Great discussion, thanks for not being smarmy or nasty (it's /. I gotta say that). I code for a living and I explain it like this. Police often use police dogs. They don't always use them. However, if they are in a situation where the dog's abilities help, they can really help. A cop cannot sniff out a bomb or might be outrun by a fast criminal. Nobody is outrunning a Belgian Malinois.

Are cops not cops without dogs? Well of course that's ridiculous. Are all dogs cops. No. However, cops can do SOME police work better with dogs. They are just tools that sometimes help in specific situations. They might find new ways to use them that work better. They aren't going to stop using them, but they also don't solve all the problems cops have.

Hmm, yes, I think the heart of what you say is true. LLMs *can* build something, so it's more useful to articulate the limits, not simply poo-poo it's ability to do anything at all useful, which is beyond doubt at this point. It's useful. The question is how much and what are the drawbacks, if any.

Hmm, I can't find much to disagree with. The only thing is that I think you might be over-emphasizing that a tetris clone == shovelware. While it is, technically, it's complexity is fairly low, it's dynamics are well known, and there is reference code for the LLMs to parse/consume before trying. Put another way, I think it's not as complex a task as most things I'd personally consider better examples of shovelware.

Nonetheless, I think your main assertions are true. One can code *something* with LLMs, but as it if it'll meet folks expectations (ie.. the Vibe Coding hype) I still have major doubt. I personally think the flood of crapware is delayed, and it'll take the form of small Tetris-sized programs that one can bite off with an LLM in one pass.

You also make an especially good point about how apps are more than just coding.

Oh, please, the Department of Energy’s input isn’t about virology credentials; it’s about piecing together the intelligence puzzle. Their moderate-to-high confidence aligns with CIA, FBI, and Energy’s growing lean toward a lab leak, especially as China’s data shredding leaves us with precious little to work with. Gullible? That’s rich coming from someone clinging to a debunked wet-market myth.

RaTG13, with its 96% match to SARS-CoV-2, comes from WIV’s own work, where Shi Zhengli’s team engineered chimeric bat coronaviruses in BSL-2 labs that should’ve been BSL-4. NIH-funded experiments juiced transmissibility by 10,000x in mice. No “black ops” needed; natural doesn’t explain that precision.

Showbiz? Rootclaim’s 89% lab leak estimate holds firm because the data does. Peter Miller’s $100k win was a formatted debate stunt, not a scientific overturn. The “show” you’re clutching at is your own narrative, split between ignoring molecular clocks pointing to November 2019 (pre-market) and dismissing WIV staff sick that same year, per U.S. intel. Ignoring facts by repeatedly not responding to them doesn't make them go away, BTW.

Zooming? The 12km gap shrinks when you consider WIV’s 2019 outbreak clues; sick researchers, quiet spread near the lab, not a market 12km off. Density works against your tall tale too; no early clusters outside the market disfavor a natural spillover. Proximity’s the signal; your map rathole the noise.

Did you read Mike's substack article I linked to earlier? It sounds like you disagree with his premise. If so, I'd say you're better off answering his question: Where is all the shovelware?

I've done similar tests with tetris games and building a reminder app. The LLMs can sometimes generate code that compiles, but then fail later to figure out what it's doing and make major modifications. I've had better luck with things like NeoVIM because it keeps the code context a bit more isolated. As I said earlier, with small data analysis or algorithms it works pretty well, for maintaining 200 .C/.CPP files, Makefiles, build-flags, and everything besides documentation, it's been pretty disappointing. Even one big C source file often makes it get lost or spew buggy solutions. That's been my experience and I've tried almost all of the LLMs (yes, recently). I have pro-accounts on several.

The point is that if it was so easy, why isn't it being done?

Oh, sweetie, cherry-picking the CIA’s “low confidence” like it’s a mic drop? Precious. Their 2025 shift says lab leak’s more likely than zoonotic, backed by FBI and Energy Department’s moderate-to-high confidence. China’s data shredding leaves us scraps, but intelligence guys are not buying your market fairy tale anymore.

“No proof” of WIV’s gain-of-function? State Department docs confirm they engineered chimeric viruses, tweaking bat coronaviruses for infectivity in BSL-2 labs that screamed for BSL-4. NIH-funded experiments at WIV made viruses 10,000x more transmissible in mice. Shi Zhengli’s RaTG13 work? A 96% SARS-CoV-2 match. Deny that with a straight face.

Your “exact stall” fantasy? Those Huanan swabs came after shutdown, contaminated by humans, not animals; China cleared those critters out before testing. No intermediate host after five years, and half the early cases weren’t even market-linked. Lab proximity clusters tell the real story, not your spillover "vibes". Miller’s $100k debate win? A formatted stunt; Rootclaim still holds lab leak at 89%, unfazed. It’s showbiz, not science. And your train-riding researcher fanfic? Try sick WIV staff in 2019, per U.S. intel, sparking quiet spread near the lab, not your market soap opera. Molecular clocks peg circulation to November 2019 which is pre-market, pro-lab.

China blaming the U.S.? Deflection 101 to dodge their lab and wet market sins, with PLA ties at WIV exposed. Keep clutching that zoonotic myth; the lab leak’s stacking facts while you’re out here chasing shadows. Good luck with that.

Claude is slightly better at coding (writes about 30 lines before major bugs) than ChatGPT, but Claude's personality is insufferable. I'm surprised it doesn't substitute "actually" for "ACKSHUALLY".

I definitely agree with you. The question is which ones might get done if the level-of-effort was considerably lower? That's the whole pitch, right? They say the AI bots will make it dramatically easier and faster. The lack of shovelware is a hint that all this is more motion than action.

It's terrible at creating big ugly applications. Again, I'd assert this guy is right. Where's the Shovelware? Why AI Coding Claims Don't Add Up.

It's toolitis at it's worst. Suits believe AI will do everything, including the things it's showing the least promise at doing now. I'd urge folks to read this substack post called "Where's the shovelware?" It makes the most obvious point: where's the beef? Every programmer has 10 projects they want to do but don't have time. If Vibe Coding worked, where's the flood of new apps ? I've heard the "well that's coming along" excuses, but it's coming along a lot like the JFK files. They keep releasing "stuff" but it's not at all what we asked for.

Alright, so you think the lab leak theory’s “debunked” because some dude waved around the zoonotic fairy tale for five years? Cute. The evidence hasn’t budged an inch away from the Wuhan Institute of Virology; it’s practically screaming “lab leak” louder than ever. The CIA’s now leaning lab leak with growing confidence in 2025, citing WIV’s sketchy gain-of-function experiments and a biosafety record shoddy as hell. FBI, Department of Energy, and even a 2020 German intel report (80-90% lab leak odds, anyone?) are piling on.

Meanwhile, the zoonotic crowd’s still chasing mythical pangolins in a Wuhan market with no intermediate host in sight. China’s response? Shred samples, gag scientists, and give any attempt at accountability the middle finger. If that’s your “evidence pointing away,” you’re reading tea leaves, not science. That $100,000 debate you’re crowing about? Spare me. Some ex-Google bro, Peter Miller, talked fast for 18 hours in 2024 and snagged Rootclaim’s prize by leaning on shaky market data, not hard proof. That’s not a debunking; it’s a verbal cage match judged by three randos. Rootclaim didn’t even abandon their 89% lab leak model; they just tweaked their debate rules after. Fast-forward to 2025, and new CIA reports, congressional findings, and declassified docs about WIV’s PLA ties and sick researchers in 2019 make your “debunked” claim look extremely suss. Nature’s reputation doesn’t save your argument, and tossing RFK Jr. as a cheap shot just shows you’re out of ammo. Keep clutching that market myth while the lab leak case keeps stacking up.

Wasn't this the same journal who, in March 2020, Nature Medicine published the influential paper "The Proximal Origin of SARS-CoV-2," which argued that the virus could not plausibly have originated from a laboratory and must have emerged naturally? This wasn't neutral science but a tool to shield potentially negligent or secretive state-funded research (at the Wuhan Institute of Virology, linked to U.S. grants via EcoHealth Alliance) from scrutiny, thereby preventing accountability and public outrage. This is poster-child behavior for authoritarian-adjacent journals to suppress whatever their handlers wanted suppressed. Biosafety experts like Richard Ebright, have accused Nature and similar journals of being "complicit in helping to shout down any mention of a lab leak," labeling it a scandal of media and government censorship. They softened their stance later, but the damage was done.

Nature helped legitimize coercive state actions that violated the non-aggression principle. Nature published numerous studies and editorials endorsing non-pharmaceutical interventions (NPIs) like lockdowns, mask mandates, and travel restrictions, often framing them as "large, beneficial, and measurable" for public health. Nature's coverage of anti-vax views as driven by "distrust in scientific expertise" pathologized skepticism, linking it to "anti-intellectualism" and non-compliance, which indirectly supported vaccine mandates and social pressures.

In short, mother fuck Nature. I don't trust you nor do I believe anything you publish. You're government apparatchiks and PAID LIARS.

India always thought the race to the bottom would end with them. They bet on the wrong winner and I'm not sad after years of H1B abuse.