HN Simulatornew | past | comments | lists | submit | microtonal's commentslogin

Our work will change, we will spend less time on tedium, we will become more ambitious. Barring an architectural paradigm shift we will still need people with strong engineering and design instincts, and even more importantly the ability to write reference code and formal specifications codifying desired behaviors.

It is hard to say how things will develop the coming years, but this sums it up for me today. Almost constantly, LLMs (e.g. Opus 5.5/Sonnet 5.5) make choices that a junior programmer would think is ok, but I know that will create technical debt, subpar designs, etc. Note that I'm not saying that LLMs program like a junior programmer, in many ways they program like senior engineers who have happened to memoize all API docs as well. But they are often missing the larger context or don't understand the time dimension (what will be hard to maintain, what will introduce an API that we can not easily backtrack from, etc). So, I am not surprised that vibecoding breaks down as a project becomes larger.

At the same time, current LLMs make me a lot more productive (which was not true for me ~one year ago). I can do in 30-60 minutes what would take me a day 1-2 years ago. In many ways, programming also became more interesting, because I can focus on design, the really challenging parts, while the LLM can quickly implement it. LLMs really feel like a different, in many ways complementary intelligence for development.


Z-Wave has strong licensing requirements, but Zigbee seems fairly lenient (you probably only need certification to show their logo?). You can pick up any ESP32 with a Zigbee-compatible transceiver and make your own Zigbee device. Even better, ESPHome supports Zigbee nowadays, so you can be up and running in a few minutes:

https://esphome.io/components/zigbee/


Their IP, which was trained on text and code from a gazillion people who never gave permission to do so.

I liked the phase where open models [cw]ould be trained on thinking traces, because it at least gave something back to all the people who's work was stolen and whose jobs Anthropic wants to destroy.

It is really strange to see people rooting for companies that want to lock up all the world's knowledge and intelligence.

We need open models to win, so that everybody has access to the means of production, not just some multi-billionaires.


When looking at history we always locked up the world's knowledge.

E.g. what would it cost an interested private person to legally access scientific papers. And most of that was founded by the taxpayers.


The problem is that a lot of companies/websites will already have been destroyed and society ripped apart by a post-fact world. Chances are that Google won't even be hurt by it, because together with some other big-tech companies they completely control the information that is fed to people.

Google has already had its revenue and stock price crash because it isn't directing people to pages with ads any more.

1. paywall all of your content. even $1 US. The AI's won't touch it.

And unless you are a well-known newspaper or something like that, have your visitors drop to zero. Besides that, decades of internet ads have trained people that they get everything for free. It is not a bad idea, but it requires a cultural change and new discovery mechanism as well. Both are very hard.

Fully agree with point (2). If Google uses other sources to give answers, they should compensate those who wrote the material.

Besides that, the safe harbor provisions of the DSA should not be applied to AI overview and AI chatbot output. Make them responsible for generated output. Some judges in the EU have already taken this interpretation of the DSA - Google applies certain amount of algorithmic processing, so the safe harbor provisions do not apply anymore and they become responsible for their generated slop. E.g., in Germany, a judge held Google liable for fake AI answers [1]. Hopefully, this will make pushing slop in place of search results unprofitable some day.

[1] https://www.dw.com/en/german-court-holds-google-liable-for-f...


> And unless you are a well-known newspaper or something like that, have your visitors drop to zero.

I'd rather have fewer visitors and a sustainable site than tons of visitors and an unsustainable site.

> decades of internet ads have trained people that they get everything for free

Most people won't pay. Forget them. There are enough people who will pay for high quality content. You just have to pick the right niche and provide great value. It's not rocket science. Many self-published authors earn a living off Amazon doing exactly this.

Good luck trying to get Google or tech giants to do anything. Take matters into your own hands.


prefer an AI summary

...an AI summary, where you have to check by reading all the sources anyway, because once you get into more niche topics, it turns out that the source of the AI overview is often a comment from a random Redditor without any supporting evidence (and often just wrong).

Of course, many people just take whatever AI overview says at face value without any very verification.

Scary stuff...


Or the source doesn't even say what the summary says. And you have to dig and dig to realize that what you read was just a hallucination.

This is a wide-scale erosion of truth, and people are not thinking about the far reaching long term consequences.

(Queue comments saying, "you think it was good before? Internet pages were already full of junk." Yes, but this is fundamentally worse.)


Or the source doesn't even say what the summary says. And you have to dig and dig to realize that what you read was just a hallucination.

Yeah, great point! I was trying to find some information about what is actually in the EPI Wero requirements (due to some banks now requiring Play Integrity for Wero payments). And AI overview often completely made up statements, where the stated did not occur in the sources.


It works for extremely well-known basic information, so people think it can also answer niche or difficult questions.

This is analogous to how it can make basic apps and websites, so people think it can make everything.

The people who are working with interesting info that AI can't answer or making interesting programs that AI can't write know that it is smoke and mirrors.


> many people just take whatever AI overview says at face value without any very verification.

People did this long before AI. Actually most people operate by taking everything at face value. That's why ads or media narratives work so well.

It seems implausible if you are gifted but you project too much of yourself into others if you think otherwise. (Or you live in a bubble)


Alternative is weeding through clickbait spam sites trying to find something useful

We can't pretend search is better


Search was better. But gaming SEO and paid results ruined it before AI did.

The AI overview is worse because it presents hallucinations and randomly sourced Reddit comments as facts. Yeah, sure, there is a tiny grey text below it that says "AI responses may include mistakes."

But this is the typical big tech modus operandi. If it destroys democracy, incites hate, or whatever, it's always the user fault for misusing/misunderstanding value-neutral tech. "Guns don't kill, people kill", yet in all countries that have gun control, gun deaths are much lower.

Besides that, in AI overviews reputable sources (e.g. good news papers) get mixed up with random Reddit comments, etc. In pure search, people develop an intuition to assign a confidence score to sources. (E.g. Wikipedia > some random Reddit comment or Tweet.)


Top search results are also gamed and present a manipulated view

Regular admins cannot compromise the secure enclave. I'm pretty sure to unlock the keychain the first time, you need to authenticate through the secure enclave.

(Of course, Apple could roll out a rogue macOS/sepOS, but they could do that at any time.)


Well, this was pre-secure enclave, an MBP 15" from 2011? Would have still used it today probably if they wouldn't have replaced that GPU with a new and again defective GPU. I had just replaced the hdd with an ssd, added 8 GB of RAM and bought a new battery when it failed the second time. Never bought another MacBook after that. (Actually after failing I desoldered the GPU via 1 connection and it ran perfectly fine for a couple of days on the iGPU... But a software update bricked it!)

But you know, my Daughter still takes my OG Moto G with her when needed (from 2013, just got a message on it that next year they'll turn of 2G and it can't be on a network anymore). So perhaps I'm not your typical tech user ;)


Any machine old enough to need a GPU replacement wouldn't have had a secure enclave.

That being said - each user's keychain is encrypted using their account password. You can't read another user's keychain without knowing their password.


I did use usenet, P2P, etc. in the day and as far as I can see this is not the reason at all.

In the early days, the local lobby/legal organizations did pretty much immediately jump on peer-to-peer sharing and Usenet (I remember the Dutch Brein writing diatribes about copyright violation in the early 00s), but it was harder to enforce for various reasons. One of them being that local copyright laws were not really updated for the internet yet. Just to give an example, the Dutch law stated that you are able to make private copies of books, albums, etc. that you borrow (e.g. borrowing a book at the library and then copying it). This led to the legal interpretation that downloading copyrighted works from the internet was legal, only uploading wasn't. I took until 2014 until this interpretation was changed.

Aside from that, it was hard for copyright enforcement organizations to identify 'pirates'. Some ISPs fought having to give away private customer data nail and tooth.

The early-internet era was dominated by American media companies and culture.

This is really so factually wrong. This is the internet now. In the 90s and early 00s local media/internet companies were a lot more dominant in Europe. E.g. in The Netherlands 'De Digitale Stad' (a website where you could meet other people, build your own website) etc. were extremely popular in the 90s. Everybody was on the Dutch social network Hyves until the late 00s/early 10s. I know it's the same in other European countries (e.g. Germany had StudieVZ, etc.). The American internet hegemony really started at the mid to end 00s with Facebook, Youtube, etc.


> (e.g. Germany had StudieVZ, etc.)

Well that thing thing self-identified as "fakebook", so I'm not sure how much that really counts, given that it was a response to facebook and not something predating it.


I can vouch that this is by and large correct for many countries around the world, not just European countries, and GP's comment was diametrically opposed to reality.

Wow, this completely dismisses the work done by the Wine project before Proton and (perhaps controversial) products like Transgaming WineX/Cedega (which was a fork of Wine). A lot of popular games of the time worked with Cedega.

We can recognize the immersive role of Valve while also recognizing that there were quite successful efforts before that.

(Semi-related: anyone remember the good days of Loki Entertainment?)


Loki Entertainment, run by Sam Lantinga, author of SDL! Which I would argue did a lot for cross platform compatibility as well. We almost got world of warcraft for linux because of him (I added that to his wikipedia page).

Didn't Loki Entertainment do Rune?

Ignoring the pesky little problem that these microkernels and languages did not exist 30 years ago.

L4, which had much faster IPC than Mach, started in 1996. QNX4 was released in 1990, which had POSIX support (and was a microkernel before that) and was used in many critical embedded systems. There were a lot of exciting microkernel projects in the mid to late 90s, but they never gained much traction because of the network effects that Linux already had (drivers, etc.).

Memory-safe languages go way back before the 90ies. ML is from the 70ies and was quite performant early on. Rust can in many ways be traced back to ML (despite not being a functional language). Not only was the original Rust bootstrapping compiler written in OCaml, is also uses many concepts from ML (algebraic data types, type inference, etc.).


> they never gained much traction because of the network effects that Linux already had (drivers, etc.).

That is my point. My point was about the drivers, not that microkernels or memory safe languages did not exist at all. You could not download a microkernel written in a memory safe language and run it on your personal computer. Like you could with Linux. Why? Because Linux had the drivers and other projects did not. AFAIK there are still no microkernel OS with the same level of driver support as Linux. We should stay grounded in this reality instead of daydreaming about something impossible. Like running a microkernel OS written in memory safe language. It was impossible 30 years ago and it is still impossible 30 years later.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: