I have been a member of HN since 2009. Back then I was fascinated by startup culture after seeing major exits like AdMob and Zappos. I used to think there was something magical or lucky about achieving startup success. But what I learned from 15 years on HN is that the secret to a successful startup is pretty straightforward (if not easy). Solve a problem that a lot of people or businesses have, charge money for your solution, talk to customers, stay lean early on, fail early and often, iterate and improve, and don't run out of money.
I was having trouble getting GPT-4o to extract data like address, email, phone, tracking number from random emails in an inbox. Sometimes it would do it perfectly and other times it would fail miserably on a similar email. Then I tried asking it to first markup the email with schema.org metadata. Then I asked it to extract the data from the schema.org markup. That worked nearly every time.
Maybe there is an extra step you can work into your prompt that would help it get to the proper classification
There is a reason it is called vibe coding and not vibe architecting or vibe systems design. I have had better results when I come up with the overall architecture first. Then I'll vibe code to stub out classes and methods, and then vibe code the tests, then vibe code to fill in those stubs, making sure the tests pass. You have to handhold the AI like you would an intern or junior dev.
Please, no more ding ding. You have a backpack on your passenger seat, ding ding, passenger seat belt alert. You're driving at 5mph on private roads, ding ding, driver seatbelt alert. Your TPMS hasn't worked for 3 years, ding ding, 'service tire pressure system'.
Please let the user decide how they want to use the vehicle instead of a one-size-fits-most model. Also, please give me the option to add a baby seat monitor - if I want one!
My car beeps occasionally, but it certainly doesn't rise to the level you describe. Some of them represent real safety issues, even if you don't think they do in that moment. For example, my TPMS has saved alerted me to low tire pressure when I didn't know I had a leak, and the fact that it beeps a couple times every time I start the car is both helpful ("oh right, I need to get that fixed soon") and far from annoying.
Accidentally leaving a kid locked in a car on a hot Summer day is beyond horrific. How many kids should die before we think the annoyance of an extra beep would be worth it?
> Accidentally leaving a kid locked in a car on a hot Summer day is beyond horrific. How many kids should die before we think the annoyance of an extra beep would be worth it?
It's not about annoyance, it's about whether it's effective at all.
If the car dings every time you turn it off to remind you "check back seat", it doesn't matter if the alert is completely unique and obnoxious and annoying, you will be trained to ignore it and it will quickly become ineffective.
There's a whole field of study here ("alarm fatigue" or "alert fatigue") that's generally looked at in terms of things like healthcare or aerospace. For example, there's a study in healthcare[0] where they found that when dealing with a system warning about drug interactions (including critical dosing errors, fatal interactions, etc) providers overrode 96% of alerts. Their "high priority drug-drug interaction" alerts were overridden 87% of the time, and on review only 0.5% of those were deemed appropriate. Other studies[1] have directly attributed this to repeated exposure desensitizing people and training them to ignore the alerts. People have died because of this.
I have a kid. I can't imagine the horror of being in that situation. I am certain that it would completely and utterly break me. I am fully onboard with a system that prevents this happening. I would be fully supportive of regulating a system that prevents this from happening. More unspecific beeps and dings is not that system.
The problem is that the driver might have so many beeps, that they decide to ignore yet another beep.
I suppose a beep that sounds very different would get their attention, like for pilots in plane cockpits. A terrible stand-up comedian suggestion would be to reuse the plane's "retard, retard!" for parents who forget their kids...
The problem is it ALWAYS dings (the car seat is always there or it detects door openings) and so it just gets ignored along with the fifty other billion warnings it gives.
Which is insane because it obviously has much more intelligent sensors - if I try to lock the doors with the windows down and the wind blowing it screams bloody murder and refuses to lock because it detects motion. Windows up and someone moving? Same thing.
An app like an AI resume customizer based on job description will quickly devolve into a million clones. Something with a network and a first-mover advantage would have some staying power even if people clone it. For example an app where you post a job description and it sets up an AI voice agent as a virtual employee ready to be onboarded. once you have thousands of of virtual employees with company email addresses and attending zoom meetings, it will be hard for clones to break into that vertical
That's the double-edged sword aspect of AI: these models are commodities so you can expect to very cheaply build almost anything out of code as the price of code generation gets driven down to zero through competition. But so can anyone else.
Exactly...so if anyone can generate the app in a few days, you've got to find a vertical where you sign up everyone first and lock them in. Then the. next 10 competitors who generate the same app won't be able to get your customers.
It is hard to predict a new advancement in EV technology, but if we get to a point where an EV is cheaper than the equivalent ICE and charges faster than you can gas up (so about twice as fast as the fastest EV today), and has a longer range than the equivalent ICE, then people will buy more of them. Then we will start to see more EV chargers than gas stations.
Advancements in AI might make it possible to have attendant-less stations if you can have automated sales of snacks and other merchandise, and automated payments for chargers and gas, maybe you don't need an attendant on duty 24/7
I meant that a lot of people still pay cash for gas, but a move toward more electronic payments or at least an automated cash kiosk would let you get rid of those last use cases for a human attendant to take cash
It’s very exciting imho, uptake of EVs and hybrids in China is contributing to noticeable oil demand destruction, and Chinese oil companies have already called the demand peak.
