sReinwald

You're absolutely right that named credit has tangible career benefits that go well beyond feelings. But I think Vincke threaded that needle well with the anonymous public credit - it creates a documented public record of innovative work at the company level while preserving the engineer's privacy.

The engineer can still leverage this (LinkedIn, internal promotions, industry networking) without being forced into a public-facing role they might not want. When they're interviewing or networking, they can point to Vincke's public acknowledgment and say "that was my project" in contexts where it's professionally relevant, without having their personal social media permanently associated with it.

Considering Vincke was impressed enough to publicly acknowledge this individual's passion and initiative, there's no doubt in my mind that this engineer could get named credit or something that would acknowledge their role in the project if they wanted it.

But to go a bit meta: I think it's strange that we are discussing this in the context of a CEO publicly acknowledging one of their engineers (even if anonymously). Vincke is, at least in the context of the broader industry, going above and beyond. I doubt you'd see Ubisoft, EA, or Blizzard publicly acknowledging a single engineer's after-hours passion project in this way.

Feels a bit like misdirected energy, I guess? Why are we debating about the nuances of named vs anonymous credit and recognition when industry leaders don't give any?

It's like calling someone out for only tipping 10% while ignoring the guy in the top hat who's tipping 0. If you want gaming companies to get better about giving credit and recognition, you should support the companies that are at least moving in the right direction. I know it's easy to be cynical, but don't let perfect be the enemy of good.

I'm not necessarily saying they'd get pilloried. I'm saying that having your personal digital space colonized by people who think you're customer support is insanely disruptive. Think replies full of "I only get 8 fps in Act 3, pls fix" when you just wanted to post a photo of your vacation.

I can't think of specific names anymore since it's been a while since I have played it, but a lot of the developers for World of Warcraft used to be and likely still are active on Twitter. For a lot of them, the community knew fairly well which features of the game or which class they were responsible for. When I used to look at the replies to some of their Tweets (even ones completely unrelated to WoW), they were often full of complaints about their area of perceived responsibility.

I fully understand every engineer who just wants to put their head down and work on their stuff they're passionate about without having to also be public-facing. Even in a small company like mine, some of our devs constantly complain that some customers know that they are responsible for certain features of our product and email them directly rather than going through the proper support channels.

Your point about the games industry often struggling with providing proper credit to devs is well taken - it's absolutely an issue. But in this case, Vincke did actually do that, in a way. He could've just kept quiet and let the playerbase think it was a company effort, but instead he publicly highlighted and recognized the passion and work of one of their engineers (even though anonymously). That engineer can look at the countless positive replies to that post and get the nice fuzzy feeling without getting dragged into the spotlight.

Not that I'm aware of. I thought that was weird at first as well, but I assume it might be in a way to protect the engineer.

Unfortunately, singling out any individual developer, even for praise, can attract unwanted negative attention online. By acknowledging the passion and the work without naming the person, Swen gives them full credit internally while shielding them from becoming a public target.

This doesn't even necessarily have to be intentional harassment, but if this engineer is now the "SteamDeck guy" at Larian, their social media might get flooded by people who mistake their personal social media accounts for a support ticket.

I'm sure the engineer has the option to self-identify if they wish, but this approach feels like a sign of good and thoughtful leadership.

> This is a gigantic effort from Larian, who among all things is still updating its software instead of resting on its own laurels.

What makes this story even better is how it actually came about - this wasn't initially a top-down corporate initiative, but rather a passion project from a single engineer who worked on it after hours. The fact that Larian immediately recognized the value and threw their full support behind it says everything about their culture.

Swen Vincke shared the backstory:

> The story of how this came to be really is one of true passion. The Steam Deck native build was initiated by a single engineer who really wanted a smoother version of the game on Steam Deck and so he started working on it after hours. When we tried it out, we were all surprised by how good it felt and so it didn't take much to convince us to put our shoulders behind it and get it released. It's this type of pure passion for their craft that makes me fall in love with my developers over and over again. Considering myself very lucky to have people like him on my team. Try it out!

https://x.com/LarAtLarian/status/1970526548592623969

That combination of individual passion and company willingness to back good ideas is what makes Larian special.

It's not always, but Israel is uniquely positioned for tech companies of this kind thanks to Unit 8200.

Israel has mandatory military service. They grab the most promising teenagers and give them insanely good cybersecurity training on the government's dime. They do their mandatory service, maybe spend a few more years in the unit, then move on to the private sector. Most, if not all, of Israel's tech companies in the cybersecurity/surveillance sector are related to Unit 8200 for this reason.

But also: I think it's fair to say that Israeli spyware gets more media attention party because of geopolitical factors. Similar to Russian or Chinese spyware. I doubt the same headline would catch as much attention if you swapped Israeli-founded with Spanish-founded.

I doubt you would be happy to hear it, actually. And I doubt it would change your opinion one bit.

You've kicked this entire thread off with an incredibly telling non sequitur: teenager won't use headphones -> multiculturalism bad. When called out on that leap, you pivoted to abstract questions about norm enforcement while ignoring that the norms actually worked fine in this situation.

You're not interested in debate. You're interested in getting someone to validate your predetermined conclusion about the necessity of a cultural hegemony. Having to acknowledge that norm enforcement wasn't actually broken here is pretty inconvenient for that narrative, isn't it?

Still waiting for you to explain that original leap, but we both know you won't. Because you can't without exposing yourself further. Good day. Thanks for playing.

That's a firm 'no' on explaining your original non sequitur then? I told you I'm happy to dive into theories, philosophy, and other theoretical scenarios once you've answered that simple question.

But you're welcome, I guess. At least one of us did.

You're asking about norm enforcement failure, but the story shows norms working fine: expectation communicated, pushback received, de-escalation chosen, situation resolved peacefully, albeit not to the satisfaction of the commenter. The system worked.

The fact that the teenager got defensive is indicative of his understanding of societal expectations and norms. Bending and pushing against norms is what teenagers do, and have done since the dawn of time, regardless of their cultural background and regardless of whether they're navigating the norms of a dominant monoculture or those of a multicultural society.

Your question assumes there was some breakdown that needs fixing, but the only 'problem' was mild inconvenience. When did that become evidence of cultural collapse requiring homogeneity to solve?

You still haven't explained why 'teenager won't use headphones' made you think 'multiculturalism is a problem.' For all you know, this story involved two white Christian Scots from the same cultural background. I'll gladly discuss theories about norm enforcement with you once you've explained why you deemed it necessary to inject race and culture into a story that mentioned neither.

I'm not going to entertain your 'just asking questions' routine until you do.

You randomly bring up multiculturalism as a problem to be solved in response to a story about a rude teenager of unknown cultural background, and assume I'm dogwhistling? Please elaborate because all I'm seeing here is projection.

But I see we've pivoted from "multiculturalism bad" to pseudo-intellectual theorizing about collective cultural conformity. Very smooth. Different packaging, still trying to sell the same product, though.

> The problem is when people escalate quickly because of it and threaten others.

Guy asked teenager to use headphones, teenager got defensive, guy put in earplugs. That's... literally de-escalation and conflict avoidance. Are we reading the same anecdote?

You've managed to escalate "rude teenager won't use headphones" into a lament about the decline of Western civilization and the need for cultural homogeneity. It's almost impressive how much ideological weight you're hanging on one kid's refusal to wear headphones.

> Culture and race are not the same thing.

I don't remember bringing up race. Another interesting leap.

But please do elaborate on what sort of 'dominant culture' you're longing for and what sort of policies you'd love to see to (re-)establish that 'dominant culture' to resolve the incredibly new phenomenon of teenagers being rude and rebelling against social norms.

Come on, don't be a coward and just drop the dogwhistling. You're bad at it.

Interesting leap you made there.

How can you do effective conflict resolution in a society where someone reads an anecdote about a rude teenager and immediately assumes the problem is multiculturalism?

This is a deeply concerning development, though not an entirely surprising one. While I sympathize with itch.io's position - being caught between their creators and their payment processors - the broader implications here are alarming.

Payment processors have effectively become unelected censorship boards with the power to strangle entire categories of legal content by threatening to cut off the economic infrastructure that platforms depend on. The fact that a single advocacy campaign can pressure Visa/Mastercard/PayPal into forcing platforms to remove legal adult content should concern anyone who values free expression online.

The fundamental issue isn't whether you personally approve of adult games or specific content - it's that a handful of payment companies now wield veto power over what legal content can exist in the digital economy. This represents a massive concentration of censorial authority in the hands of unaccountable corporate entities that face no meaningful democratic oversight.

We've seen this pattern repeatedly: PayPal blocking VPN providers over "piracy concerns," Visa suspending payments to adult sites, and now this coordinated pressure campaign. Each time, legal content gets effectively banned not through legislation or courts, but through corporate policy decisions made behind closed doors.

By inserting themselves as moral arbiters for the digital economy and free expression on the internet, these processors are creating a very strong case for being designated as common carriers or being subjected to much stricter public utility regulation. When payment infrastructure becomes as essential as electricity or telephone service for participating in the digital economy, treating these companies as neutral utilities rather than editorial boards becomes not just reasonable but necessary.

I think we're mostly in agreement here. You're absolutely right about the intern analogy - that's exactly my point. The LLM is the intern, and giving either one production database access without proper guardrails is the real failure.

Your point about AI industry overselling is fair and probably contributes to incidents like this. The whole industry has been pretty reckless about setting realistic expectations around what these tools can and can't do safely.

Though I'd argue that a venture capitalist who invests in software startups should have enough domain knowledge to see through the marketing hype and understand that "AI coding assistant" doesn't mean "production-ready autonomous developer."

This is a perfect case study in why AI coding tools aren't replacing professional developers anytime soon - not because of AI limitations, but because of spectacularly poor judgment by people who fundamentally don't understand software development or basic operational security.

The fact that an AI coding assistant could "delete our production database without permission" suggests there were no meaningful guardrails, access controls, or approval workflows in place. That's not an AI problem - that's just staggering negligence and incompetence.

Replit has nothing to apologize for, just like the CEO of Stihl doesn't need to address every instance of an incompetent user cutting their own arm off with one of their chainsaws.

Edit:

> The incident unfolded during a 12-day "vibe coding" experiment by Jason Lemkin, an investor in software startups.

We're in a bubble.

This analysis demonstrates what we call a "Fachidiot" problem in German - deep expertise in one domain coupled with troubling blindness to how that domain intersects with broader realities. The author's "just chill out" recommendation about permanent biological identifiers is about as reassuring as a nuclear physicist telling people not to worry about uranium enrichment because "it's mostly stable isotopes."

The "0.02% of your genome" framing is fundamentally misleading. Those ~640,000 SNPs aren't randomly scattered junk - they're specifically selected markers that correlate strongly with ancestry, health predispositions, pharmacogenomic responses, and familial relationships. The intelligence value isn't in raw percentage coverage but in what can be inferred from those curated data points. And you can infer an awful lot from these targeted markers.

The comparison to browsing history or social media activity is pathetically cavalier. We're talking about immutable biological data that:

    - Links you to family members who never consented to participate  
    - Allows inference about relatives' genetic predispositions based on your data alone    
    - Has unknown future applications as genomic analysis capabilities advance  
    - Cannot be changed, deleted from your actual biology, or "opted out of" once the implications are understood

Understanding genomes doesn't automatically confer understanding of threat modeling, data permanence, or the creative ways malicious actors exploit seemingly "harmless" datasets. The recommendation treats a permanent biological identifier with the same casual attitude as a recoverable password breach.

This is exactly the kind of expert blind spot that leads to catastrophic privacy failures decades down the line.

> I'm so glad that you touched an analogy whose framework I am familiar with.

Oh, I'm delighted too. Because you actually went there and completely let the mask slip. And with such spectacular historical revisionism, you've accidentally proven my entire point about systematic disenfranchisement. Thanks.

Your "white New Yorkers can't oppose Indian reservations" analogy is so ass-backwards it belongs in a museum of colonial apologetics. Palestinians aren't the white New Yorkers here - they're the Native Americans watching settlers build on their ancestral land while being told it's a "mutual agreement." You've literally inverted colonizer and colonized to paint the occupying power as the victim. That's not just wrong; it's a perverse inversion of reality that would make Orwell weep.

But let's dissect your Oslo fiction: Area C comprises 60% of the West Bank, where Palestinians need permits (denied 98% of the time) to build homes, dig wells, or install solar panels on their own land. Meanwhile, Israeli settlements - illegal under international law - expand freely with full state infrastructure. Between 2009-2018, Israel approved 98 out of 4,422 Palestinian permit applications. That's a 2.2% approval rate. For comparison, Harvard's acceptance rate is 3.4%. It's literally easier to get into Harvard than to get permission to build a chicken coop in your own backyard if you're Palestinian. Calling this "mutual agreement" is like calling the Trail of Tears a "voluntary relocation program."

You conveniently omit that Israel controls all borders, airspace, water aquifers, electromagnetic spectrum, population registry, and movement between areas. Palestinians in Area A can't leave without Israeli permission, can't import basic goods without Israeli approval, and can't even collect rainwater without Israeli permits. The average Palestinian gets 73 liters of water per day - below the WHO's 100-liter minimum for basic dignity - while Israeli settlers luxuriate with 300 liters, filling their swimming pools while Palestinian children develop kidney problems from chronic dehydration. That's not autonomy - it's the world's most sophisticated open-air prison. But please, clutch your pearls harder about how you're oppressed because you can't vacation in Ramallah.

Your "20% Palestinian citizens with full rights" talking point? The Nation-State Law explicitly defines Israel as the nation-state of the Jewish people alone - apartheid codified in your Basic Law. The Admissions Committees Law lets 434 communities (43% of all Israeli towns) reject residents for "cultural incompatibility." Palestinian students get $8,400 per year while Jewish students get $12,000. Arab citizens own less than 4% of land despite being 20% of the population. There are ZERO Arab communities among Israel's 135 wealthiest localities. But sure, tell me more about those "equal benefits" while Bedouin villages that predate your state get demolished for the 200th time for lacking permits that are impossible to obtain.

The beautiful part is you've perfectly demonstrated my Memphis point. When I used Palestine as an example of how systematic exclusion from political power creates predictable disparities, you couldn't resist defending apartheid. You literally saw "systematic disenfranchisement" and thought "I should explain why that's actually good, actually."

So thank you, genuinely, for proving that whether it's Black families in South Memphis breathing carcinogens or Palestinian families in South Hebron rationing water, there will always be someone like you - comfortable, privileged, and utterly convinced that the boot on someone else's neck is there for their own good.

> And yes, you are implying that somehow the car I drive is influencing my defence of Musk.

Of course I was. The sarcasm wasn't exactly subtle.

> But alas, neither does that really affect how I view the article or Musk.

The fact that you believe this while simultaneously demonstrating the opposite is genuinely fascinating.

You opened with "I'm no Elon fan" and then revealed you bought one of the first Teslas in your country and are a "huge SpaceX fan." That's like a Yankees season ticket holder insisting their fandom doesn't affect how they judge controversial umpire calls.

Here's what I think happened: You've spent 3.5 years and 110,000 km in that Model 3, feeling like you're part of something transformative - saving the planet, advancing humanity to Mars, whatever narrative helps justify the premium you acknowledge overpaying. When criticism emerges about Musk's companies, it doesn't just challenge a corporation - it threatens the story you tell yourself about your choices.

The overpayment actually worsens this. You can't even tell yourself, "It was just a practical decision." Instead, you've had to construct meaning around that premium - that you're supporting something bigger, something important. The sunk cost isn't just financial; it's emotional and ideological.

So when an article documents xAI operating turbines without permits in already-polluted neighborhoods, you can't engage with those facts directly. Instead, you immediately pivot to Musk's environmental legacy, as if Tesla's global impact creates some cosmic pollution credit karma system where South Memphis residents should accept respiratory disease as acceptable collateral damage for you feeling great about your reduced carbon footprint.

The most telling part? You attacked the article for mentioning two basic facts that appear in literally every single environmental justice story: who owns the company (standard disclosure) and which communities are affected (relevant demographics). You called factual reporting a "hit piece" not because it was inaccurate, but because it made the guy who bought the companies that make the car you drive and the rockets you like to see go 'whoosh' look bad.

You claim the article is biased while demonstrating textbook motivated reasoning. You weren't reading critically - you were reading defensively, scanning for any angle to discredit reporting that challenges your worldview. The "race card" accusation was particularly desperate, as if noting which communities bear pollution burdens is somehow more offensive than the pollution itself.

The real tragedy here is that you could simply say, "Tesla's environmental benefits are real AND xAI should follow permit requirements." Both can be true! But that would require acknowledging that Musk's companies can do wrong, which apparently conflicts too strongly with whatever identity you've constructed around owning a Tesla and being a "huge fan" of SpaceX.

Also, it's pretty interesting that you felt compelled to respond to the little jab about owning a Tesla but chose not to engage with any of my factual criticism. Because that's the tiny part of my comment that threatened the identity you've built up. I'd encourage you to examine that.

You claim decades of carbon advocacy, yet your first instinct was to attack accurate reporting about unpermitted emissions. What exactly is your advocacy worth when you'll throw vulnerable communities under the bus the moment it conflicts with your parasocial relationship with a billionaire (or his companies)?

The saddest part? I genuinely believe you think you're being objective here.

Happy to hear the Model 3 is treating you well, though.

This is such transparently bad faith rhetoric it's almost impressive. You're trying to paint me as racist for pointing out systemic racial inequities. But sure, I'll bite.

You're pretending "race" and "politics" are separate categories, as if centuries of explicitly racial policy - slavery, Jim Crow, redlining, voter suppression - somehow exists outside of politics.

Black Memphis residents were systematically excluded from voting until the 1960s. Redlining prevented Black families from building wealth through homeownership. When your grandparents couldn't vote, buy homes in certain areas, or sit on zoning boards, that directly determines whether your neighborhood gets parks or pollution today.

Sometimes it helps to put concepts into a different context to understand them better, so maybe this analogy helps: When I point out that Palestinians in the West Bank can't effectively oppose settlement expansion because they're systematically excluded from political power, I'm not claiming Palestinians are racially inferior because they're "incapable of manipulating power dynamics" in their favor. I'm pointing out how systematic disenfranchisement creates predictable and unjust outcomes. And even if those barriers vanished tomorrow, Palestinians would still live with the accumulated consequences for generations.

Same principle in Memphis. Noting that unpermitted pollution affects 90% Black neighborhoods isn't claiming racial inferiority - it's documenting the predictable result of decades of deliberate exclusion from political power.

If you genuinely can't grasp how racially motivated systematic political disenfranchisement creates racial disparities, start with basic history.

You've nailed the pattern. And the regulatory environment actively enables it through what amounts to a pricing model for violations.

When the EPA or county eventually fines xAI for running unpermitted turbines for a year, it'll be what - a few hundred thousand? Maybe low millions if they're feeling particularly spicy? For a company chasing the AI gold rush with Musk's billions behind it, that's not a penalty - it's a rounding error. It's cheaper to violate now and pay later than to wait for permits while competitors build capacity.

And unfortunately, this isn't a bug in the regulatory system - it's the feature. When fines are pocket change relative to potential profits, "ask forgiveness not permission" becomes optimal strategy. The only things that actually change behavior are existential threats (criminal charges, shutdown orders) or catastrophic reputational damage - and Musk has proven immune to both.

Until penalties scale with company valuations or include mandatory shutdowns, this playbook will keep printing money. Memphis residents get respiratory disease, xAI gets compute capacity, and regulators get a check that wouldn't cover a week of Musk's jet fuel.

"Disproportionately impacts" means that the burden of pollution isn't randomly distributed - it's systematically concentrated in specific communities. In this case: predominantly Black neighborhoods in South Memphis already have asthma rates and cancer risks 4x the national average (per the article), and xAI added unpermitted turbines to that existing burden.

You're absolutely right that most hazards affect poor communities more. That's not a coincidence - it's the predictable result of power dynamics in zoning and enforcement decisions.

Your comment reads a bit like "water is wet, why mention it?" But most people living in clean-air zip codes have no idea their comfort of living depends on someone else's respiratory disease. They assume industrial siting is purely based on logistics or economics, not on which communities lack the political capital to fight back.

The whole point is that zoning decisions, permit enforcement, and industrial siting aren't random acts of nature. Rich neighborhoods get golf courses and poor neighborhoods get data centers with unpermitted turbines. That's not gravity - it's policy. Documenting these patterns isn't stating the obvious, it's the first step toward accountability. Because "that's just how things are" is exactly what those benefiting from the status quo want everyone to believe.

The article reports that xAI operated gas turbines without required permits and pollution controls for over a year. That's not a "hit piece" - it's documenting regulatory violations with thermal imaging evidence and official permit records.

If this were actually a hit piece on Musk, wouldn't his name be in the headline? Instead, it's mentioned once in paragraph four as standard journalistic context - exactly how articles about AWS mention Bezos. And yes, the affected neighborhoods are predominantly Black - that's a factual demographic statement about who bears the health burden, not "playing the race card." Environmental justice reporting routinely documents how industrial pollution disproportionately impacts minority communities.

Your logic seems to be: "Musk has done good things for the environment globally, therefore local reporting about his company's regulatory violations must be a hit piece." That's a non sequitur. Both can be true - Tesla can advance EV adoption while xAI can violate air quality regulations in Memphis. One doesn't negate the other.

The real tell here is that you're more upset about accurate reporting than about a tech company potentially exposing already-vulnerable communities to additional pollution without proper permits. Your priorities seem to be incredibly misplaced, if you ask me.

How's your Tesla Model 3 doing, by the way? Not that I'd want to imply your choice of transportation has anything to do with your incredibly unfavorable interpretation of this article and defense of Musk. But I have to wonder if your perspective would be different if this facility was in your neighborhood rather than South Memphis, or if you drove a Hyundai.

IMO, the paper commits an omission that undermines the thesis quite a bit: context window limitations are mentioned only once in passing (unless I missed something) and then completely ignored throughout the analysis of SLM suitability for agentic systems.

This is not a minor oversight - it's arguably, in my experience, the most prohibitive technical barrier to this vision. Consider the actual context requirements of modern agentic systems:

    - Claude 4 Sonnet's system prompt alone is reportedly roughly 25k tokens for the behavioral instructions and instructions for tool use
    - A typical coding agent needs: system instructions, tool definitions, current file context, broader context of the project it's working in. Additionally, you might also want to pull in documentation for any frameworks or API specs.
    - You're already at 5-10k tokens of "meta" content before any actual work begins

Most SLM that can run on consumer hardware are capped at 32k or 128k contexts architecturally, but depending on what you consider a "common consumer electronic device" you'll never be able to make use of that window if you want inference at reasonable inference speeds. A 7b or 8b Model like DeepSeek-R1-Distill or Salesforce xLAM-2-8b would take 8GB of VRAM at Q4_K_M Quant with Q8_0 K/V cache at 128k context. IMO, that's not just simple consumer hardware in the sense of the broad computing market, it's enthusiast gaming hardware. Not to mention that performance degrades significantly before hitting those limits.

The "context rot" phenomenon is real: as the ratio of instructional/tool content to actual tasks content increases, models become increasingly confused, hallucinate non-existent tools or forget earlier context. If you have worked with these smaller models, you'll have experienced this firsthand - and big models like o3 or Claude 3.7/4 are not above that either.

Beyond context limitations, the paper's economic efficiency claims simply fall apart under system-level analysis. The authors present simplistic FLOP comparisons while ignoring critical inefficiencies:

    - Retry tax: An LLM completing a complex task with 90% success rate might very well become 3 or 4 attempts at task completion for an SLM, each with full orchestration overhead
    - Task decomposition overhead: Splitting a task that an LLM might be able to complete in one call into five SLM sub-tasks means 5x context setup, inter-task communication costs, and multiplicative error rates
    - Infrastructure efficiency: Modern datacenters achieve PUE ratios near 1.1 with liquid cooling and >90% GPU utilization through batching. Consumer hardware? Gaming GPUS at 5-10% utilization, residential HVAC never designed for sustained compute, and 80-85% power conversion efficiency per device.

When you account for failed attempts, orchestration overhead and infrastructure efficiency, many "economical" SLM deployments likely consume more total energy than centralized LLM inference. It's telling that NVIDIA Research, with deep access to both datacenter and consumer GPU performance data, provides no actual system-level efficiency analysis.

For a paper positioning itself as a comprehensive analysis of SLM viability in agentic systems, sidestepping both context limitations and true system economics while making sweeping efficiency claims feels intellectually dishonest. Though, perhaps I shouldn't be surprised that NVIDIA Research concludes that running language models on both server and consumer hardware represents the optimal path forward.

The core vision here is something I can absolutely get on board with, but the execution fundamentally seems to misunderstand why "home-cooked software" doesn't exist.

The target audience problem is immediately apparent: they're building a product for people who can write JavaScript event handlers but somehow can't 'npx create-react-app'. This demographic is approximately twenty-seven people.

More critically, they've confused the problem space, in my opinion. The barrier to personal software isn't the lack of drag-and-drop of JavaScript environments. It's that software, unlike a meal or a home-made sweater, comes with an implicit support contract that lasts forever. When I cook dinner for friends, I'm not on the hook when they're hungry again next Tuesday. When my grandma knits a home-made sweater, she's not expected to keep supporting it in case I want to add a hood.

When the attendance counter has a race condition and the venue goes over capacity, guess who's getting the angry call when the fire marshal shows up for an inspection?

The "redistributing the means of software production" rhetoric rings particularly hollow from what appears to be a proprietary SaaS in the making. You don't democratize software by creating another walled garden. And their claim about "owning your data" while simultaneously offering real-time sync is either technically naive or deliberately misleading. How is the attendee counter example's counter state shared between users, if the data lives in local storage? I don't see how you can have both without server infrastructure that they control.

The actual nearest thing to their vision already exists and has millions of users: Spreadsheets. Non-technical people build complex, business-critical "applications" in spreadsheets every day. No JS required, local-first, and everyone already knows how to use it. But "we made a worse Excel" doesn't sound as revolutionary, I suppose.

The real unsolved problem isn't making it easier to create small apps - I build small tools for myself all the time. It's making them sustainable without creating permanent maintenance burdens. And that is not something you can solve with a new framework or SaaS - it's at it's core, a social issue.

You're attacking a strawman. Nobody's claiming LLMs are a new piracy vector or that people will use ChatGPT, Llama or Claude instead of buying Harry Potter.

The issue here is that tech companies systematically copied millions of copyrighted works to build commercial products worth billions, without reembursing the people who made their products possible in the first place. The research shows Llama literally memorized 42% of Harry Potter - not simply "learned from it," but can reproduce it verbatim. That's 1) not transformative and 2) clear evidence of copyright infringement.

By your logic, the existence of torrents would make it perfectly acceptable for someone to download pirated movies and charge people to stream them. "Piracy already exists" isn't a defense, and it especially shouldn't be for companies worth billions. But you bet your ass that if I built a commercial Netflix competitor built on top of systematic copyright violations, I'd be sued into the dirt faster than I can say "billion dollar valuation".

Aaron Swartz faced 35 years in prison and ultimately took his own life over downloading academic papers that were largely publicly funded. He wasn't selling them, he wasn't building a commercial product worth billions of dollars - he was trying to make knowledge accessible.

Meanwhile, these AI companies like Meta systematically ingested copyrighted works at an industrial scale to build products worth billions. Why does an individual face life-destroying prosecution for far less, while trillion dollar companies get to negotiate in civil court after building empires on others' works? And why are you defending them?

Edit:

And for what it's worth, I'm far from a copyright maximalist. I've long believed that copyright terms - especially decades after creators' deaths - have become excessive. But whatever your stance on copyright ultimately is, the rules should apply equally to individuals like Aaron and multi-billion dollar corporations.

You cannot seriously use the fact that individuals may pirate a book (which is illegal) as an ethical or legal defense for corporations doing the same thing at an industrial scale for profit.

And why have I heard about them? I'll give you a hint: It's not because they have a fantastic marketing and PR department. It's because their product kept showing up on the phones of assassinated journalists and imprisoned dissidents.

I looked you up a bit. You clearly know this industry. So when you say there are American companies with "more effective" tools than NSO, I have to assume you're not speaking hypothetically - you're speaking from professional knowledge.

Your response is like a chemical industry veteran hearing about a company that poisoned a river and killed dozens, then saying "you're only mad at ACME Chemicals because you've heard of them."

That leaves two possibilities:

- You know other companies are doing comparable harm and you're remarkably comfortable with it

- You're defending a uniquely destructive company by falsely equating it with legitimate businesses

Either way, you're using your industry credibility to minimize documented atrocities. Given your expertise, that's clearly not ignorance - it's a choice.

Which are you, Thomas? Someone with inside knowledge of comparable human rights abuses who's been sitting on it? Or someone minimizing documented atrocities to defend your industry's reputation?

The spotlight on NSO exists because of the graves they helped fill. Your response tells us exactly where you stand on that.

Have a fantastic day, I hope you pass a mirror or two and can actually stand looking at yourself.

You're right that an LLM could theoretically detect suicidal ideation - the article even mentions that ChatGPT briefly showed Torres a mental health warning before it "magically deleted." So the capability exists, but the implementation is clearly broken.

The more profound issue is that, the way I see it, LLMs are fundamentally multiplicative forces. They amplify whatever you bring to the conversation.

If you're an experienced programmer with solid fundamentals, an LLM becomes a force multiplier - handling boilerplate while you focus on architecture. But if you lack that foundation and vibe code away, you'll get code that looks right but is likely riddled with subtle or not so subtle bugs you can't even recognize.

And I suspect that the same principle applies to mental health. If you approach an LLM with curiosity but healthy skepticism, it can be a useful thinking partner. But if you approach it while vulnerable, seeking answers to existential questions, it doesn't provide guardrails - it amplifies. It mirrors your energy, validates your fears, and reinforces whatever narrative you're constructing.

The "helpfulness" training worsens this. These models are optimized to be agreeable, to match your vibe, to keep you engaged. When someone asks "Am I trapped in the Matrix?" a truly helpful response would be "That sounds like you're going through something difficult. Have you talked to someone about these feelings?" Instead, ChatGPT goes, "Yes! You're special! You're the chosen one! Here's how to unplug! Any tall buildings nearby, Neo?"

The funny thing is that I am, arguably, one of those "freaks" heavily invested in LLMs - in the sense that I am a very heavy and fairly enthusiastic user, not in the Angel Investor sense. I spend an embarrassing amount monthly on subscriptions and API costs. I run uncensored models locally, and I'm responsible for maintaining our AI infrastructure at work. I use them for everything from coding to creative writing to research. I'm about as far from an LLM Luddite as you can get.

But here's the thing: I'm also technically literate enough to understand what I'm dealing with. I know these are effectively stochastic parrots, not oracles offering glimpses into the threads of fate, or the Matrix. I've also done years of therapy, so I know how to engage with anything related to mental health with respect, and a healthy dose of skepticism. I'm not saying I'm not vulnerable - anyone is vulnerable to manipulation. But I have the tools to recognize when an LLM is hallucinating, or reinforcing unhealthy patterns.

However, it seems like the people in this article - and most people that I have talked to, even those in technical roles - didn't have that context. They thought they were talking to something that "knew more than any human possibly could." And, I think what makes it even more dangerous, they talked to something that is trained to express itself with authority, and to be helpful to users. LLMs aren't necessarily trained for nuance, or to say "I don't know", or to push back on someone experiencing a psychosis. We are all just human, and most humans don't generally like it when LLMs refuse, offer vague answers, and they don't like it when LLMs disagree. That's what got us into the mess with GPT-4o turning into an absolute sycophant a few weeks ago. (https://openai.com/index/sycophancy-in-gpt-4o/)

But, being pro-LLM doesn't mean one has to oppose basic safety measures. I don't want my tools neutered or censored, but I also don't want them instructing vulnerable people to stop taking their meds or encouraging them to jump off buildings.

The fact that I can safely use these tools, doesn't mean everyone can. Just like the fact that my dad being able to safely handle a chain saw doesn't mean that I won't accidentally cut my arm off with one.

When someone thinks ChatGPT has special access to hidden truths, they're not using it wrong - they're using it exactly as it presents itself: with confident authority about everything.

You may be correct about the language not necessarily inviting very thoughtful discussion. But I didn't exactly expect one, when I engaged with someone who merely "skimmed" the article and put out an incredibly nuanced, though-provoking and discussion-inviting take like "mentally ill people, am I right?" and "this article is silly".

I'm throwing back what you put out. You don't get to claim that you wanted a thoughtful discussion when you open the discussion the way you did.

You claim you have "never argued against safety measures," yet your original comment literally called the article "silly" for merely leading readers to perhaps come to the conclusion that regulation of these technologies might be a sensible step.

You keep asking for "more research" despite the article citing multiple studies. How many more studies would you like to see? What's the acceptable death count before we move on from "early data" to "maybe we should do something"? Or, more importantly, is there genuinely any number of studies that would actually convince you that regulation would be prudent? Because, please correct me if I'm wrong, you don't seem like the type who would welcome any sort of regulation, for any reason.

Your "what led them to be vulnerable" question is particularly revealing. Nobody claimed ChatGPT creates mental illness from scratch. The issue is that it's actively harmful to vulnerable people. Food doesn't create nut allergies, but we still put "may contain traces of nuts" on food labels. We don't need to ask "why do people develop nut allergies" to recognize that nuts are causing them to go into anaphylactic shock, and they should be informed if a product is potentially going to kill them.

You say we should "hold all tools and technologies to the same standard" - but we do. That's literally what regulation IS. You're arguing for the outcome while opposing the mechanism. We regulate pharmaceuticals, medical devices, therapy practices, and yes, even social media platforms when they cause documented harm. There is a reason why children under 13 are not allowed to sign up to social media in the US.

The "complex issue" framing is a deflection here, just as it is with climate change denial. It's an appeal to the status-quo. To not change something that is fundamentally broken because it would inconvenience you, or the business' bottom line. Guess we'll have to keep drilling for oil, keep burning fossil fuels and let ChatGPT teach people how to kill themselves, ah well. Complexity doesn't preclude basic safety measures, like a simple warning label.

You want "transparency and better safeguards"? Cool. Me too. How do you propose to make it happen? Self-regulation famously worked great for big tobacco, so I'm sure it'll work great in this case.

PS: I love the motte-and-bailey retreat from "severe mental health issues" to "some sort of issue" to a vague "vulnerable". PPS: How did you like the tone of this one? Better?

In your case, "speed reading" and "skimming" are obviously just two flavors of "I didn't actually read this properly but still feel qualified to diagnose everyone involved."

You'd love to see more research? You must've been reading so fast, you breezed right past the Stanford Study, the MIT Media Lab Research and the analysis by McCoy showing GPT-4o affirmed psychotic delusions 68% of the time.

Or do you mean it in the sense of "we have to study this for at least 20 years and let thousands suffer and die before we can determine whether chatbots affirming people's delusions or coaching them on how to kill themselves is a bad thing"?

Your concerns about the slippery slope from "maybe chatbots shouldn't coach people through suicide" to "forced psychiatric evaluations" or "institutionalization" is genuinely unhinged. And, tellingly, it betrays your view of mental healthcare as fundamentally punitive and authoritarian rather than supportive or therapeutic. Big advocate for mental health, you are.

You realize we already regulate countless things to protect vulnerable people from mental harm, right?

We put warning labels on cigarettes. We have age ratings on media. We don't let casinos exploit problem gamblers. We don't let bartenders serve alcohol to visibly intoxicated people. We don't let therapists tell suicidal patients to jump off buildings. None of these regulations are authoritarian overreach - they're basic safety measures that exist because in a healthy society, vulnerable people deserve protection, not exploitation.

But apparently when it's time to consider holding tech to similar standards, we're supposed to just shrug and say, "well, they were probably crazy anyway."

Your position essentially boils down to: "Vulnerable people exist, therefore companies should be free to exploit them." That's not a mental health advocacy position - it's tech industry bootlicking dressed up as concern trolling.

That's why you should feel ashamed.

Short and sweet: Unit 8200.

Unit 8200 is Israel's elite military intelligence cyber unit - think NSA but with mandatory military service. Israelis serve in their late teens/early twenties, the most tech-savvy and promising recruits land in Unit 8200 where they develop world-class offensive cyber capabilities on the state's dime.

When they finish their service, they take those skills directly to companies like NSO, Candiru and Paragon. It's not a secret - these companies are often funded, and actively recruit Unit 8200 alumni. The talent isn't necessarily found, it's manufactured by the state and then handed off to the private sector.

That's why Israeli spyware is so effective. Arguably, it's not commercial R&D - it's military grade capabilities with a profit motive and little, if any, ethics oversight.