criley2

Strongly disagree, and you've misread what you've linked. These linked cases are situations where people are staying in one chat and posting thousands and thousands of replies into a single context, diluting the system prompt and creating a fever-dream of hallucination and psychosis. These are also rarely thinking and tool calling models, relying more on raw-LLM generation instead of thinking and sourcing (cheap/free models versus high powered subscriber only thinking models).

As we all know, the longer the context, the worse the reply. I strongly recommend you delete your context frequently and never stay in one chat.

What I'm talking about is using fresh chat for questions about the world, often political questions. Grab statistics on something and walk through major arguments for and against an idea.

If you think ChatGPT is providing worse answers than X.com and reddit.com for political questions, quite frankly, you've never used it before.

Try it out. Go to reddit.com/r/politics and find a +5,000 comment about something, or go to x.com and find the latest elon conspiracy, and run it by ChatGPT 5-thinking-high.

I guarantee you ChatGPT will provide something far more intellectual, grounded, sourced and fair than what you're seeing elsewhere.

Apologies, it sounds like you have no experience with modern models. Yes, you can push and push and push get it to agree with all manner of things, but off-rip on the first reply in a new context it will provide extremely grounded and rational takes on politics. It's a night and day difference compared to your average reddit comment or X post.

In my years of use and thousands and thousands of chat uses, I have literally never seen chatGPT provide a radical answer to a political question without me forcing it, heavy-handedly, to do so.

You could say the same about Apple, word for word.

When I look at the state of how humans have manipulated each other, how the media is noxious propaganda, how businesses have perfected emotional and psychological manipulation of us to sell us crap and control our opinions, I don't think AI's influence is worse. In fact I think it's better. When I have a spicy political opinion, I can either go get validated in an echo chamber like reddit or newsmedia, or let ChatGPT tell me I'm a f'n idiot and spell out a much more rational take.

Until the models are diluted to serve the true purpose of the thoughtcontrol already in fully effect in non-AI media, they're simply better for humanity.

"Please include ACTUAL EVIDENCE!"

"I tripled my output (I provide no evidence for this claim)"

Never change, HN.

A lot of people are picking apart the many problematic parts of this, but I'll choose to target "Regurgitating content from the trending influencer of the week."

This is fundamentally how humans work. Our civilization is far too large and complex for you to be a bona fide expert in everything. And even worse, the more time you spend dedicated to expertise in one thing, the less well rounded you are in the ~infinity other things.

So what do you do if you cannot have authentic "taste" in everything?

You trust people. You find "influencers" that seem to succeed and have knowledge and you trust their opinion.

In practice, the vast majority of everything we think and do are not because of our personal expertise, but because of our un-expert belief in other people who do these things. Taste isn't just a measure of your personal expertise, but a measure of your ability to find and listen to experts.

So to call this out as a negative is harmful. You need to assemble a group of influencers not just professionally, but in every part of your life, who themselves are experts and can lend you their taste so you can make decisions without the paralysis of needing personal expertise in all things. That, or you need infinite time to become an infinite expert.

It's not lazy, it's by design. We have chat messages because the actual knowledge is stored inside of people, and chat messages are the most searchable way to see what people know outside of being able to ask them personally.

So why don't all of these people simply write it down in a notion/document store and meticulously keep it all up to date?

Because the business does not want that. We demand efficiency, so we understaff engineering departments sufficiently that there is always a little crunch, so that slightly-too-few engineers have to work slightly-harder-than-they-want to make the business successful. The end result of this intentionally engineered "lack of time" is that things like maintaining meticulous documentation are ignored, and the only time the knowledge is shared is in a frantic slack message.

The business is designed to do this. It's not laziness. It's the standard operating procedure to increase efficiency and profit.

Is this a thing? I've been using a Pixel 9 Pro Fold for one full year now and my inner screen looks pretty flawless. I don't see a single scratch, and I've never used any kind of protector on the inside. This kind of sounds like a "sour grapes" excuse where a really good thing is presumed to suck only because you can't have it. Personally, as someone who isn't really interested in a full tablet, the foldable is really really nice.

As a counterpoint: Using a foreign model means the for-domestic-consumption censorship will not effect you much. Qwen is happy to talk about MAGA, slavery, the Holocaust, or any other "controversial" western topic.

However, American models (just like Chinese models) are heavily censored according to the society. ChatGPT, Claude, Gemini, are all aggressively censored to meet western expectation.

So in essence, Chinese models should be less censored than western models for western topics.

>There have been endless court decisions, eg there's loads of case law, where geoip is specifically determined to be a best-effort for blocking

Bullshit. Absolute hogwash. Cite your case law. Cite a SINGLE court which says geoip is "BEST EFFORT". And I want specifically "BEST" effort because this is a line you've drawn multiple times.

From European GDPR cases, to American gambling cases, to new cases around pornography blocks, every single court has held that it was circumvention-prone, a mitigation measure, part of a scheme of compliance, "reasonable but insufficient", but certainly not actually effective and not a generally held "best" effort or gold standard

Tip: Use AI to judge your comment. It's embarassing to make a real human sift through this. Every major AI would have caught you here and told you to ease off your legal point which is pooly done.

P.S. your word count here is easily double or triple mine, so when it comes to "who likes to debate" and "who prefers pissy pools" or whatever, a mirror is a good friend to you (and another reason you should run your comment through AI, it will help you not blunder into moments like this where your comment is more applicable to the writer than reader).

> something that you do, or a thing that you give someone, that expresses your feelings or intentions, although it might have little practical effect

Perfect definition for the geo block, since it's trivial to bypass and billions worldwide use the technology to bypass such a check.

Thank you for providing a dictionary definition that perfectly captures how the businesses efforts are "token", since literally billions of humans can bypass it with minimal effort.

I disagree that "performative" and "token" are the opposite of "best effort".

The opposite of "best effort" is clearly "worst effort".

You seem to take offense with the idea that the company is doing "the minimum viable legal requirement" and you insist that "no, by doing what the judge says, it's actually an earnest and good attempt!"

If you actually think a company puts in even 0.1% more effort than a court requires of them, then I think you are very naive. Clearly the company could prevent VPNs from working if they wanted to invest the effort, like Netflix and China do, but they literally can't be bothered if the court doesn't require it.

I consider "minimum viable legal requirement to get past the judge" to be "performative and token" because they do NOT actually care if users access it, they want them too, they are only checking a liability box forced on them by the court and their legal department, doing the literal minimum.

>> judge in the area will feel that it was reasonably done

> No ... It's doing a best effort to comply

Generally when you repeat my statement back to me, you do so in agreement.

They don't actually care about the block or ban, they just want to put in enough token effort that a judge in the area will feel that it was reasonably done. It's performative for the legal system.

If you think that predicting economic crash every single year since 2012 and being wrong (Except for 2020, when he did not predict crash and there was one), is good data, by all means, continue to trust the Boy Who Cried Crash.

Ah Michael Burry, the man who has predicted 18 of our last 2 bubbles. Classic broken clock being right, and in a way, perfectly validates the "no one can see a bubble" claim!

If Burry could actually see a bubble/crash, he wouldn't be wrong about them 95%+ of the time... (He actually missed the covid crash as well, which is pretty shocking considering his reputation and claims!)

Ultimately, hindsight is 20/20 and understanding whether or not "the markers" will lead to a major economic event or not is impossible, just like timing the market and picking stocks. At scale, it's impossible.

>stack up the habitual offenders and send them to jail for a few months to few years setting them on straight path.

I'm not sure if you have been to an American jail but they do not set folks on the straight path. They are basically Crime University, and the folks on the inside trade all kinds of information about how to crime more effectively, where to crime, what tactics police use and what neighborhoods are safest or most dangerous for police activity.

I was thrown in lockup for a weekend for not changing my tags after moving and letting it escalate out of control and what I saw in that inner city lockup truly shocked me. Folks had incredible amounts of illegal goods on them (despite having been searched and thrown in jail) and were openly performing transactions, sharing "industry secrets" and coordinating for future work once they were out.

If you have spent any time in an American jail or prison, I think you would be disabused of the notion that you can simply lock a criminal up for a few months and "fix" them. I would suggest that it's the opposite, a few months in jail turns a newbie criminal into a true amateur or journeyman with networking, education and future opportunities.

Your prompt is pretty basic. Both Claude Code and Github Copilot having similar features. Claude Code has `init` which has a lot of special sauce in the prompt to improve the CLAUDE.md. And github copilot added a self-documenting prompt as well that runs on new repos, and you can see their prompt here https://docs.github.com/en/copilot/how-tos/configure-custom-...

Reading their prompt gives ideas on how you can improve yours.

Better to work with the tools we have instead of the tools we might one day have. If you want agents to work well today, you need to build for the agents we have today.

We may never achieve your future where context is unlimited, models are trained on your codebase specifically, and tokens are cheap enough to use all of this. We might have a bubble pop and in a few years we could all be paying 5-10X current prices (read: the actual cost) for similar functionality to today. In that reality, how many years of inferior agent behavior do you tolerate before you give up hoping that it will evolve past needing the tweaks?

That's a really cool mandoline, thanks for sharing. Not only is it safer but it's accessible for folks with motor issues. I'm still using a cut resistant glove and ignoring my plastic guard on a standard one.

Enjoyed reading this. I've followed J Kenji Lopez-alt for a while and I've practiced the "aim below" method for a few years now.

I also like that the article ends with the perfect Kenji-ism. "Yes, technically my method is statistically ideal, but like, it's home cooking and it doesn't matter, heterogenity isn't the enemy". Reminds me of Adam Ragusea (https://www.youtube.com/watch?v=5cWRCldqrxM), we're not making fancy french cuisine, we don't need a perfect brunoise!

A few quick pulses doesn't make mush and is fine for a lot of applications. Otherwise, food processors have dicing kits https://i.imgur.com/cXbZ9aC.png

I enjoy the art of prep with my beautiful wa gyuto, I truly do. But if you put a 5 pound bag of large onion on front of me to dice, I will prefer the machine...

I have a well documented copilot-instructions.md (and have used githubs new agentic self-documentation prompt) and the reality is that it takes about 15-20 minutes to build and start multiple react, reactnative and expressjs projects.

Github now appears to support defining setup tasks in a Github Action that runs prior to the agent, so that's the next avenue of research.

Regardless, the website agent will always be slower. My local is already running and fully ready to go so the ide agent can hit the ground running on any task. The website agent has to spin up a machine and install and build. It will take time.

Curious to try this against the Github (website) Agent. The website Agent is definitely dumber than the vscode agent (because it has to spend 20 minutes figuring out how to build and start my monorepo apps) but on the flip side, it doesn't take up my computer and thus any value it creates is additive.

We have tried out Gemini code review vs Copilot code review and Gemini is consistently offering better code review tips. It has officially caught multiple potential bugs, even a few that reviewers might have missed, so it's definitely been additive.

Observability looks way worse. Github Agent has a full UX built into the Github PR that lets you dig into the agent behavior. This requires you to egress text logs and make sense of it yourself.

Also curious about customization. Github just rolled out "agent writes its own instructions" https://github.blog/changelog/2025-08-06-copilot-coding-agen... which is super cool, how do I customize this one and teach it how to start and manage apps across my monorepo?

Everyone is suggesting that AI rejected this candidate but that brings up two points:

1) Is the hiring AI so incompetent that it did not realize it had a "S-tier pull" in the process and should have immediately prioritized the find?

2) Was the candidate's submission so bad that a reviewing AI couldn't even tell the massive relevance he had to their work?

I suppose, alternatively, Anthropic could just not really care about Claude Desktop enough to hire a specialist for one part of the stack. Perhaps they're looking for much more "full stack AI" who can do a lot. They have 350-400 total engineers, is that enough to hire a specialist for Claude Desktop?

I guess my question is: Did the AI fail, did the candidate fail, or did the AI work well and we just don't know the criteria it was succeeding in using.

>The whole point of going iPhone is not to have to deal with these kinds of situations.

This is actually really funny because Android users have had the ability to use any browser they want for like a decade+, including browsers with adblock built in, and browsers with fully featured extension systems supporting all major desktop ad blockers, and it all just works. One click download, no setup, nothing.

This is one of those places where Apple has intentionally made a terrible UX for you to steer you into their walled garden / first party products. You have to use Safari, you have to dig around in settings, you have to make sure your versions all line up, it's pointless rigamarole that will mean the majority of users stick with stock Safari, just as intended.

In many ways, things "just work" on any platform Apple product managers aren't allowed to muck with...

The concept that you're talking about in the US is a "common carrier" e.g. a taxi can't deny some people or a hotel can't refuse some people.

In the US, payment processors are not common carriers and operate on a contractual regime that allows them to refuse or terminate service for non-compliance, risk management, or policy reasons.

Mobile companies here are common carriers and are much more strictly regulated.

>Incorrect, it applies to companies too as companies are citizens according to citizens united ruling over a decade ago.

This doesn't even make sense. If a corporation is a person, then 1A Freedom of Speech means that the government cannot restrict the corporations political speech.

The corporation is absolutely allowed to restrict their users free speech, including political speech, because A) the bill of rights only binds the government, not corporations and B) it would actually be against free speech to compell a private corporation to engage in speech it does not agree with.

Should you be forced to post political or sexual content that you disagree with on your accounts or on a wall at your house? Of course not. Similarly, if you start a business, you cannot be forced to post political or sexual content you disagree with. Your freedom of speech as a business is what matters here.

The idea that we have "speech anarchy" where all people can say anything they want and punish anyone who doesn't reproduce their speech is insanity.

I mean, they can burn it all up in a fire in a few months. xAI loses $1 billion per month, so this round would be an 8 month runway for that company. OpenAI loses $2 for every $1 it brings in, but perhaps they're only losing $500M/mo, so maybe this $8b could last them more than a year...

I'm not sure I agree that flat design removed the technical barrier to entry. First off, not everything is flat. Second, and I think this is really important, the ability to deliver a beautiful design system and the ability to use a design system to create a nice UX are two fundamentally different skills. The artist that delivers the most beautiful gradient (which apparently using the gradient setting in photoshop is a Big Scary Skill™ that flat design solved) often is not an expert at how best to deliver iOS UI.

And your resident mobile designer who knows everything about iOS and Android probably isn't the best at rolling brand new design systems with or without really pretty gradients.

Because these are two different skills, I don't think the style of the design system really impacts the barrier of entry. Most UI designers aren't fiddling with the finer details like that. They're composing already defined "atoms" into the "molecules" of components and pages.