The US just paid a German energy company over a billion dollars to stop building wind farms it already had permits for.
What the article says
- The Trump administration is paying German firm RWE $1.2 billion to abandon offshore wind leases off California, Louisiana and New York.
- RWE says there is no realistic path to build these projects in the US right now, so it is walking away.
- In exchange, RWE plans to pour close to 20 billion dollars into other US energy investments over six years, including gas.
- This follows similar deals with TotalEnergies and Duke Energy, part of a broader push to kill offshore wind and boost fossil fuels.
- The Interior Secretary framed it as ending costly subsidies, even though the payout itself is effectively a subsidy for gas.
What HN is saying
- Commenters piled on the irony of an official denouncing subsidies while handing out a billion dollar payout to redirect money toward gas.
- Several pointed to Trump's long running personal grudge against wind turbines after a fight over the view from his Scottish golf course.
- One detailed rebuttal argued the project was already troubled before politics got involved, citing lost subsidies, turbine delays and rising costs, calling the settlement a reasonable exit for both sides.
- A side thread spiraled into comparisons with the fall of Rome and broader despair about US institutions, energy policy and climate direction.
- Someone flagged that RWE plans to reinvest part of the money into a gas export terminal in Louisiana, undercutting any climate silver lining.
A New Mexico judge just ruled Instagram and Facebook count as a public nuisance for kids' mental health.
What the article says
- A New Mexico court ruled that Meta's apps are a public nuisance because they harm children's mental health and strain schools, hospitals, and police.
- The judge used an old public nuisance law, not a new tech regulation, to order Meta to pay into a fund meant to address that harm.
- The ruling stops short of banning the apps but is being covered as the first of its kind to hold a social platform liable this way.
- Meta is expected to appeal, and the case is likely to become a test of whether nuisance law can reach how apps are designed.
- Note: the article text itself could not be pulled, so this is drawn from the headline and discussion.
What HN is saying
- People argued over whether the payout is trivial or massive, since New Mexico is small and poor, making it one of the largest corporate penalties ever per resident.
- Several pointed out the ruling rests on an old public nuisance statute meant for things like pollution, and expect it to get thrown out on appeal.
- The sharpest split is between readers calling this deserved accountability for designing addictive apps for kids, and others saying parents should manage screen time and warning the ruling threatens encryption and speech.
- One commenter noted a strange side effect, if the money reaches affected teens directly, kids end up paid handsomely just for having had an account.
A viral essay says AI killed the hard part of coding, so now taste is the only skill that matters. HN spent 500 comments arguing it was written by AI.
What the article says
- The author argues that writing software used to be hard because building anything took real effort, and that effort acted as a filter on quality.
- AI has made production nearly free, so the bottleneck moves from making things to judging what deserves to be made at all.
- He calls that judgment taste, and says it used to be built slowly by making bad things and living with the failure.
- People who skip that struggle can generate fluent work immediately but never develop the instinct to know if it is any good.
- Having taste no longer makes you ship faster than someone without it, since the market cannot see or reward the difference.
What HN is saying
- A huge share of the thread argued about whether the post itself was written by AI, with several readers saying detection tools flagged it and pointing to specific AI sounding phrases.
- The author added a postscript insisting he wrote it himself, which some commenters said read as even more AI generated than the original piece.
- Commenters split on whether taste is even a coherent idea, with some invoking Susan Sontag and Kant and others calling the whole conversation pretentious.
- One sharp pushback said taste is not a durable advantage since competitors can copy any interface or feature within days once someone proves it works.
- Several programmers said the real problem is not taste but that AI generated code and writing tend to be bloated and low on signal, forcing people to read far more to understand far less.
A cheap new open model is matching pricier rivals on tough reasoning tests, and it's converting Claude users.
What the article says
- The page is mostly a benchmark chart testing how well the model handles visual puzzle style reasoning tasks.
- Run at its strongest setting, it solves nearly all of an easier puzzle set and more than half of a much harder one.
- Each of those solves costs only a few cents, which is the real headline here.
- It's plotted against costlier rivals like Claude Opus and Gemini's small Flash model and holds up well.
- The article itself is just numbers and charts, most of the real detail comes from people testing it in the comments.
What HN is saying
- Commenters are struck by how cheap and capable this model is, some running multiple coding sessions all day for just a few dollars total.
- Opinions split hard on quality. Some say it now rivals top paid models, others call it worse than tiny free ones, and effort setting seems to explain a lot of that gap.
- Several people switched over after their Claude account got banned or after maxing out subscription limits, and say pay as you go now costs them less overall.
- A recurring worry is that DeepSeek has warned prices will rise soon, so nobody's sure how long the bargain lasts.
- A few report the model looping or losing the thread mid task, though switching providers or prompting more carefully seemed to fix it for some.
A Chinese model briefly topped a top AI benchmark, then the score flipped minutes later.
What the article says
- Qwen3.8 Max, an open weight Chinese model, briefly ranked first on Artificial Analysis's agentic capability leaderboard, just ahead of Anthropic's Opus.
- People watched the numbers change in real time. One screenshot showed Qwen on top, a few minutes later Opus was back in first.
- The benchmark's own team explained they had just shipped a planned update to how answers get graded, which reshuffled the order.
- On a separate coding specific index, Qwen hasn't even been tested yet, so its real coding strength is still unproven.
- Qwen costs almost as much to run as GPT5.6 despite being open weight, since it uses far more tokens to reach its score.
What HN is saying
- Most agree Chinese open models like Qwen and Kimi have genuinely closed the gap with the top US labs, not just on paper.
- Several coders trust daily use over leaderboards, describing Opus 5 as slow, overthinking simple tasks and burning huge amounts of tokens.
- Others push back, saying Qwen and similar models are inconsistent, needing more careful prompts and more back and forth to behave.
- A loud side thread is anger at Anthropic's pricing, with people describing burning hundreds of dollars in credits for barely any finished work.
- A few doubt the leaderboard itself, since the score swung within minutes right after launch under heavy traffic.