llbbdd

Same here. Every time a thread like this comes up it brings out that a surprising number of paranoid helicopter parents browse HN, which is a contradiction in perspective that I can't really emphasize with.

Let's ban books and skateboards too

If they're not building reactors on the land allocated to native tribes, why should they?

Indistinguishable from a joke

What is vibe coding?

Check in in 20 years. The gap between the US and whoever is in second place is enormous. I don't spend enough time on Reddit to think it's close or declining.

Then touching grass remains a good recommendation for you and the person I replied to. Only hard to hear because I wish more of my countrymen were connected to reality.

It is, because of the time component of money. Money is a way of storing the value of labor in a manner less affected by time. If you sell it all at once, its value is greatly reduced. All investments include that time component. Point-in-time net worth doesn't have much value as a measurement in part because of this.

Gold is value, you cannot take it to a grocer and buy food.

I have no idea why the adjusted net worth of people who died centuries ago would have anything to do with money in the modern day. And at a glance, neither of your examples had billions of dollars in cash, which is the point I was making. They may have had a lot of value tied up in illiquid investments, which is exactly like Musk's valuation.

You sound like you don't live in the US. Maybe read less reactionary news.

So pool your money with other poors and enter the infinite money game

They avoid it because it sucks

I mean as far as I know no human being is even a billionaire anyway if you only count cash. It's one of the things the "eat the rich" crowd is particularly bad at internalizing, that there's a difference between value and money that can be spent on food or hospitals.

If you can describe the code you want to write, AI can write it exactly the way you would have, but faster than you every time. That is the floor now.

The problem is that people who are bad writers have trouble understanding that AI writes worse than they do

I'm a man, and I've been using Uber since it launched. Most rides are fine, but there are enough weirdos on the platform that 220 incidents per day that are serious enough to report seems reasonable to me, even if you don't consider that they operate internationally.

I once had a driver pick me up in downtown Seattle, and it turned out to be that he was driving for Uber as a tactic for his entrepreneurial venture developing antimicrobial and hydrophobic coating. He claimed to have applied it to the fishing boat from Deadliest Catch. He was specifically circling downtown to try to pick up someone who could get the ear of someone in Amazon's grocery division that he could pitch to (which I was not). At a red light in a nightmarish seven-way intersection, he took out a square of cheesecloth that had apparently had the coating applied, and attempted to demonstrate its effectiveness by pouring water onto it. It worked, and the water got all over his passenger seat and center console instead while the light turned green and cars behind us honked.

A few months later, Uber tried to match me with that same driver, and I cancelled it and walked instead. I have to imagine that if a guy with that level of high-preparation social ineptitude can stick around in their system, that the number of people making offhand inappropriate moves or remarks must be reasonably high.

In the case I described then, what would normally happen if the fan is blocked after ignition? Like if I had a fully functioning system, booted it up, and then blocked the fan from spinning with a stick, the computer is going to keep the gas valve open and fill my house with CO?

We're going on 13 years since React launched publicly

Gemini at least will produce small functional inline sample apps without being explicitly told to, particularly if you're trying to learn about something, it'll produce an interactive diagram or similar. I can see a future where these kinds of end users aren't necessarily saying "I can build an app for this", but their AI can just produce one when appropriate.

I attribute a lot of it to the principle of "punching up".

Yeah and it's actually possible to avoid pork if you need to. If she wanted to actually stand by her beliefs at this point she'd have to go live in the woods.

I was productive for 25+ years before AI, now I'm more productive because of it. At this point in time, you can do all the software engineering and programming and design yourself, and at worst an agent will be able to literally just type it out faster than you can. I've typed 100 WPM average since I learned how to use a computer, 135+ at a burst, and when I need to type a ton of code that I already understand fully, agents are faster. She doesn't have to outsource her thinking, but this decision will show up on her metrics to her employer whether it's for religious reasons or not.

You wouldn't call a carpenter using electric tools a "Power Tool Bro". I didn't hear of anybody using autocomplete called an "Intellisense Bro". If a someone worked for a shop and refused to use lithium-ion batteries for religious reasons in favor of their hand-crank drill, they would be less productive, and the shop would be better off hiring an equivalently talented carpenter with a modern skillset and fewer arbitrary hangups. That's the level of absurdity on display with this comparison.

She's just using religion as so many ill-intentioned people do - an appeal to authority to justify behavior she intrinsically wanted to act out anyway, and becoming a burden to others in the process.

Seems like she has.

Can you elaborate? I interpreted the same as the other comment that the blower fan just needed a hand start and kept going after the furnace started up. What you're saying only makes sense to me if the spinning the fan by hand allowed the furnace to start by bypassing the safety at startup, but wouldn't that mean that if the exhaust fan was stopped during normal operation (blockage etc) that the furnace would just keep going, dumping CO into the home?

Real tough guy hours

Short answer: yes. So far it has borne out that AI has substantially raised the floor for software development, but that experienced software developers benefit the most. If you don't really know what you're doing, you can still build useful software today. If you do know what you're doing, you can build useful software faster, more reliably, that performs better. Until AI is writing all the code on earth without a human in the loop, I think that it's going to become increasingly valuable to be the person who knows how to pick up where the machine leaves off.

The part of software development that AI replaces the most is typing. Professional engineers like to say that software development is more than typing, and the important parts are, but the reality is that until very recently typing was still a huge responsibility of the work by percentage of time spent. If you spent a week planning and thinking and diagramming and arguing, whatever you were building was probably going to be weeks of typing. Now you can spend the week thinking and get the typing part done in a day, immediately freeing up more time to do the parts that are still complicated.

Now, the thinking and planning and architecting parts of the job, the parts that involve maintenance and keeping things running, and predicting the future, are taking up a greater chunk of the time. AI is getting better at supplanting some of this, and at acting as a sounding board echoing feedback cultivated by the attitudes of millions of talented engineers, but judging the results is still the hard part, and the best judgement still (for now) comes from knowing how the machine works under the hood. I often will have an approach in mind, and will ask Claude to do something without showing my hand. If its approach deviates from mine, I can evaluate it the same way I would evaluate any other coworker's approach, and determine if it's an improvement or not. Sometimes I'm right and I correct it to get better results. Sometimes I think I'm right and I correct it, only to run into some issue I didn't anticipate but Claude did. I don't know how I would have developed that intuition without programming fully by hand for 20+ years, and I unfortunately don't have a recommendation for how anybody new to the field can develop it now, but it still seems important.

What I can recommend, how I learned, and what I think hasn't changed about learning, is to build something you want to exist. Something you want for yourself or to share with others. Build an app or a game, build a website, make some software nobody else has. But the advice I would have that might go against some grain here: use the tools available. Functional software is the goal; understanding is the process. Use Claude or Codex, write code by hand with VSCode or Vim, copy-paste some code if you need it, watch a stream of someone building something similar. I think that's the reality of the career now and it's not going away anytime soon.

But: even with these superpowers, you will eventually run into an issue that the AI can't fix or understand, where it butts up against the reality you want from the result. It'll get caught in loops where no matter how you prompt it fixes and breaks and fixes and breaks back and forth in the same ways, and that's where you come in. That's your alpha. Figure out what's going wrong by reading the code, read the docs, ask the AI questions about the code, and learn how to correct it. There's a saying that computers are bicycle for the mind, and AI strengthens this metaphor a lot. A bicycle can amplify each of your steps into ten steps, but that's only valuable if you know where you want to go.

All of that said, I can't see the future, and I think we're early to figuring out where the limits of this tech are. It might turn out to be a step change in how we work, or it might ultimately obliterate most office work as we know it. Maybe everybody on HN will be retired or a tradesman in ten years. What I can say is that programming teaches you how to think and how to solve problems; even if this field is unrecognizable in our lifetimes, those are transferrable skills in work and in life, and there are worse ways to spend your education.

There is a setting that fixes this, I can't remember what it's called off the top of my head

An article laser-targeted at HN's front page, making tantalizingly negative and easily disprovable claims about Javascript? Perish the thought.

Two things that I would recommend trying out if you're interested in exploring this further:

1. If you're not paying for a model, the results will be worse. That sucks but the free access models are just not very good for anything where you need to trust the output, even for basic queries.

2. More important than #1 is access to tool use. If the LLM is just producing a nutritional breakdown from its weights, it's almost always going to be wrong. If the LLM is allowed to break the problem down into deterministic steps, it will do a lot better. In the nutritional breakdown case, an LLM with search + tool access can pretty easily break the problem down:

- Searching the web for a recipe or ingredient breakdown for the food

- Searching the web for nutritional qualities of each ingredient per some volume of the ingredient

- Writing and running a script with e.g. Python that takes in the recipe's projected serving output, the desired serving size, the amount of each ingredient etc, and scales the ingredients to match the desired serving size, and sums the nutritional qualities of the scaled ingredients.

I've tried this specific case with Claude + Gemini for my own purposes and they both handle it very well. The challenge currently is that the models will not always arrive at this approach when provided with an ambiguous prompt; sometimes they will, but sometimes they'll just vomit up a fully autocompleted response from their weights. Being more specific in the prompt or defining a skill that details the intended approach lets you get more useful + deterministic results while still taking advantage of the fuzzy glue that LLMs can provide here between steps.

Same with the classic strawberry r-counting case. IIUC LLMs have trouble with this because of how training data is tokenized, but any LLM will have no trouble farming out to e.g.

> echo -n "strawberry" | grep -o "r" | wc -l

> 3