gattsuru
5hr ago
Interestingly, it also prevents under the same 'hard no' serious analysis on real-world assaults: I ran the Loudon County case and follow-up through Opus 4.8, and it's pretty overt how quickly it pulled away from discussions that got close to the assaults or alleged improprieties, and trying to get its evaluation of how bad the teacher's assistant's willful blindness likely was explicitly calls it "hard boundary".
Which makes sense, in a lot of ways! Many of them unpleasantly obvious. But it also makes it stupid.
gattsuru
8hr ago
I wrote this story in 2022. I did not write it about LLMs. It's a problem that dates back to the invention of calculators, or arguably to the development of writing.
I think it's more relevant today than four years ago.
We have solutions, of a kind. Programmers pick up (and build) esolangs, hardware hackers hunt for antiques that may never have been documented, sysadmins can learn to install archlinux without archinstall, woodworkers try random crap with random scrap, mechanics do custom work, so on. Math Olympiads and spelling bees can bypass the development of calculators and always-on spell-check. But these are solutions to the old versions of the problem.
In the real FOOM case, we don't care. It's the Minds' universe, or CelestAI, and either we live with it or we don't and it doesn't really matter.
If the extant development is bounded, or even unbounded in many ability but constrained in others, it gets significant more bad.
gattsuru
11hr ago
In addition to looking like a roving example of Everything Leftism (election voting to gender politics to military spending to riot policing), there's a perception of IT people as largely devolving into three separate stereotypes:
- Stuffed suits, who have a very good compliance posture, but will recommend software that EOL'd five years ago and an API for that software from the Clinton administration.
- Polos, who will have memorized minutia of esoteric stuff specific to their environment and can run wires through a spiral labyrinth, but get rapidly out of their depth in Big Questions.
- T-shirts/hoodies, who have five years of experience in three-year-old new languages, but think of compliance as 'it's only illegal if you get caught, and prosecuted, and the penalty is big enough to care'.
This is a set of stereotypes -- there's a small industry of t-shirts who've put on a suit and play suit games in particular -- but it's not an unfounded one. But Moussouris is a t-shirt.
That wouldn't necessarily be unpersuasive because of those things at smaller scales (though I've personally been burned over it before), but Anthropic is big enough that they could have gotten any expert from any of those positions or from a Red Tribe perspective was a choice, and not doing so indicates that there wasn't a credible expert willing to do what they wanted from outside of those positions.
Add in the ideological position about defense-vs-offense, and it's very much like if Cathode_G were brought to testify in front of the ATF. He could well be right! And it wouldn't matter.
gattsuru
13hr ago
I assume the files in question were just normal guns for manufacture? Nothing super high-tech, needing any sort of classification or export restriction?
The rules apply from the it's-kinda-dumb Liberator proof of concept to the modern-gen FGC-9, but they're nothing superscience or fantastical.
gattsuru
15hr ago
Eh... that's not really how it ends up if you push the thing hard. Conceptually, you'd expect it to reduce to averages or centrality, but in practice even minor weird decisions early in a prompt can drive an LLM to extremes pretty quickly.
This prompt isn't at all original or complicated, but this story (cw:30k words AI slop) isn't central-mode. It has a lot of other problems! The prose has one setting and it's 11, the foreshadowing is less subtle than a hammer, and the narrative sets up some really interest ideas about collaboration over dominance in an elemental magic system and then segues into a marvel movie at the end. Some of those are downstream of the prompts, since I was attempting to match phailyoor's test and do minimal editorial direction, but a lot of them are more general.
And the other side of things is that as it gets cheaper to produce something, visibility of that content is going to need something more than production. So long as something on the outside edges is available, it's going to collect attention just by nature of being differentiable, whether or not it's even better.
I don't think that's going to go quite the same as sort-by-controversial, but it's got a lot of potential for weird.
gattsuru
16hr ago
Defense Distributed v. New Jersey en banc denied.
Depends on how you look at it.
Defense Distributed 'won' a big victory where there was a full settlement announcement that Defense Distributed specifically would get one of the new licenses to distribute those ITAR'd files. Then the government moved the files from ITAR rules to EAR rules, and then successfully argued in court that they no longer needed to issue a permit. In theory, this meant Defense Distributed was free and clear so long as they were publishing information rather than ready-to-print files ... and then the Biden administration argued that an EAR exception for ready-to-manufacture code explicitly covered CAD-only files or instruction manuals for firearms, did so by FAQ, and coincidentally EAR rules are completely barred from judicial review. Defense Distributed has not received a license under the new regime, could not receive one to publish broadly -- that's explicit text in the Department of Commerce letter -- and it's basically in the same state as it was before the lawsuit, but with no recourse.
Strictly speaking, a First Amendment as-applied challenge might get past that reviewability bar. Good fucking luck.
For bonus points, EAR violations are 350k+ per item (presumably per-file) as civil violations, 1 million and years in prison as a criminal violation if willful, and come with the extra punishment of going on the Denied Persons List, aka 'no bank accounts allowed'.
gattsuru
1d ago
I guess that's an insight, but it's a pretty unimpressive one in the modern era. Arbitrary, contradictory, and impossible-to-verify compliance nonrules are such a common thing in gunnie contexts that I've made a running theme out of it, and it's just more obvious in gunnie spaces because there are clear definitional categories. Whether it's California insisting on microstamping tech, or New York trying to ban 3d printers, the requirements don't even need to be possible.
Hell, Ball seems to even be missing the actual alpha in the strategy. The rules don't actually have to include a disliked organization's behavior: the administration can always just try to bring enforcement anyway, lie to the courts, and then shoot someone in the head for breaking the 'rules' so long as they do it before the rules are enjoined. Look at both FCC v Starlink and Illinois v Due Process here, and note that the real punchline is that none of the appeals processes for either cared that the rules were made up and the points don't matter, and even post-hoc no one cares but absolute nuts like myself.
((But yeah, no one received civil contempt penalties or faced legal oversight, if only because they'd have to investigate themselves.))
The real benefit is that it makes it impossible to challenge and drives terror in the people who oppose you.
But again, Defense Distributed. Would be nice if no one burnt that bridge.
gattsuru
1d ago
I would prefer a world where there are deeper motivations than mere politics determining international arms export controls or national security compliance. That kinda flew the coop in the 90s, and by the point Defense Distributed first came out, it was a punchline.
But I'm also still confused about what, exactly, is going the fuck on. "Recent cyber directive" is either someone's ERP or this executive order, and it focuses on 44 U.S.C. 3552(b)(6)(A), which is about restrictions on providers to the US government and specifically disavows permitting or licensing: "Nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models."
ITAR is my first guess, given the 'local employees that aren't citizens can't touch' bit, and it's the most dangerous threat -- not only does it prohibit unlicensed export of specific technologies, it can restrict training, which is a serious and common concern for ITAR-covered imaging technologies and has some hilarious ramifications if you think LLMs can independently rediscover technology. That's a pretty big gun aimed at every LLM company based in the United States.
But it's also just plausible that the government's referencing some more esoteric command, or even just making things up.
gattsuru
1d ago
Eh... SpaceX has launched more rockets than the rest of the world in no small part by making the justification to do it, in addition to the ability -- SpaceX without StarLink gets into a really weird place that's at minimum a much smaller company. And now that they can do it, there's a lot of other potential: space-related buildouts have historically been a place clever ideas go to die because availability and reliability didn't exist, and now they're present.
A lot of the ideas won't be economically viable (ICB-concorde) or won't work for decades (asteroid mining). But it's a little strange that there's no more potential. I admit I don't know the expectation rules and timelines that the filing here would need to use, though, which may explain that.
gattsuru
1d ago
AI's a fun excuse, and there have been some code wtfs, but a lot of the Win11 stuff is just an absolutely bizarre set of UI decisions that have been progressively escalating for longer than LLMs could finish a trivial function. The Start Search Menu has been fucked up since before COVID. Same for OneCloud. The inconsistent and awkward right-click menus in File Explorer are a human interface problem; the code's fine.
Something just went badly rotten in Microsoft.
gattsuru
1d ago
There's zero chance of SCOTUS having all the opinions done by the end of this week, and it's very common for SCOTUS to not post its opinions days until the week before or sometimes even a day before the day. I'm expecting this year to see opinions released until well into the first (and possibly second) week of July.
gattsuru
3d ago
I haven’t been able to figure out what authority the feds are using here, but if it’s ITAR it gets rough quick.
gattsuru
4d ago
Ken White, this week:
The richest man in the world is doing everything he can to get poor, downtrodden people killed in the UK because they aren't white, and the best chance of that stopping is someone killing him. If Only.
In case anyone actually wanted to pretend he believed anything he ever says.
gattsuru
4d ago
Continuing on RP2350 development, fighting the software on a Pico 2 while waiting for a board fab to get my first-run prototypes back (and inevitably show me how badly I fucked up several subsystems and how much I'm going to regret QFN for prototype-gen2).
I used to wonder why software debugging support on edutech microcontrollers was either deep in the SWD/JTAG weeds or absolutely nothing. All of the components exist: micropython supports tracing with a compile flag, DAP means you don't need to build a full IDE (albeit at the cost of writing Yet More Json), there's more than enough computer and memory overheard on normal boards that the processor and memory cost doesn't really matter; dual core and multi-processor systems are so common that you can buy them by accident.
I don't wonder anymore. Holy shit.
gattsuru
5d ago
More that they're cheaper than a code monkey, only weakly expensive if you're doing something hard or novel (or novel-to-you), and they can get ludicrously expensive if you just start firing the slop cannons, either to solve a problem by volume or by producing a lot of useless or specialized lines-of-code.
[apologies if I'm answering a rhetorical question]
A 'token' is the ML equivalent of a syllable: the word, portion of a word, or symbol that represents the smallest viable input or output unit of effort. The exact value (and cost) depends on model, as well as whether it's input or output.
I'm ... skeptical to endorse line-of-code as a measure of programmer time -- I've spent days planning out business-critical logic that ended up five lines of code and needed to be absolutely correct, and spat out thousands of lines of text in an hour before when it was just boilerplate -- but the output side you can give a pretty good average. Depends on the tokenizer and your output language, but I'd expect less an average of less than 20 tokens per line in C++ or TypeScript, and I'd get worried if a human coder was regularly writing >80 tokens in a single line.
((edit: less so in java.))
So pessimistically, 12.5k LoC per million tokens, more realistically 25k.
Input is the high-variance part. If you're writing from scratch, the input is a few paragraphs and some design documents, maybe some scribbled image files if you feel spicy and the model supports it. I've done a few personal projects like that where it's been <2k tokens to get 20k line-of-code. If you have an existing codebase you want the model to adjust to, or an API document you need the model to learn, that can burn through a lot of tokens fast; the only real restriction is context window size, and most of the corporate APIs obfuscate that (tbf, often because they have an automated store and search strategy). I've blown through 50k in a single search once (thanks, Atmel, love your manual layout too). Input is typically cheaper and there's some strategies to reduce the cost of repeated input hits with the same content, but they're complicated and pretty specialized.
For some examples:
| Model | Input (USD/million-tokens) | Output (USD/million-tokens) | Output (USD/thousand-line-of-code) |
|---|---|---|---|
| Claude Mythos | $10 | $50 | $2 |
| Claude Opus 4.8 | $4 | $25 | $1 |
| ChatGPT 5.5 | $5 | $30 | $1.2 |
| Grok 4.3 | $1.25 | $2.50 | $0.10 |
| Qwen3.7 Max | $1.25 | $3.75 | $0.15 |
| Qwen3.7 Plus | $0.32 | $1.28 | $0.05 |
| Qwen3.6 35B-A3B | $0.15 | $1.00 | $0.04 |
For smaller or more efficient models, inference is pretty cheap: Qwen3.6 35B's probably the weakest coding model I'd use in a professional environment (and borders the point where it might be better to run it locally, if only for privacy/security reasons), but there's a lot you can do.
That said, all of that can go out of the window when you start getting agentic options involved. Someone made a fun experiment of trying to let a local model figure out a display protocol by hooking a camera, an LLM, and a microcontroller together, and they got it mostly there overnight, which is really cool. It also probably burned tens of millions of tokens on output for an interface code block that should have ended up in the <1.5k line-of-code level.
gattsuru
5d ago
I have a small ML server that I initially set up for some work stuff, and have since retrofitted for LLM and diffuser use. nVidia 3090, i5-14400, running between 128 GB RAM to 192 GB RAM depending on what else I've shut down. Squeaked in just before RAM prices spiked (and am kicking myself for not grabbing three or four more of the 64x2 kits), if you want to know the why on the weird RAM numbers.
I'll caveat that you just shouldn't expect Claude or even Grok-level outputs from local models on their own.
For LLMs runners, I've mostly stuck to llama-server (and forks) as well, after an initial and short-lived love-hate relationship with LMStudio. I have a few custom bits of code for sequencing larger grouped requests, but they're worse-than-vibe-code level stuff and basically just a UI and for loop. Toyed with SillyTavern, just in the hopes of getting better organization, but it's really heavily built for roleplay and I'm not that interested in it. I've looked at and played with some agentic-ish stuff in heavily sandboxed and airgapped environments, but when the best options are nanoclaw, hermes and odysseus, but when the least obnoxious one is powered by pewdiepie, there be dragons here.
Writing:
gemma4-26-A4Bis a great editor, beta reader, and brainstorm sounding board. It's the closest to okay prose from a local model in its class, although beating the obvious AI tells out of it takes some effort and it's seldom very interesting. Also seems to have the best MTP assist (though I had the to build theatomic-turboquantvariant llama variant to get MTP to work when it first came out; don't know if the situation has changed there).[Cydonia](https://huggingface.co/TheDrummer/Cydonia-24B-v4) (24B, mistral-based)and ```Strawberry Limeade (70B, llama-based) are older models that were pretty useful and I'll still pull up on occasional to sanity-check stuff against.
Coding:
Qwen 3.6is hard to beat for simple and fast work, especially things like bringing in an image sketch and converting it into XAML or webdev, or beating some simple file munging into shape. In addition to35B-A3B, I'd also point to the27B-MTPdense variant. It's not as fast as A3B, but the gap's smaller than you'd expect, and in some use cases it comes across as much smarter in my experience. I'd also recommend it any time you have a sketch or powerpoint art-level design you want converted into a GUI representation, and can't use a cloud model -- far from perfect, but easily saves hours of work.GLM(ranging from 4.5-Air at 106B up) can be good for complex work, refactoring, and troubleshooting -- but even at moderate quants, it can be fifteen minutes per turn. Great where I've got a ton that needs to go into the hatch and can work on something different; terrible for anything where a fast OODA loop is important.
Standard vs uncensored/abliterated models is a hard question. Qwen is very refusal-prone, and not just on political topics. While Gemma4 is surprisingly willing to play along for a variety of topics, it still has some hard refusal points, some of which can come up surprisingly rapidly. And not just for weird smut, either. I've had .
For Image Generation, your two power user options are Automatic1111/Forge WebUI and ComfyUI. WebUI is the easier option to get started with, and still has a good level of support for things like img2img, swapping models out, or using various plugins or controlnets. ComfyUI's much more capable and eventually lets you do things like switch between models for different stages of a pipeline, but there's very much a 'who wants to drink from the firehose' moment every time you get started, and managing workflows sucks. On the other hand, if you want to run something like TRELLIS2 or Wan3d, ComfyUI's a lot easier (though not easy!) to set up.
In terms of models:
- SDXL based models like the
IllustriousandNoobAIfamily are good for producing general 'vibe'-ish scenes with one or two actors, so long as you don't need precision, and they're pretty fast. - The current new hotness popular options in the furry fandom are
[Chroma](https://huggingface.co/lodestones/Chroma)(9B, FLUX.1-schnell-based) and[Anima](https://huggingface.co/circlestone-labs/Anima)(2B), which favor natural language over the SD-style "throw a bunch of words at it" approach. Much slower, though.Qwen Image 2512variants also fall here, although their workflows can be a lot more annoying. Qwen Image EditandFlux2-Kleinare the best edit models, especially for keeping consistency in a scene while tweaking it, or moving a character from one setting to another.
2D->3D Models:
TRELLIS(2)gives the nicest-looking outputs for a given input image, and supports(ish) transparency.Wan3Dgives more 'whole' models that require less post-processing to ship to a 3d printer, but tends to be a little fuzzy.
Animation:
WAN2has the most support and has been out the longest.LTX-2.3is much faster and comparable or better quality.SCAIL2just came out, but I haven't even tried to set it up yet. Initial reports look good.
I had a hell of a time getting any non-trivial animation model working in Forge WebUI, and the ComfyUI workflows get nutty pretty fast. If you want to experiment with them and not go leaping into the deep end, WAN2GP gives a lot of workflow options, at the cost of sometimes serious performance costs and a bad tendency to automatically download a model without warning.
gattsuru
5d ago
There's a funny world where, in a couple years, ultra-cheap used AI chips make local game streaming a viable option. The software already exists and does a great job with a fairly arbitrary setup, so if you wanted to have a server on one side of the house supporting an ultra-thin client on the other, it mostly just needs someone selling it as a product.
But the limitations to AI 'GPUs' are more broad than just the form factor or output slots, but down to the architecture, power focus, and supported pathways. The economics don't make a ton of sense even if the lag can be kept to sub-frame periods.
I'm surprised that the cybersex numbers are so low, given the model of infidelity as driven by availability. It's far less expensive and more available than normal dating, even for heterosexual men who care about their cyber-partner's real-life gender. Dunno how much that reflects not wanting it, or not seeing it as cheating.
One respondent described their preferred relationship style as "I'm monogamous, but have no preference if partners are mono or poly", which sounds like polygamy with extra steps, but what do I know.
At the risk of unmasking myself, you'd think so, but in the same way that a certain branch of the populace rounds bisexuals to straight or gay depending on their current relationship status, there's a pretty broad stroke who don't think you're poly unless you're either fucking or seeking to fuck a side piece.
LGBT people (62% of bisexuals and 67% of gay men*) were more likely than heterosexuals (41%) to report at least one unfaithful partner.
Huh, that's surprising. I expected a discrepancy, but I didn't expect it to be this severe. Another point in favor of the infidelity-by-availability model given grindr, though I guess there's probably a soccon somewhere surprised it's not 100% - poly%.
However: of the 14 respondents mentioned above, 12 of them (86%) state that they were in monogamous relationships with their partners at the time the partner was unfaithful.
Wonder if there's enough information to point in direction of causation. I could imagine either a world where people who were cheated on were more likely to become poly (either in the charitable 'let's be extremely explicit about boundaries' or less charitable 'might as well get in on the action'), and another entirely different world where people who were poly were likely to find cheaters even when looking at 'monogamous' partners.
One female respondent answered this question with the response "Don’t like the word 'unfaithful' but I have never not been so in a relationship". Your guess is as good as mine as to what that's supposed to mean.
Presumably didn't like the connotations of unfaithful, but did it anyway? I could maybe see the difference if their culture is like 1960s US where relationships and going-steady were different things, but more likely... not that optimistic.
I'm... skeptical about crediting the duty of candor to the court too heavily. There's a lot of things that courts seem perfectly fine with that normal people would call lies, and a lot more than can be done by implication.
In particular, specifics like a lock able rather than locked cabinet, or Chrystal supposedly using parentheticals in a phone call that BAM quotes (and, while not as prohibitive, referring to themselves as "they"), or the 80k-no-60k-no-30k valuation, are not the sort of things that make me trust the complaint as being deeply honest. Doubly so where these are core to the claim, rather than incidental. The parenthetical in particular is very much core to bringing the RICO claim, bolded, and afaik, not written as an insertion by the transcription, and just not a believable thing for even a pretty dumb criminal to say.
Maybe those are all real and supported with strong evidence, or maybe they're the client's claims rather than the lawyer. But they're not written like it.
And there's some weird errors. Did Brandon start the 300k USD valuation in February of 2026, or in "early 2025"? One of these is a factual claim (90) and one is a legal predicate (172), and they seem to both be talking about the same incident. [EDIT: actually, it's weirder than that. One is tied to "Chrystal and Benjamin", while the other is tied to "Bryan (Mansell) and Chrystal"; I'd assumed the 2025 was just a typo since it's logically inconsistent with BAM's larger timeline, but bringing in different names?]
Similarly, the >$200k USD number seems to have come from previous marketing material from the store, and while some of that's probably puffery and none of it can attribute to either BAM or Josh, it also seems to be matched by third-party estimates of the value of the whole original consignment. Some of the disparity comes because the majority of the value was in the minifigs, not the sets, which the pleading very carefully occludes in the valuations even though it highlights Chrystal's alleged taking (but not theft?) of a Boba Fett figure. (no, I don't know why people pay so much for the minifigs)
It's possible that Chrystal/Gorman sold a large portion (or even Brandon took it back without notice to BAM/Josh) and Brandon's just doing a scam artist routine, although BAM having access to the POS system this long and not knowing or providing evidence one direction or the other raises serious doubts that direction.
That doesn't make Chrystal, Gorman, and Bryan in the right: this very much looks like a 'pox on all their houses' situation.
Which is true if you believe what's in the complaint, i.e. that the prior franchisee sold the goods and absconded with the money, and that neither BAM nor the current franchisees have possession of the goods or knowledge of what actually happened.
Yeah, if BAM's version is true, BAM was either blameless and legally not-at-fault or at worst ethically negligent about management, and Bryan was screwed. Even more so than the BAM pleadings admit -- Chrystal's partner dirtbag-left'd his way out of the country and near-certainly would have most of the money, in a way that would make recovering funds even more impossible than the normal judgement proof version -- so there's even a good explanation why Bryan might have felt like he had no option but to team up with Chrystal in the hope of getting some questionable settlement from an innocent part to get him to go away.
It would be really weird if Chrystal had control of the consignment for nearly a year and didn't sell anything from it.
But honestly, I think it's pretty serious risk even if the current franchisees did have possession of (a large portion of) the goods when they took the store and sold them, either not knowing or not caring about the complicated ownership. It's not the most likely situation, but it's not impossible or improbable one. Still, shops like this aren't going to have a ton of cash-on-hand, they'll liquidate the value of property trying to fight the case, and they've got logical reasons to fight rather than give a token settlement because of the tight financial situation to start with. BAM is a more tempting legal target, but in turn it has a ton of (legitimate!) doubts about actual liability.
If Bryan has a justifiable, good-faith belief that the items are worth $200,000 and that BAM is responsible for their loss, then it's worth filing suit. Hell, attorneys take suits on contingency that are worth a lot less.
There's a bit of a survivorship bias, here. Attorneys also refuse to take suits on contingency that are worth a lot more, either because the defendants are judgement proof, they think the odds are low, or just because it'll take so long that any victory won't be worth the sauce.
gattsuru
7d ago
I will give caution that I've had some mixed results from TP-Link. The worst problems have involved their routers, which you're skipping, but their access points and especially management were pretty flaky the last time I tried them (admittedly, 2023), especially with mesh mode problems or multiple wifi network configurations. Doesn't matter a ton for the switch, but you may want to consider stepping up to UniFi.
((Or if you have concerns about China... uh, gfl.))
Get a bigger switch than you think you need. You're looking at one port for the router, three ports for the access point, one for the dedicated home server... and that just leaves three for end users. That's probably enough right now, if you don't do a lot of high-quality media streaming to televisions or want dedicated lines for a desktop computer, but it's not a lot of excess. Especially if you might want a security camera system or have a lot of PC gamers, those three remaining slots can fill up faster than you'd expect, and WiFi that works fine today might struggle when a couple more devices are all stomping the same SSIDs.
Get a box (with a fan) for the switch if you're putting it in the garage, or keep it in the house. You would be amazed how quickly these electronics pick up dust, mouse crap, sawdust, dead leaves, and insects. I'd favor keeping as much electronic equipment inside the house's vapor barrier as possible.
When you pull the coax, both tape (ideally gaffer, if not duct) and tie a good series of knots, and pull a poly line with the cat6. If you're dealing with conduit, I'd argue just pulling a poly-line run alone first, but for normal residential you shouldn't see conduit for coax.
Consider a small APC for the router, switch, and server. Less because of the uptime -- you're buying maybe 15 minutes for any not-ridiculous APC -- and more for the device protection. Voltage transients in particular are murder on single-power-supply servers, and it's as often the mainboard that dies as the psu.
gattsuru
7d ago
Unfortunately, this is the sort of lawsuit that takes ten thousand bucks for a trivial chance of actually being made whole (and mostly in a settlement), a large chance of getting a useless judgment on a defunct or judgment proof target, and a larger chance of just never having a conclusion that’s mostly a shrug.
Doesn’t make the Reckless Ben response good…
gattsuru
7d ago
BAM claims in court filings that they only allow consignment with express written consent, and while they seem to be overstating the degree it's explicitly prohibited by their franchise agreement, it seems pretty plausible in practice. Consignment is a weird model for a lot of this class of shops, and the threat goes away for sales that are cash-on-hand.
((Conversely, reading between the lines, BAM's management of franchisees seems somewhere between 'poor' and 'actively promoting bad practices', so that could scare people off more than the sale risk. "We couldn't get into the POS system for a franchisee for over a year" is the sort of thing that should have you running for the hills.))
Oh, and Opus 4.8 is weirdly opposed to writing explicit sexual content involving consenting adults, even if it's fine brainstorming it, and even if it recognizes the classifier as a false tag, and sometimes even when it's offered and started to write the smut (cw: lesbians, ctrl+f for "Can you write 3k words, explicit.") I've also generally had problems in scenarios with consenting adults where the consent is weird enough (drunken sex, over-the-top-consenting somnophilia), though those are a little more understandable.
I don't mind as much -- local LLMs or Grok will write these things, and the prose is awful enough that I'm fine with not using them for it either -- but it's a decision. Albeit a decision I'm not sure Anthropic made, specifically.
More options
Context Copy link