Hacker Newsnew | past | comments | ask | show | jobs | submit | StochasticLi's commentslogin

it's mostly about AI training at this point. the software for this only supports CUDA well.


don't illegal immigrants commit less crimes per capita?


Parent is clearly not interested in that kind of statistic. Instead their framing implies that even 1 terrorist/rapist/murderer/etc that makes it across the border is a problem worth solving via authoritarianism. The only successful response is to draw this out and make it clearer for everyone else.


Please stop with the straw man BS. That's not what I was saying at all and I think you know that.

HN has reached Reddit levels of partisan nonsense. Sigh.


It's right there in your framing. The most charitable take available is that you're approaching the problem like a software engineer, and accepting that framing as defensible. It is not. One of the most important facts to remember at the level of government is that while we must abstract human problems in order to solve for them, humans themselves are not abstractions. When perfect becomes the enemy of the good (enough), human suffering increases.

All that aside, feel free to walk your framing back and I'll change mine accordingly.


Please stop defending masked thugs that don’t identify themselves kidnapping us citizens and trafficking them to foreign countries.

Previous administrations followed this thing we call “due process”. This admin is not.


No, that’s simply incorrect. ICE is following exactly the same process that has been in place since 1996. It can’t be due process under Biden but then suddenly unconstitutional under Trump.

ICE is not “kidnapping” anybody, and is not “trafficking” them. They are federal police officers who arrest criminals and deport them according the legal process defined by Congress in 1996.


I didn't realize Biden was instructing ICE to grab people off the street and deport them to a gulag in a country they aren't from before they can have a hearing or make any appeals. Would you please provide references for this?


These are presumably threat actors choosing an infiltration route on purpose. At least, that's how I understand the logic behind it.



Grammarly has a function that replaces phrases that are typical for LLM output. Good luck coding that in a few hours.


Last I read, they run it off of MSFT/OpenAI.


I'm sure they haven't created their own model. Probably fine-tuned it + constantly updating a database of the most occurring phrases from all LLMs.


I'm paying for most AI models top tiers and Grammarly. Grammarly is a phenomenal tool. It's not that LLMs can't do it. Well, they can't, but the more important thing is Grammarly's UI.


Agree with this. I love that Grammarly is integrated into everything, and I don't have to switch to an AI chat for minor text edits.


Exactly. Also, it's incredibly affordable, especially when you buy it on Black Friday.


What do you like about their UI and what are your main use cases?


I can see room for improvement about the UI, but I like that it works everywhere and I use it to, obviously fix grammar and mistakes, to add/change our company's specific vernacular (finance), and to change common LLM sentence forms so that the readers don't notice that content was made by AI. Ya know, all these em dashes, "testaments to", "this isn't just X, it's Y", etc.

My favorite function of the UI is that, on any website, pressing TAB just fixes/improves what I just wrote before.


I think we can reasonably expect they will become non-stateless in the next few years.


If agents are stateful a few years form now it will be because they accrete a layer of context engineering.


Why?


That's where the research is going.


I don't think it matters. Here is why I would pay 10x more for using Apple devices than other devices:

1. The ecosystem is just much better, ie. the quality of apps is in another league compared to apps on Android or Windows.

2. Many amazing apps that are at the core of my company's workflow simply don't exist on other platforms at all. With no comparable equivalents.

3. We get many apps earlier. Even up to a year earlier. Example: Clubhouse. Running a company, my marketing team can't afford to be late to a new social media platform, risking that it becomes the new status quo. I can name a dozen apps that were 6 to 12 months late to Android vs. IOS :(

If these change I would happily move to Linux, or Windows (once they make the OS usable).

We use Linux servers for running Python/Node Scripts, and APIs.

We use 1 Windows machine to use a very specific data science app that only exists for Windows.

Before anyone says I'm a fanboy. I'm here to generate revenue for my company. I don't give 2 sh*ts about AAPL.

---

But this is just me.

Why it won't matter to the public: The kids don't care. The want an iPhone.


The CUDA ecosystem is their moat.


Jacob & Co. Watches recently


I would like to know what inference speeds they are achieving exactly on what hardware. I skimmed and searched the article and didn't find that info.


Consider applying for YC's Winter 2026 batch! Applications are open till Nov 10

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

Search: