My experience a year ago (back when half of HN was still in denial about what was already working, let alone what was to come) was that Python was the linqua franca of LLMs. You could achieve almost anything that fit in 700 lines or less if you told it to write it in Python.
Times change, and I work more in R&D space than on legacy codebases, but I still ask it to write something in Python then convert it to the actual language on occasion. I don't know if I'm tricking the context window, forcing alternate pathways, or both, but it works.
> Times change, and I work more in R&D space than on legacy codebases, but I still ask it to write something in Python then convert it to the actual language on occasion. I don't know if I'm tricking the context window, forcing alternate pathways, or both, but it works.
My experience with LLMs is that they perform best in one of two modes - either one carefully scoped context or translating between two different contexts without modification - so this modality lines up with that fairly nicely: think in the programming language the LLM thinks "best" in and then translate that to the one you want.
That said, there's often enough structural and conceptual differences between languages that a direct "transliteration" between, say, Python and Go is going to result in some fairly crummy Go, so I'm curious what you see in terms of the fidelity of that translation - do you mostly get "Python written in Go," or does the LLM really do a proper conversion from one language to the other?
I have strict context in place on what I expect from the final language (C# or C++) and I'm frequently left with my jaw open. Used my preferred json library on C++, used LINQ appropriately in C#. Mapped AWS libraries appropriately and used existing credential stores. Certainly better than what I got when I asked for the native version first, which is why I do the hurdle. It feels hacky but it works. In a year it probably won't be necessary.
Some HN readers are blind to this, others are obsessed with it.
ChatGPT claims 900 million active weekly users. You really think a dietician who writes for the Minnesota Reformer (whatever that is), trying to get the word out about his current "evidence-based" whatever, isn't getting a little robo-coaching along the way?
This one sure smells like a human article that went through Claude 4.6 with a "proofread, identify passive voice, increase clarity, adapt to house-style.md, and make it fit in X words" prompt. Maybe the editor did it.
The issue is the quality of the writing, which still needs work whether an LLM was involved or not. Most sources (Forbes, Business Insider) require the author to sign a waiver that indemnifies. That's the chilling effect, not the AI tells.
Hollywood has extraordinarily well-defined controls for keeping things legal and everyone in the chain compensated. Plus a separate Oscars category for it.
OP says he has movies in his head and doesn't have to pay royalties. I told him that if he produces a derivative work, he has to pay royalties. Your comment doesn't follow, but I'll address it.
A trivia host doesn't have to pay royalties to ask questions, and the players don't have to pay royalties to answer them. If that turns into "movie night" at the bar then they have to pay royalties to screen the full film. If a professor plays clips in film class, he doesn't.
Your implication is that an LLM is little more than an brilliant film scholar or exceptionally well-read librarian, and that the matter is settled. The billions of dollars in play across a dozen active court cases say it isn't.
Most artists considered it a one to one exchange. They appreciated attribution and were flattered to inspire people. Some got gigs. Some got laid. The money flowed to DeviantArt, hosting providers, and ad providers. The artists were okay with this. They were the ones paying.
Then DeviantArt built a tool to automate the "make a similar image yourself" part and here we are. It removed all the fun parts: the personal contact, the attribution, the inspiration.
Artists realized they unwittingly contributed to the death of not only the community, but the art form they love. Lawsuits pending.
At a minimum, I do see a lot of AI-as-researcher tells here. You can get Claude to draft very similar essays (of surprisingly quality) if you feed it a target market/philosophy, a few articles for style, then ask it to dig up dirt on any published author in the humanities. It connects the dots and writes stuff that feels just like this article, right down to the meandering. The rough edges and sudden shifts in register is the author editing, then asking for a revised draft.
Claude says: "Verdict: Heavily assisted, possibly lightly edited from an LLM draft. The primary sources are real and the Kierkegaard scholarship is accurate, which suggests a human who knows the material. But the connective tissue and virtually all the 'writerly' prose is machine-generated."
I've written essays in this exact format and I recognize specific tells. He's using Claude Sonnet 4.6 Pro (now Adaptive) as a research assistant then tweaking the output. Know it, done it, smell it.
"The piece moves in a pattern that LLMs default to: historical episode, philosophical summary, contemporary relevance, theological application. Each section is self-contained, cleanly closed, and bridges to the next with a meta-sentence. A human essayist leaves more mess in the transitions."
Now that I've pointed it out, you'll see more stuff like this. It's everywhere.
Evidence degrades, memories fade, witnesses die. Generally the worse the crime, the longer the statute of limitations. Murder in most places has no limit.
Also, if someone hasn't committed a crime in, say, 20 years, there's questionable need to lock them up for three years to deter the behavior. Goal is to optimize the overall system even if some people slip through the cracks.
The state isn't in the business of revenge. It prosecutes crimes on your behalf because private vengeance escalates and frequently lands on the wrong person.
> Financial aid covers the full cost of attendance -- including tuition, housing, food, books and personal expenses -- for most families with incomes up to $150,000 a year. Most undergraduate families with incomes up to $250,000 will pay no tuition.
The morgue manager at Harvard Medical School spent five years selling donated body parts online. The Cornell president just backed his Cadillac into a student asking him a question in a parking lot. This isn't high-trust culture. It's people who stopped believing anyone was watching.
> It's people who stopped believing anyone was watching.
Which, in the era of social media, video surveillance, smartphones and dashcams, is crazy. Once you leave your home, you have to assume everything you do is recorded and might end up online or in court.
These people don't operated under that assumption because they run the systems. They don't care if it's recorded because the system will never make a big deal out of it. Ain't no different than a cop driving drunk because his coworkers won't prosecute him.
Times change, and I work more in R&D space than on legacy codebases, but I still ask it to write something in Python then convert it to the actual language on occasion. I don't know if I'm tricking the context window, forcing alternate pathways, or both, but it works.
reply