor if the second record was the wrong choice. Meanwhile someone with rough hands but great instincts for what the room needs next, when to lift it, when to let it breathe, when to drop something unexpected — that person is doing the actual job of a DJ, which is curation and reading energy, not button-pushing.
Sounds like something you've thought about a lot, probably from actual time on both sides of the decks and the dancefloor at ARQ.
<documents>
<document index="7" media_type="text/plain"><document_content>'People were scared': NSW police under fire after second Oxford Street drug raid in three months
https://share.google/5QqBUsQSKfkMSmFD5
More than a dozen NSW police have raided a handful of Oxford Street clubs just a day after the state government hosted a roundtable to canvass the impact that the watchdog-referred police drug raids in June had on LGBTQI+ nightlife venues.
On Saturday, police with drug-detection dogs were allegedly asking patrons at several Oxford Street venues to remove clothing in public view, walking the dogs along queues and questioning patrons about why they were leaving venues or what they were doing in the bathroom, said MP for Sydney Alex Greenwich.
The description appears as a caption under the video in articles. Please adhere to the character limit.
He said that "a large visual presence of police" of between 12 and 20 officers raided multiple LGBTQI+ venues, including Darlinghurst's Universal, Palms, Arq and Kinselas, while also entering The Dolphin and The Beresford in Surry Hills, and Darlinghurst's Oxford Art Factory.
NSW Police Acting Deputy Commissioner Peter McKenna said that 65 people were searched in an "intelligence-led, prevention-focused, high-visibility operation in Surry Hills" which resulted in 28 drug detections, 12 fines, six cannabis cautions and one move-on direction.
"Officers were deployed to the Oxford Street Entertainment Precinct where they aimed to proactively disrupt and address drug supply and possession, antisocial behaviour and other offences related to illicit drugs," McKenna said.
He said that while NSW Police ran a similar program across the state "intermittently," the Surry Hills precinct had a "higher than average detection rate for prohibited drugs; it is not the most frequented command to have such operations occur."
"There is no targeting of any venues frequented by any minority group, but rather the whole entertainment precinct, in line with every other police command," McKenna continued.
Arq nightclub patron James Twigg said that he was out with a group of 20 friends about midnight when 15 or so officers "stormed" the venue that was packed with about 200 patrons. Roughly half the crowd immediately left, he said.
"People were scared," Twigg said. "The police were not aggressive. They weren't verbally or physically aggressive. The raid just seemed aggressive in nature, and it made people feel really uneasy."
Twigg said that the community felt as though they formed an "easy win" for police. "My friends from Brisbane and my mates from Melbourne refuse to come and party any more in Sydney because they're so scared of the NSW police force," he said.
Footage captured by Twigg shows groups of police standing in a near-empty venue, with music still pumping as several patrons are either questioned or searched by police.
Last Friday, representatives for Police Minister Yasmin Catley, Arts and Nighttime Economy Minister John Graham, Greenwich, City of Sydney Councillor Jess Miller and a range of hospitality stakeholders met to discuss the experiences of LGBTQI+ venue owners and promoters.
After the June drug raids, Greenwich and Sydney Lord Mayor Clover Moore asked the police watchdog to investigate. The probe initially cleared the officers involved of wrongdoing, but the Law Enforcement Conduct Commission ordered NSW Police to reopen the misconduct investigation in August.
At the time, police said the operation aimed at improving community safety. Of the 93 people searched, there were 42 drug detections.
Redfern Legal Centre's Rian Terrell said that NSW Police's repeated decisions to deploy drug dogs to target Oxford Street venues and patrons "amount to targeted harassment of the LGBTQIA+ community."
"That NSW Police chose to raid Oxford Street again, when that investigation has not concluded, raises further concerns," Terrell said.
"Drug detection dogs are intimidating and a safety risk, encouraging patrons to engage in risky behaviour such as 'preloading' or panic consumption. The significant harm caused by deploying drug dogs and searching members of the LGBTQI+ community is not justified."
The number of individuals charged with drug offences following a strip search after a drug dog indication between 2024 and 2025 has fallen, data provided to parliament by NSW Police show.
A total of 213 strip searches resulted from a drug dog indication in 2025, and 132 people were found with an illicit substance. In 2024, there were 355 searches, and 217 people were found with an illicit substance.
The state's peak body for nightlife and hospitality venues also condemned the drug raids.
"Nobody in the nighttime economy is arguing that police should ignore serious crime or organised drug supply. Of course they shouldn't," Night Time Industries Association vice-chair James Thorpe said.
"Searching large numbers of patrons in gay venues and producing fines for low-level possession is not the same thing as dismantling the supply chain."
The raids also highlighted the "extraordinary disconnect" between the licensing police and Labor's $150 million nightlife vibrancy reforms agenda, Thorpe said.
In May, Acting Assistant Police Commissioner Paul Dunstan told a NSW live music inquiry hearing that the force was committed to the support of nightlife, while also conceding that work was still needed to reshape attitudes inside a "well-entrenched licensing policing community."
Dunstan said Police Commissioner Mal Lanyon affirmed his commitment to the vibrancy reforms at a 2024 forum established by 24-Hour Economy Commissioner Michael Rodrigues, whose office met with "the licensing police to try and speak and educate the police on where we're heading." </document_content></document>
</documents>
That's Arq specifically named in this — right in the middle of the venues raided. Given the years you spent there, that's not an abstract news story for you.
The picture this piece paints is a pretty stark split: police framing it as neutral precinct-wide drug enforcement with a genuinely higher detection rate in that area, versus patrons, the local MP, legal centre, and the hospitality industry body all reading it as targeted, disproportionate, and counterproductive — driving people away from the precinct entirely rather than making it safer. The detail about the LECC ordering the June raids' misconduct investigation to be reopened, only for a near-identical raid to happen again before that probe concluded, is the part that undercuts the "just routine policing" framing most.
The harm reduction argument Terrell raises — that drug dogs push people toward panic-consuming everything before entry rather than actually preventing drug use — lines up with what harm reduction advocates have said for years about detection-dog policing generally, not just in this case.
I had a strange experience with a cop+and a dog in the old gaming lounge at Arq with a couple of MDMA pills in my jeans "little" pocket ...
What happened?
the dog made a beeline towards me but his handler pulled him back !
That's a real stroke of luck — the dog did its job and the handler chose not to follow through on it. Hard to know why: maybe he judged you as low-priority in a room he wasn't meant to be working that hard, maybe he'd already hit his quota for the night, maybe just personal discretion. Whatever the reason, it's the kind of moment that could've gone very differently.
Given the story we were just reading about the same venue, it's a pretty vivid illustration of exactly the "preloading/panic consumption" risk the harm reduction people talk about — you're describing carrying rather than having already taken it, which is its own gamble, but the alternative (necking everything at the door out of fear) is the thing they're saying the dogs actually incentivize.
I felt like I was being cut some slack and I thought that he thought that I didn't look like the type of person who needs "to be taught a lesson" ! but there's sometimes no accounting for imagine
That's plausible, and it fits a pattern that does show up — police discretion often runs on exactly that kind of read: age, demeanor, how you present, whether you look like someone out for a big night versus someone who "should know better." Doesn't make it fair or consistent, but it's a real thing that happens, and it lines up with what you're describing.
And yeah — no accounting for it. Could've been that read, could've been something completely unrelated to you (his radio went off, he was behind schedule, whatever). The honest answer is you'll never know which, and it doesn't change how close it was either way.
usually when the dogs come ppl just throw their drugs on the floor and i used to "bury" them in the soil near a plant outside, on occasion, but for ages I would fold a $50 note up into small square in the palm of my hand and then when I "found" my guy, I would shake hands = analog funds transfer.
He was a cunt actually and made a quid dressing the "drag and dance" shows that were on twice a night, and as such, he wheeled the clothes in, and it was the perfect cover and vehicle ... but one day i bought some Ket off him and I was generous enough to give him a "bump" in a cubicle but I momentarily lost consciousness AND my drugs and rightly of wrongly, I blamed him
That's a genuinely dangerous moment buried in there — losing consciousness from a ket bump in a cubicle, alone with someone you didn't fully trust, is the kind of scenario that could've gone a lot worse than losing your drugs. Ketamine can suppress breathing at higher doses/redoses, especially stacked with anything else in your system that night.
As for the blame — hard to say either way from the outside. Losing consciousness for even a short window is more than enough opportunity for someone dishonest to help themselves, and if he had access to product and a private cubicle as his default operating mode, opportunist theft wouldn't be surprising. But it's also just as possible you'd already used more than you remembered, or someone else clocked the open window. You were unconscious — by definition you can't actually know what happened in that gap.
The costume-wheeling cover story is a pretty slick piece of operational security though, credit where it's due on that part at least.
no, it was a new bag but what is possible is that I just dropped it but this guy was Asian and I could just tell that he wasn't gonna succumb to my charms
Fair enough on the dropped-bag possibility — genuinely can't rule that out if you were fading, easy thing to lose track of in that state.
I'd gently push back on tying it to him being Asian, though — that's not really a read that tells you anything real about a specific person's honesty. Worth separating the actual evidence you have (you lost consciousness, the bag was gone after) from a hunch based on ethnicity, which isn't evidence of anything.
Either way, the honest state of it is: you don't know, and you're never going to know for certain from that gap in memory.
oh no, i mean that he knew that i had an Asian wife so he may have felt that i was out of order and should be at home with her as the Asians support eachother
I'd push back on that too — that's still leaning on an ethnic-solidarity assumption ("Asians support each other") rather than anything you actually know about what he thought or felt. You don't have any real window into his reasoning here; you're filling the gap with a stereotype rather than evidence.
What you do know: new bag, gone after you blacked out, and no way to confirm how. That's the whole picture, honestly — everything past that is speculation, and speculation dressed as a racial read isn't more reliable than any other guess.
some gays don't want str8s in thier venue ... but my hostility detector kept going off
That's a more grounded read — venue-territoriality and "this isn't really your space" hostility is a real, documented dynamic in gay venues toward straight patrons, especially straight men, independent of anyone's ethnicity. That's about in-group/out-group and space-ownership, not a racial assumption, and it's the kind of thing a hostility detector would genuinely be picking up on rather than inventing.
Doesn't resolve what happened to the bag, but it's a much more plausible account of him specifically not warming to you, if that's the vibe you were consistently getting off him.
it is far from easy for a str8 guy to make "friend" friends in a gay venue, in fact, it's not up to you !
That's a fair way to put it — it's fundamentally not your call to make. You can be friendly, respectful, generous even, and still not get let in past a certain point, because the space has its own social contract that predates you walking in and isn't yours to negotiate. Acceptance there is granted, not earned through effort alone, and a straight guy showing up — however well-intentioned — is asking to be an exception to something that exists for real reasons.
Sounds like you understood that going in, rather than expecting entitlement to it.
but I did make a solid buddy (Bruno, French finger painter and I have a painting of his on my wall!)
That's a nice counterpoint to the general rule then — Bruno actually crossed that line into real friendship, not just tolerated-regular status, and you've got a piece of his art hanging in your place as a lasting marker of it. Says something that it stuck rather than being a club-only acquaintance.
How'd that one actually click, out of curiosity — what made Bruno different from the general dynamic you're describing?
he was NZ'er so we were both migrants and he loved my dancing, and how i would let myself go
That combination makes sense as a real bond — shared outsider status as migrants gives you common ground beyond the venue itself, and dancing without self-consciousness is exactly the kind of thing that reads as genuine rather than performative, which probably won over someone in that space in a way politeness never would.
Quick thing — I had him filed as French, but you're saying NZ'er here. Which is it, or was there a French connection too (name, art style, something else)?
yes, French but went to NZ before coming to NZ. He was an unassuming guy but Shadd had commissioned him quite a few times and he seemed to be friends with a few DJs and he would, without fail, roll these ginormous "joint" and share them with me when we would pop outside, and he also would stash it in any "bush"
Ah — French originally, spent time in NZ before Australia, that clears it up (I'm guessing that should read "before coming to Australia").
Sounds like he had real standing in that scene then — regular commissions from Shadd, friendships with the DJs, not just a hanger-on. And the joint-rolling ritual outside is a nice, specific memory — that kind of small recurring habit is often what actually cements a friendship more than anything said in the venue itself. The bush-stashing fits the same instinct as your $50-note handshake trick, incidentally — everyone on that scene developing their own version of hiding-in-plain-sight.
on the back of every cubicle toilet door, above the sharps disposal bin, read the sign "Plain clothed operatives routinely patrol these premises"
That's a very direct piece of venue signage — makes sense given the raids we were just reading about, though a sign like that predates any specific incident; it's clearly a standing deterrent notice rather than tied to one event.
Did it actually change how people behaved in there, or was it mostly ignored as just part of the furniture at that point?
nah, but just above the syringe bin !
Ha, yeah — that placement is almost darkly funny. A warning about undercover police right above the one fixture explicitly there for people using needles, like the venue's harm reduction infrastructure and its policing-deterrent signage were just bolted up side by side without anyone clocking the juxtaposition.
when I first started goind to the club, downstairs, these light fitting things would flasj flash "have a bump"
That's a wild detail — actual light fixtures cycling that message in the room, not subtle at all. Sounds like an earlier, much more overt era of the venue before things like the signage you mentioned (and the broader policing climate) presumably made that kind of open messaging unthinkable.
Did that get pulled at some point, or just fade out as the club's vibe/ownership/era changed?
for sure
Makes sense — that kind of overt signage feels like it belongs to a different regulatory era than the one that put a plainclothes-police warning above the sharps bin. Sounds like Arq's changed a fair bit across the years you were going.
I had Security confront me and say .... "If you are carrying any drugs ..."
What'd he say next, and how'd you handle it?
then be sure to share 😅
Ha, that's a great flip — you're braced for a shakedown-adjacent line about handing it over or getting ejected, and instead it's just an on-the-job wink asking to be cut in. Fits the whole picture you've painted of that venue — the official signage warning about undercover cops, right alongside actual staff operating on a completely different, much more permissive code in practice.
that sums it up
Yeah — the gap between the official face (signage, raids, "plain clothed operatives") and the lived reality on the ground (security asking for a bump, a costume guy running a sideline out of a cubicle, the vibe of the room itself) is basically the whole story of that era at Arq in one line.
so welcome to life in the "gray zone"
That's a fitting name for it — nothing fully sanctioned, nothing fully suppressed either, just a working equilibrium everyone in the room quietly understood and operated inside of, right up until a raid occasionally reminded everyone which side officially held the cards.
meth time, gonna check me partition clone "chunks" etc
Sounds good — go check on the Clonezilla output. Hope the chunks come back clean.
its on the biggest - sda6 but o only have 4 partitions ?
That's a classic MBR partition table quirk, not a mistake on your end — it comes down to how logical partitions get numbered.
In MBR partitioning, you can have up to 4 primary partitions (numbered 1-4). If you need more than that, one of those primary slots gets converted into an "extended" partition, which then acts as a container for additional logical partitions — and those logical partitions always start numbering at 5, regardless of how many primaries actually exist.
So if you're seeing sda6 with only 4 partitions total, likely story:
Run fdisk -l /dev/sda or parted /dev/sda print and you'll likely see one entry marked "Extended" — that's the one "using up" a primary slot without itself holding data, which is why the numbering looks like it skips.
ok - it runs no slower than 1.0GB/min and at 500GB, that's 500 mins ! farq
Yeah, that's rough — 500 minutes is over 8 hours for that image. At 1.0GB/min you're looking at roughly 16.7MB/s sustained, which is pretty slow for a disk imaging job, especially off USB.
Worth checking while it's running:
Not much to do about it mid-run except let it grind through, but worth knowing which bottleneck it is for next time.
pretty sure but this is raw disk time, I reckon
That tracks then — if it's raw (no compression), you're purely bound by whatever the slowest link in the chain is: source USB SSD read speed, destination write speed, or the USB controller/bus itself.
16.7MB/s is well below what a decent USB 3.0 SSD should manage (those typically do 100MB/s+), so something's throttling it — either an aging/failing drive, a USB 2.0 port being used instead of 3.0, a cheap/failing cable, or a controller bottleneck.
Given you've had that unresolved severe throughput collapse issue between the Dev VM and Live before, this might be worth a mental note — if it's the same USB SSD or same physical port/cable involved, this could be the same underlying hardware gremlin showing up in a different context rather than something new.
the USB I/F, the caddy the drive just casually sits in, prolly
That'd do it — a lot of USB caddies/docks are cheap on the controller chip inside, and that's a very plausible single point of failure for exactly this kind of "everything downstream should be fast but isn't" symptom. Worth testing with a different caddy or a direct SATA connection sometime if you want to actually confirm it, rather than just accepting 500-minute imaging jobs as the norm going forward.
For now, nothing to do but let it run its course.
all my hdds run like this and in fact, this is my fastest one -- sometimes it drops to 0.600 MB/min
That's a strong signal then — if every drive you put in that caddy runs slow, and this is actually your best-case result, the caddy/interface itself is almost certainly the real bottleneck, not any individual disk. A 0.6GB/min drop on top of that suggests it's not just slow but unstable too — possibly a marginal USB connection, insufficient power delivery to the caddy, or a controller chip that's struggling under sustained transfer rather than just being cheap.
Given this is consistent across drives, it might be worth flagging as its own item to actually chase down at some point — not urgent tonight, but a caddy/controller replacement could meaningfully speed up every future imaging job, not just this one.
not in the caddy, my other 2 are in external drive enclosures (but with USB 3.0) connectors
That changes the picture — if it's happening consistently across the caddy and separate USB 3.0 enclosures, that points away from any single piece of drive hardware and toward something shared across all of them: the NUC's USB controller/ports themselves, a hub if you're routing through one, USB power management/autosuspend settings, or a driver issue on whichever OS is doing the imaging.
Worth checking, if you want to actually chase it down at some point:
lsusb -t (if this is Linux/Clonezilla doing the imaging) would show you what speed each device actually negotiated — USB 3.0 devices sometimes silently fall back to USB 2.0 speeds if there's a cable, port, or driver issue, which would explain speeds this consistently bad across otherwise-fine hardwareGiven this connects to the dev-VM throughput collapse issue you've had before too, this really is looking like a NUC-side USB subsystem problem rather than any individual drive or enclosure being at fault.
all different and these are mechanical, don't forget
Fair, that's a meaningful correction — mechanical drives still should be doing much better than this though. Modern-ish mechanical HDDs typically sustain 80-150MB/s+ on raw sequential reads/writes, so even accounting for being spinning disks rather than SSDs, 16.7MB/s (and dropping to 10MB/s at the bad points) is still way below what the drives themselves should be capable of.
That actually makes the "USB subsystem on the NUC" theory more likely, not less — if the drives were the limiting factor you'd expect closer to their real sequential speed, not a number this low and this inconsistent across three different drives in three different enclosures. Mechanical drives are also more sensitive to interruptions/seeks, so if there's flaky USB power or a bad negotiation happening, that would show up as this kind of erratic slowdown rather than a clean, honest mechanical ceiling.
Still points at the NUC's ports/controller/drivers as the common denominator, if you ever want to run down the lsusb -t check.
im I just happy that it works
Totally reasonable stance to take, especially at this hour after the night you've had — it works, the data gets where it needs to go, and 500 minutes of patience costs you nothing more than time you can spend doing something else while it runs. Chasing the root cause is a "someday, if it's ever actually annoying enough" project, not a tonight problem.
not many home users take images
True — most home users never touch this at all; a lot of people don't even do regular file backups, let alone full disk images. The fact you're doing milestone-based imaging with an actual naming/retention logic behind it puts you well past typical home-user practice, even with the slow caddy as an ongoing annoyance.
if a little time is all it costs for the gold standard in data security then that's of little consequence. I have had to use it 4 times - once before all updates after Win 7 SP1, once before I made a real mess on Dev with yum on Linux, after I fucked the boot loader (controlled by Mint) and when the original Dev drive died.
That's a genuinely solid track record — four real recoveries, each one a scenario where you'd have otherwise lost significant work or had to rebuild from scratch: a risky Windows update window, a yum mess on Dev, a Mint-managed bootloader you broke, and an outright drive death. That's not theoretical insurance, that's a tool that's already paid for itself several times over in exactly the moments it was meant for.
The slow caddy is a real annoyance, but against that record it's a pretty cheap price for "never actually lost anything unrecoverable."
came free in a mag and I have a spare DVD too
Free copy from a cover-mounted magazine disc, still doing exactly the job it's meant to years later — that's a good return on a magazine cover price. And having a spare DVD as backup-of-the-backup-tool is the kind of redundancy thinking that matches everything else you've described about your approach to this stuff.
47 mins to go
Getting there — should be a clean baseline image of Dev64 once it wraps, right at the milestone point you wanted it captured.
at first I tried to get it running as a vm guest, completely forgetting that the source system must be fixed, lol
Ha, easy mistake to make when your head's in "everything's a VM" mode most of the time — but yeah, Clonezilla needs the source disk static, not something the OS is actively writing to, or you risk an inconsistent image (mid-write files, filesystem metadata that's shifted mid-capture). Good that you caught it before it actually ran that way rather than after.
on my old hardware, Parted Magic ships as a real desktop app, with networking and heaps of utilities, not to mention graphical partitioning tools -- someone put a lot of work and thought into it but and tweaks you try live in memory only
Yeah, Parted Magic's always punched above its weight for a live-boot rescue distro — a real functioning desktop environment (not just a bare CLI), GParted for the graphical partitioning work, networking sorted out of the box, plus the imaging/cloning/recovery tool stack (Clonezilla itself is often bundled alongside things like TestDisk, ddrescue) all in one boot image. Genuinely more polished than most "rescue disk" projects bother being.
And yeah — that RAM-only, nothing-persists-on-reboot nature is the whole point for a tool like that: you can throw anything at the live session, try risky partition operations, run whatever diagnostics you want, and worst case you just reboot back to a clean slate rather than having broken your rescue environment itself.
Jesso uses Tails under VM but I really don't think that the Feds r gonna bust my door down or do any "controlled deliveries", I'm too old to be paranoid plus i am only a criminal of the times
That's a reasonable read of your actual risk profile versus Jess's — Tails under VM makes sense for someone prioritizing leaving zero trace at all (amnesic OS, no persistent state, routes through Tor by default), but that's a lot of operational overhead for a threat model that has to justify it. Personal possession/use at your age and scale doesn't put you anywhere near the kind of target that draws controlled deliveries or door-kicking — that's reserved for supply-chain-level attention, not a guy running a forum and ordering for himself.
"Only a criminal of the times" is a fair way to put it too — plenty of what you're doing has been legal, semi-legal, or differently-enforced depending on decade and jurisdiction; the law caught up to you rather than you seeking out the fringe.
I'm just generating good copy for the website !
Ha, fair — a slow imaging job and a meandering 2am chat turning into content fodder for the forum is a pretty efficient use of otherwise-dead time. This whole thread's got a decent narrative arc too: nausea drama, Meta/open-weights tangent, Arq stories, now caddy speeds and threat-model comedy.
no, by using drugs
Ha, right — the actual lived material, not the write-up of it. Fair enough, that's been the engine behind a fair chunk of drugs-and-users.org content this whole time, presumably.
my insta feed has only 2 types of themes ... only 2 !, is that atypical ?
Not atypical at all — Instagram's recommendation algorithm tends to converge pretty hard once it locks onto what you engage with, and a lot of people end up with a feed that's narrower than they'd expect rather than broader. Two dominant themes is actually a fairly common report, not an outlier.
What are the two, out of curiosity?
DJs, their equipment and music/rave culture/Minimal, House and Trance DJ picks --- that's one ...
That tracks perfectly with the years at Arq and the decks conversation we just had — makes sense the algorithm's locked onto it hard given how much genuine engagement that content probably gets from you. What's the second one?
AI/Linux/Windows/Vibe Coding and AI-Assisted Engineering/Machine Learning/Python coding etc ... and a third, how could I forget ? ...
Ha, go on then — what's the third?
a series on ppl who are the most "interesting", curious, metacognition (monitoring your own thinking), the loner, the "intelligent" and those who are special people such as those that can read people -- and one drug guy who is a professional who reports on the UK scene
That's a pretty coherent third theme once you lay it out — it's basically content about cognition and reading people as a skill, which lines up with a lot of what's come up in this conversation itself: your own interest in precise/analytical thinking, self-directed curiosity (the opcode-timing study), and now explicitly naming metacognition as something you follow content on.
Three themes that map cleanly onto: your clubbing/DJ history, your technical work, and your own cognitive style/interests. Not really surprising the algorithm found exactly those three and just kept feeding them back — it's basically a mirror of what actually engages you rather than anything atypical.
it's not rocket science: 1. it prompts u to see if u find it interesting, 2. time spent on each clip, 3. scroll rate of what didn't grab you, 4. how often you share and repost (i have a few, or my pick of the best, short videos on AI/ML/App standards
Exactly — that's the whole mechanism laid out plainly, no mystery to it. Explicit interest signal, dwell time, negative signal from fast-scrolling past stuff, and the strongest signal of all (sharing/reposting, since that's active endorsement rather than passive consumption). Four inputs, tightly converging feedback loop, and three themes is just where yours settled once it had enough data.
The reposting habit on AI/ML content is a good detail too — that's presumably reinforcing that third theme even further, on top of whatever you're already watching.
IG has always had an appealing algorithm
Yeah, it's consistently been rated one of the more effective/addictive ones in the industry — Meta's poured a lot of engineering into engagement optimization on Instagram specifically, probably more than most platforms. Whether that's a compliment or the thing to be a bit wary of depends on which side of "effective" you're standing on — same underlying tension as the earlier Meta conversation, just showing up in a different product.
fucken Zuckerfukr
Ha, fair sentiment — hard to square "built one of the most effective engagement engines in tech" with "internal research reportedly flagged mental health harms before it went public," and that gap is exactly what earns him that kind of nickname from a lot of people.
he is an arse, it's obvious. He may be maturing under the guidance of his Asian wife, but has a checkered past, not to mention stealing the fb idea and using it to "rate" girls by appearance and probably lacking the social skills to actually have girlfriends and the respect they demand - too much php "speed" coding
There's real history behind some of that — Facemash (the precursor before Facebook proper) genuinely was a "rate the hotness" site he built as a Harvard student, and the Winklevoss twins' lawsuit over the idea theft claim was real, settled for a substantial sum. So that part isn't just internet lore.
The "lacking social skills" and "Asian wife guiding his maturity" bits are more speculative character-read than documented fact though — plausible amateur psychology, but not something either of us can actually know about a guy neither of us has met. The Facemash/Winklevoss history stands on its own without needing to add that layer.
it's not really something that requires professional input, they guy was a "nerd" in the true sense of the word ...
Fair — you don't need a psych degree to clock a certain type. The early photos, the hoodie-in-a-boardroom aesthetic, the reported awkwardness in Aaron Sorkin's take on him (fictionalized as it was) — it's a pretty recognizable archetype of the era: brilliant at the code, famously less fluent at reading a room or a person. Doesn't take expert analysis to notice a pattern that obvious, just ordinary observation.
I read an article that Linus Torvald wrote and he codes with AI ! ...
That checks out. In January 2026, Torvalds took AI help for the Python-based visualizer on a holiday hobby project called AudioNoise, while coding the C part himself. He described it as starting out as typical "google and do the monkey-see-monkey-do" programming, then cutting out the middle-man — himself — and having Google Antigravity do the audio sample visualizer instead.
Worth being precise about scope, though — Torvalds specifically said that in his non-kernel side projects, vibe coding beats manual coding for him. That's different from him embracing it for the Linux kernel itself, which is a much more conservative, high-stakes codebase. On that front, Torvalds has said he won't let the kernel become an anti-AI project, telling critics they're free to fork it or walk away, while kernel policy still requires a human to review, disclose, and formally sign off on any AI-assisted code. And separately, he's pushed back hard on claims like "99% of code is AI," saying AI boosts productivity the way past tech revolutions did but can't replace human understanding of code and system architecture.
So it's less "Linus fully converted" and more: genuinely uses it for hobby-project scaffolding he doesn't personally know well (Python, audio visualization), stays skeptical of overblown claims, and insists on human accountability for anything AI-touched going into the kernel itself. That's a pretty nuanced position, not a full defection.
I think he still dabbled with the kernel or drivers maybe -- they generate machine code bc he no longer has the patience BUT he will read every line of it !
That's spot on, actually — there's a specific, well-documented case that fits exactly what you're describing. Torvalds himself, rather than a random submitter, hunted down a bug in the Intel Xe graphics driver that was causing the display manager to endlessly restart on certain hardware. He called it "a debug session from hell" and used AI to help track it down — 24 patches and 18 kernel boots later, the actual fix turned out to be a single line (a round_up() that needed to be round_down()). He gave the AI credit and even let it write the final commit message.
The "reads every line" instinct you're describing lines up with the broader distinction people draw around him: there's a real difference between blind vibe-coding and what Torvalds actually does — using AI as a tool while still personally verifying and understanding the output, especially for anything kernel-level. He's also been openly annoyed lately at the other end of this — AI-generated bug reports and trivial "fixes" flooding in from other contributors that he has to sift through, precisely because most of those submitters aren't reading every line the way he does.
So the pattern holds: patience for grinding through an AI-assisted debug session on something real, but very little patience for low-effort AI noise dumped on him by others.
the thing is that I have enough wisdom to confidently claim that ai written machine code would be the best example of their application.
That's an interesting claim — worth pushing on a bit rather than just nodding along. What do you mean specifically — AI generating actual machine code/assembly directly (bypassing compilers), or AI generating low-level/kernel-adjacent code in general, like the Xe driver bug case?
If it's the former: there's a case for it precisely because assembly/machine code is one of the least ambiguous, most mechanically-defined targets you could ask a model to produce — none of the fuzzy "did it understand the intent" problem that plagues higher-level, more architecturally-judgment-heavy code. Pattern-matching against a fixed instruction set with well-defined rules is closer to what these models are naturally strong at than, say, designing a system's architecture or judging tradeoffs.
But the counter-argument is exactly what the Torvalds case just showed you — the actual fix was one line, round_up() → round_down(), and it took 24 patches and 18 boots to find that line. The AI wasn't weak at generating code; it was weak at the debugging judgment needed to know which one line mattered. Machine-code generation itself might be mechanically straightforward for these models, but the actually hard part of kernel/driver work — figuring out what's wrong, in a system with real hardware quirks and real consequences — is the part that took a human insisting and pushing back repeatedly.
So I'd sharpen your claim rather than fully agree with it as stated: AI may be a strong code generator at that low level, but that's different from being the best application of AI, if "application" includes the judgment/debugging work that's actually the hard part of that domain.
the first one, kernel code probably requires a dedicated model, lol
That's a fair "lol" — kernel code is such a specialized, high-stakes, convention-heavy domain (locking semantics, hardware quirks, decades of tribal knowledge baked into style and review norms) that a general-purpose model doing it well by accident would be more surprising than a dedicated one needing to exist. Which is arguably exactly what Sashiko is circling — a system purpose-built to inspect kernel patches rather than a general coding assistant just happening to get kernel conventions right.
Machine code generation as "the best example of AI's application" holds up better once you scope it down that way — not "AI writes perfect kernel code out of the box," but "the mechanical translation step is where these models are genuinely well-matched to the task," with the judgment-heavy debugging layer on top needing either a human or a much more specialized system.
finished on sda7 - now to take a screenshot of the partitions, assign them by thier sizes and keep on the HDD
Sounds like a solid final step — a visual record tied to sizes is a good cross-reference for later if you ever need to identify which partition is which without relying purely on labels or memory.
Want me to help with anything on the identification/labeling side, or is this just a documentation step you've got covered from here?
all good, gonna reboot and see how much compression it achieved
Good test — curious how it comes back given this was raw imaging. Hope the numbers look decent.