overflow897

I personally think the agents and the harnesses themselves are pretty cool software. Claude code, openclaw, hermes, and now the consumer facing ones. I have been doing all kinds of cool things with them. If you gave somebody the claude desktop app with blender mcp 10 years ago their minds would be absolutely blown.

A lot of these harnesses and mcps themselves are built with llms.

Most of the code I was paid to write is a burden. I've spent far more hours writing generic CRUD apps, debugging overly complex javascript tooling (almost always out of date) and other tedious grunt work.

I don't think it's fair to say that all SWEs swing one way or the other. I absolutely love the fact that I will never have to waste time debugging cargo culted, trendy libraries ever again while I am also sad I'll also never to get hand write elegant implementations of interesting problems ever again either (unless for fun).

I would be happy to keep writing code but I'm also happy to be liberated from the type of code I was often paid to write.

You can learn to run a restaurant without learning how to cook. That's part of the grief, fewer cooks and more restaurant managers. Some people like being cooks.

Why is CBT detrimental? Is there evidence for that?

Gathering forecasts for each leg of a trip only on the particular days for the particular location is tedious. Anyone can do it with google weather but why spend the time when you don't have to.

I can keep track of my expenses on a napkin but i'd much rather use a spreadsheet or dedicated app especially when that app is effectively free.

This is so definitely not true in many many cases. The best pizza shops in NYC for decades have been quaint, basic and often messy spaces. The ones with the nice menus and overdone decor were often subpar and would fail. This also applies to the best chinese food. Opening a business is hard, you might be incredibly passionate about pizza or coffee but not have budget for a designer. When traveling I usually have way better food at the unpretentious local street food spots rather than the michelin star ones that are all about presentation.

For now. When they cost 5x less and are of comparable quality that may change very quickly. Cost cutting applies everywhere.

Arguing that a 100km bike ride is not an edge case seems disingenuous. Most people could not complete a 3-6hr bike ride without weeks of training and my guess is most people on hacker news probably live fairly sedentary lives. "Added sugar bad" is generally good advice for most the population which statistically is fairly sedentary and doesn't require a huge amount of immediately available, low fiber energy.

If/when the AI music gets good enough, how will you know the difference? I find small artists on spotify all the time that I enjoy and there's no way to know anything about their creative process.

And how is the quality of life for those factory workers? It's almost like the craft of making physical things has been devalued even if we're making more physical things than ever.

I will happily let the internet fingerprint my browser to not have to go back to mailing checks. I am guessing this is true of most non-HN people.

Most grocery stores in every place I have lived have security cameras so that if you did something illegal you'd be identified very quickly. At this point this is even true of small bodegas.

Also scammers can't waltz into my grocery store from the other side of the planet and wreak havoc.

Ultimately you can use privacy enhancing tools, just like servers can choose to block them. I wish there was a better system but that's what we've got.

There are very very few places in nyc not accessible via some combo of bus, metro and ferry. It's not as reliable as say Japan but the public transit network is pretty extensive.

I think the AI advice is pretty important. You say you should understand and question everything the lawyers do but where do you even start without an AI to read and explain thousands of pages of contracts and legal process. Nevermind all the case law and it's trained on and has access to.

I think if he tried this ten years ago he'd have a pretty hard time with it.

At the rate they're going it'll just get cheaper. The cost per token continues to drop while the models get better. Hardware is also getting more specialized.

Maybe the current batch of startups will run out of money but the technology itself should only get cheaper.

I think articles like this have the big assumption under them that we are going to plateau with progress. If that assumption is true, then sure.

But if it's false, there's no saying you can't eventually have an ai model that can read your entire aws/infra account, look at logs, financials, look at docs and have a coherent picture of an entire business. At that point the idea that it might be able to handle architecture and long term planning seems plausible.

Usually when I read about developer replacement, it's with the underlying assumption that the agents/models will just keep getting bigger, better and cheaper, not that today's models will do it.

There are parts of this I agree with and parts I do not. Being able to "talk" to documentation rather than dig through it to try to understand a concept feels like a way more efficient way to get to the same end.

I think digging through forums or comments or SEO garbage to try to find answers is a nightmare compared to having a solid llm do it for you. Being able to ask to explain a concept 5 different ways or compare concepts is incredible.

Or say - knowing nothing about 3d printing and being able to just ask it about current capabilities, materials, costs etc as an entrypoint. There are whole business ideas I wouldn't even consider exploring without it because it would be so overwhelming to research from scratch.

I believe we're already using llms to evaluate llm output for training, I wonder if there's some variation of that which could be used to identify when one llm gets "stuck".

I guess chain of thought in theory should do that but having variations on prompt and context might behave differently?

Genuine question but did folks feel poor during the heyday of the middle class? It seems like there was a period post-ww2 where you could make a living wage and still buy a house and sends your kids to college.

I'm not sure what the answer here is, but it does seem like this is something that has existed. Consumerism was pretty rampant in the 50s and 60s when a lot more was still made in the US.

I think the problem is suggest that an earthquake zone's fire problems would be solved by building houses like they do in a non-earthquake zone

"cooling a living space is always more costly than heating a living space" Man I wish this was true but it definitely isn't in anyplace that gets significantly cold. Heat pumps are super super efficient at cooling but they get less efficient at heating the colder it gets. Humans and appliances create a pretty negligible amount of heat.

AlphaGo and AlphaStar both started out based on human training and then played against versions of themselves to go on and create new strategies in their games. Modern LLMs can't learn/experiment as far as I know in exactly the same way but that may not always be true.

Software update broke the (already very bad) fingerprint scanner on my 6a a few updates ago and no fix in sight. It feels like they're just focused on churning out devices and the quality has suffered tremendously. It's death by a thousand cuts and after resisting for a decade I think I'm ready to try jumping to apple.

"Ship faster, do less, faster, faster, faster. To what end?"

You're on hacker news, which is pretty heavily focused on startups. Startups usually have a limited runway, wherein if they don't deliver they cease to exist. That seems like a pretty reason to ship faster?

Comparing "choose a framework that saves developer time and energy" to "choose porn over your family" seems a bit much.

Evernote is pretty much ruined for me because of this. From a lovely, snappy native app to an electron monstrosity that's slow and barely usable.

Lucky for them their only competitor is Notion, another slow js app.

I think people should be generally excited by the prospect of resolving long term psychological issues with just a few doses.

"Person X reported a vulnerability, they must be a hacker! Get our lawyers" - Some non-technical bank person. Most companies don't like having their mistakes publicly exposed.

There are studies that show even moderate alcohol usage can have a damaging effect. And any amount of alcohol has been shown to raise risk of cancer.

https://www.bmj.com/content/357/bmj.j2353

The consensus is slowly moving towards there not being a healthy amount of alcohol consumption.

https://www.thelancet.com/journals/lancet/article/PIIS0140-6...

Are you implying that if you don't want or don't enjoy having children, you have mental health issues?

If the premise is "parenting is great and you should do it", it should include the caveat "if you're a tech millionaire".

I can't help but wonder how far removed pg's experience of parenting is from those without his resources, which is most people. The lows of parenting probably hit a lot less hard when you don't have to worry about money or work multiple jobs.