dvfjsdhgfv

I believe the current game everybody plays is:

* make sure the model maxes out all benchmarks

* release it

* after some time, nerf it

* repeat the same with the next model

However, the net sum is positive: in general, models from 2026 are better than those from 2024.

> "order of magnitude" seems to also be silly-speak very often, trying to sound more technical than "ten times".

But these are two different things. If I hear "ten times" I assume the person actually means ten times; when they say "an order of magnitude", I'm aware they might mean 8 or 12.

LLM-based software is just another layer to be hacked.

> 1) One can no longer trust things out on the web.

I assume you mean software, because we haven't been trusting other things on the web already for decades.

As for software, everybody interested knew about inherent insecurity of supply chain of modern software but the solutions proposed were too expensive. We need an order of magnitude more money lost for organizations to start switching from today's security theater to a model with security built in.

This is mostly a US thing. Netanyahu and Putin are two war criminals according to International Court of Justice. Although Trump threatened the ICJ, this doesn't change that basic fact.

This is quite normal for politics-related news on HN. Many people flag them and they disappear from the FP. Whatever contentious issue you take, there always people who will dislike an even or how it's presented and will flag the submission.

By the way, what are you using it for? I bought Max and Pro plans for Claue and Codex, developed a few apps with it, and after the initial excitation ("Wow I can get results 10x faster!") I felt the net sum is negative for me. In the end I didn't learn much except the current quirks of each model/tool, I didn't enjoy the whole process and the end result was not good enough for my standards. In the end I deleted all these projects and unsubscribed.

While I agree with your points, this one could be more nuanced:

> Infrastructure: Bare Server > Containers > Kubernetes

The problem with recommending a bare server first is that bare metal fails. Usually every couple of years a component fails - a PSU, a controller, a drive. Also, a bare metal server is more expensive than VPS.

Paradoxically, a k3s distro with 3 small nodes and a load balancer at Hetzner may cost you less than a bare metal server and will definitely give you much better availability in the long run, albeit with less performance for the same money.

Really? I can think of a few reasons I wouldn't trust AI-generated recipes.

It depends on the use, I'm not fixed on "productivity" measured by LoC but on code quality. So when using LLMs to challenge my code I'm less productive but the quality of my code increases.

I use codex for cleaning after cloude and it always finds so many bugs, some of them quite obvious.

> One that I retired was used for serving ftp

It must have been a while ago - FTP was practically killed the moment browsers stopped supporting it.

I wouldn't paint the image in such black terms. LLMs can be good in finding bugs and potential issues. And if you like, they can be like IntelliSense on steroids. Even agentic workflows can be good, e.g. for an initial assessment of a new large codebase. And potentially millions of other small tasks like writing one-off helper scripts etc.

> I had to build some tools around linting entire projects

OK, everybody is doing that. And everybody is doing their best at making LLMs more reliable when working on non-trivial tasks. Yet, it looks like nobody came up with a universal solution yet. This is particularly true for non-trivial projects.

This is the exact opposite of my experience. Maybe it was true 10 years ago when K8s was new and trendy so many engineers wanted to try it out. Now it's just boring tech at large orgs.

> The EU themselves "froze" (essentially confiscated) Russian assets, both state and private.

This is a very weak punishment for what Russia is doing every day.

> Most of the time that's probably not very important

Well... Maybe, but what's the point of an answer if you can't trust it? For ultra-fast answers for unimportant stuff I keep Cerebras tab open.

Dang needs a raise - or a team of helpers.

> Who is "we"? Definitely not Google or any other major tech company, they're all actively encouraging this.

Google has been fighting aggressively to replace its search results with snippets, now generated by LLMs, to avoid sending traffic to other websites. If they continue, they will basically lead Google Search to a tipping point where a good competitor can take this market by storm. Microsoft also believed Windows is indestructible and now they have a rude awakening.

> Not sure I understand your question

I'm not sure why you are not sure - S3 was using Cadence's Xtensa 32-bit LX7 dual-core microprocessor, but the article on S31 only mentions "dual core" without too much detail.

Tangentially related but Iran wasn't much of a threat to the USA before Trump decided to attack it. And apart from Israel, nobody is backing this war. The sooner he realizes it makes zero sense, the better for the whole world. It seems that apart from Russia and the USA, other countries are not so eager to start wars. And what is happening now is a bitter lesson also for China: starting a war is easy, winning it is nearly impossible. So I hope we won't really start to build all infra in under-earth bunkers after all.

> YC is, ultimately, not responsible for what these startups choose to do.

Formally they might not be (depends on the case), but morally they are.

> You are paying to be using that limit some of the time.

This makes zero sense. I'm paying to use that limit all of the time. If that's too much for Anthropic, they are free to lower the limits or increase the price. Claiming otherwise would be false advertising.

> Pasting a big batch of new code and asking Claude "what have I forgotten? Where are the bugs?"

It's actually the main way I use CC/codex.

Not sure if the CEO of the USA knows that, he confused Spain with Brazil.

This is repeated ad nauseam, but do they really? Can you honestly say that Nadella's setting a hard target for conversion of local users into Microsoft ones caused their shares go up?

> Like, imagine documentation on object oriented programming being removed because it offended some functional programming folks.

Let's not pretend we are fundamentally different from people living in other epochs, just biases change. We literally changed branch names of git repos because some people in one big country felt the naming could be offensive to another group of people.

Most larger orgs I worked for used Gitlab rather than Github.

Anyway, the core value of Github has always been collaboration - this is where people were. If people go to other platforms, this core value dwindles. And switching platforms is not that difficult.

I basically agree with the practical conclusion:

> For now at least I’m keeping my Claude Pro subscription, but given the persistant rumors of undisclosed rate limiting, and ever improving performance of local LLMs, I can easily imagine that I’ll cancel my subscription before the end of the year.

This trend was obvious when they started removing physical buttons. My thought was, man these people do put so much faith in software.