redsymbol

The Amiga, and this era of computing, were before my time. But it is surprisingly fun to browse through some of this stuff, and get insights into the very early days of software development.

My favorite so far is the "Iff Specs Disk", with technical documentation on an ancient container and data format called IFF (Interchange File Format).

Like in the file ea.iff.85 [0], and this section explaining why common data file formats are useful the first place:

"The problem with expedient file formats - typically memory dumps - is that they're too provincial.

By designing data for one particular use (e.g. a screen snapshot), they preclude future expansion (would you like a full page picture? a multi-page document?).

In neglecting the possibility that other programs might read their data, they fail to save contextual information (how many bit planes? what resolution?). Ignoring that other programs might create such files, they're intolerant of extra data (texture palette for a picture editor), missing data (no color map), or minor variations (smaller image).

In practice, a filed representation should rarely mirror an in-memory representation. The former should be designed for longevity; the latter to optimize the manipulations of a particular program. The same filed data will be read into different memory formats by different programs."

Imagine having to explain to a professional developer today why having something like JSON is valuable. Most of us find it obvious.

But there was a time when this kind of thing was not obvious at all; the need for it had to be discovered the hard way, and then solutions invented from first principles.

We really are standing on the shoulders of those who came before us, in more ways than we realize.

[0] https://amigafreeware.downer.tech/17bit/17bit/61/contents/if...

What's the quality level of what they are producing this way? Are they ending up with good pieces of writing?

The most interesting part of the official announcement [0]:

"As people gravitate toward firsthand perspectives..."

That implies Google is seeing a shift away from people relying on just AI-generated summaries, and checking the cited sources more.

I have certainly found that when I ask a chat tool something, and then go to the human-written source webpage, it's not uncommon I get a different insight than what the AI implied. Not always of course, but often enough that I have noticed it.

[0] https://developers.google.com/search/blog/2026/07/search-con...

This is the right way to use it! You're not making it write for you. You are instead using it to help you write your own words. Massively different.

I wrote a post on linkedin last year titled "Do not use AI to write."

Boy, was it controversial. I could not believe how hard some people were pushing back in the comments.

Quoting from myself there:

"When you write your own words, you are forging your own voice. It is distinctive, conveys your unique world view, and connects with others in a way that is specific to you alone.

If you use an AI tool to write for you instead, you lose all of that."

That seems blindingly self-evident to me, but apparently a lot of folks disagree.

Something else I said:

"Writing is hard because thinking is hard. When you write, you forge your thoughts, distinctions, mental models and even feelings into the clarity of precision that the written word demands. When you outsource your writing to an AI tool, you lose more than you know."

I guess a lot of people don't want to bother with all that.

That's interesting, thanks for sharing this. Yes, I can see that as a big part of the value a publisher can add, in making the book better.

Yep, linked there for anyone who is curious! Given the topic, I was concerned they might push me to stuff AI into it, but like I said they did not do that at all.

This is absolutely true, speaking as someone who has both self-published and also published with a big publisher. Each choice has pros and cons.

In my case, I self-published and sold a book for several years, and then published an updated version with O'Reilly.

I decided to do that because I came to realize people judge self-publish books as less vetted and lower quality.

That may be often true. But in many cases, a self-published book can be much better than those released by a former publisher. I certainly believe it was true in my case.

But in the end, I decided my highest best opportunity was to go with a well-regarded publisher, for the authority that would bring.

And it changed things. People treat me differently now, like they consider me more of an authority. Even though it's essentially the same book; it just has an O'Reilly logo on the cover now.

Whether that should be the case is up for debate...

But it absolutely made people listen more seriously to my message. and I believe it has massively increased the positive impact of that book on the world.

Financially, I think it's been about even. For me it was worth the tradeoff for other reasons, but I don't think that is always the case for every author and every book.

That's great! I am not surprised.

That was definitely my experience.

Haha good question. No, but I did not ask; I wanted to give them as much freedom as I could bear on aspects of the process I was not too attached to, so I let them pick.

I will say I was very happy with the animal they came up with! If I was not, I would have asked them to change it, and I bet they would have. They showed me a preview version early on, so there would have been plenty of time to do so.

I had written and self-published three books, and in 2024 decided to publish the most successful one with O'Reilly. It went up for sale in December 2024.

The whole experience was wonderful. I had basically none of the problems that this fellow experienced with his publisher, and I am delighted about how it went.

I did some things differently. For one, I had already been selling the book on my own for a few years, and was essentially on the 3rd self-published edition. Because of this, they were able to see what the almost-finished product was.

I told them I would not make massive changes to the book, nor would I contort it to the AI trend (the book barely mentions AI at all), and they never pressured me once.

Their biggest contribution was their team of editors. This book has code on just about every page. I had 3 technical editors go through it, finding many bugs. How many? Let's just say "plenty".

And the feedback from the non-technical editors was, to my surprise, even more valuable. Holy crap, I cannot express to you how much they improved the book. There were several of these folks (I had no idea there were so many different specialties for editors), and all of them were great.

(They also accepted my viewpoint when I disagreed with them, immediately, every time. The final published version of the book was 100% my own words.)

From all of that, I made improvements on what must have been almost every page, and rewrote two chapters from scratch. I also added a new chapter (I volunteered for it, no one at any point pressured me to do that). The result was making a book that IMO is at least twice as good as what I was able to accomplish on my own.

I do not resonate with the article author's comments about compensation. He negotiated a pretty good deal, I think; it's not realistic to get much better than what he did, since the publisher is a business with their own expenses to pay, etc.

I was pretty disciplined about meeting deadlines that we agreed to for certain milestones. That helped my relationship with the publisher, obviously.

All in all, it was a great experience, and I am glad I did it this way.

Reading the article, it sounds like my publisher (oreilly) was better to work with than his, but I think he could have done some things differently also. In the end, though, I agree with him that it was best to walk away in his situation.

Unfortunately, just changing ad platforms is no magic bullet. Tiktok advertising is no easier to ramp up on, in fact it will probably be more difficult and expensive.

If are worried about the significant expense and high risk of paid ads, which is very reasonable, I recommend you do not do paid ads at all.

I think a better alternative for you is to learn to create organic content, e.g. videos for IG and tiktok, or text posts for platforms like X. And learn to stimulate sales of your app that way.

This also has a long learning curve, but dodges the financial risk of paid ads. Once your organic content is dialed in enough to make consistent profit, you'll be much better positioned to start with ads and be successful.

Good luck!

I would not proceed unless you are willing to invest $50/day for a minimum of 6 months.

There is also the investment in learning, which would require 5-10 hours per week that whole time.

I say this because anything short of that is likely to lose money and accomplish little.

Not trying to gatekeep, so feel free to try it however you like. Maybe it is worth you trying to set up one ad just to see if you like the process enough to continue.

Is it possible to start small with Meta and then ramp up?

It depends. What kind of offer are you selling?

Generally the learning curve is massive, long, stressful and expensive. I lost tens of thousands of dollars before I could consistently make ads which were profitable.

It's not like learning coding. Imagine if every time you saw a stack trace, $250 was deducted from your bank account. Learning media buying is kind of like that.

Another path is to get hired by an ad agency. They will train you and give you a budget to manage; if you lose that budget, the worst that can happen is they'll fire you (rather than losing all your stuff).

But that would be a switch in career path, which I do not think you are asking about.

I sometimes use Instagram and recently I have seen a ton of ads for a local tech event... but the event already passed a good while ago

As someone who buys ads on IG: this is almost certainly because the advertiser has misconfigured their settings for that ad campaign.

I have spent over $100k of my own money running ads on Meta, over $10k on Linkedin, and also significant amounts on Google (GDN/Youtube/keyword search) and X/Twitter.

My experience:

For most ad campaign types, YES, Meta works far better than any alternative.

In my view, this is because its underlying machine-learning algorithms are better. Not the generative AI mentioned in the article, but the core ML-based learning strategy that every modern ad platform has been based on for over a decade now, allowing it to learn to find buyers for whatever you are selling. Meta is just better at that than anyone else - far better. And has been for years.

Linkedin is quite interesting and can reach audiences no other ad platform can. The problem with LI is that it is tremendously expensive. Think $50-200 per qualified lead, so it only makes sense for very high ticket offers. I've never been able to make LI ads work for lowticket offers (i.e. under $500 gross margin per sale).

I have attempted to run ads on X/twitter several times in the past 5 years, and each time gave up because of how terribly it performed. Literally zero sales from the ad spend, using the exact same ads which worked great on Meta. X is kind of infamous among media buyers for how bad it is. But I haven't tried it in over 8 months, so maybe it has gotten better now. Grok/xAI is quite good, so I am not sure why they haven't made the X ad platform a priority. If they did, I bet they could eat Meta's lunch.

Google is really 3 ad platforms: Youtube, Google Display Network (GDN), and keyword search. These are all radically different in many dimensions; the only thing they have in common is that Google lets you set them all up through the same web interface.

Mostly I run ads selling mid to high-ticket stuff to sophisticated buyers, and I think most of those people have Youtube Premium, which makes YT ad-free. So I haven't tried running ads on Youtube much. My media buyer colleagues who are into it are mostly selling mass-market offers that most people on HN would consider kind of scummy. They tell me it can work well, but requires a lot of ad spend over weeks to start working. In contrast, I can often get a Meta ad campaign profitable within 48 hours.

Keyword search works great, but only for "bottom of funnel" buyers (i.e. people who are actively looking for a solution they can pay for right now). There are only so many of those, so generally you cannot scale up ad spend (and thus absolute profit) very high. In contrast, Meta can create new "top of funnel" buyers, which is a larger audience by many orders of magnitude, which allows you to ramp up some Meta campaigns to a very high level of profitable ad spend.

GDN has been tricky for me. I think it has a lot of potential, but it has a very different model, and so far I have not been able to make it profitable for my offers. It certainly seems easier to set up a successful campaign with Meta.

Short answer to your question: Yes, even with the things the article is justifiably griping about, Meta is just way, way better than any alternative right now.

Media buyer here (among other things), so I feel you. I think it's possible the American FTC might start going after advertisers at some point, because Meta's generated text can in some cases mislead or flat-out lie about what is actually being sold.

The only defense is to have checklists and continually go over running campaigns to make sure everything is configured properly, which as you note is really hard to turn off. Very easy to miss a detail in ad manager's complex UI.

Yeah, as someone who spends a lot of money every month buying Meta ads, I had that thought looking at some of the article's examples.

But in its current form, I think that may happen mostly for very direct-response ads, while creating branding problems that would be expensive for many companies in the long run.

Also, some of the AI-generated creatives and copy that Meta has suggested to me actually misleads or flat-out lies about what is being advertised. Which makes me wonder if the American FTC will go after some companies for running misleading ads at some point, if they are not careful about what suggestions they accept (the ad manager UI currently makes it extremely easy to accidentally approve something you shouldn't).

It's an interesting situation. I think that despite the issues, Meta is the best (and fastest-learning) platform for mass-market purchase conversions. Linkedin has powerful targeting options, but the fact that it's tied to GMT drives me nuts. For LI, typically I use manual bids so I can easily dial down delivery at off times (e.g. if the ad objective is booking a sales call).

All in all, these ad platforms are tremendously complex software systems, and as someone who has been a fulltime software engineer I have a lot of sympathy. But with so much money is on the line, the standards are high.

Sure, happy to chat. @redsymbol on twitter, or my email is in my HN about page

I'm a software engineer who is also a media buyer, which is the term for someone who purchases and configures advertising campaigns. I have spent over $100k of my own funds on Meta ads the past several years, which makes me moderately experienced. I spend a lot of time in the "ad manager", i.e. the webapp that Meta has for configuring these ads, allocating budget, and so on.

I think the current UX of the ad manager will make Meta the target of a class action lawsuit, and there is nothing they can do to avoid that now.

Why: many aspects of the ad manager UI will activate settings that had previously been disabled. The details vary over time, but right now three specific examples come to mind:

1. Promo codes 2. Site links 3. Related media

I won't explain these here (you can ask a media buyer and/or an llm). But these are features of Meta's ad system that are useful in certain situations, but for many types of ads, it is better to disable them.

The problem: If you disable them, and then edit the ad creative (i.e. change the image or video), in many contexts they are silently re-enabled.

This is not noticeable unless you navigate through the complex web interface to check, and disable them again. I now have a detailed checklist, but before that, I would often find I had activated ads with these accidentally active.

The outcome is to increase the cost of the intended result of the ad campaign. In other words, it makes the ads more expensive.

It has certainly caused many media buyers to spend significantly more than they otherwise would to get the same results from their ad campaigns.

These three specific examples have been happening for many months, maybe all of 2025. If they disabled the auto-enable right now, that is still a potentially massive amount of ad spend which has been wasted by many companies around the world.

That is why I say a class action lawsuit is inevitable at this point. There is simply too much money on the table for that not to happen.

Why did this happen? My best guess is poorly thought out internal incentives. I.e., someone in some layer of management has their compensation tied to the percentage of running ads with site links activated, for example. So that person(s) is forcing the design/engineering teams to implement a UX that inflates those metrics. That is the best explanation I can imagine for what I am seeing.

Why?

Not trying to challenge you, and I'd sincerely love to read your response. People said similar things about previous gen-AI tool announcements that proved over time to be overstated. Is there some reason to put more weight in "what people on HN said" in this case, compared to previous situations?

This is where list comprehensions came from! Like we have in modern Python, Haskell and more. SETL was the first language to use comprehensions in its core syntax.

It currently only supports Python 3.11. That is a big reason.

There is another big difference: natural languages have ambiguity baked in. If a programming language has any ambiguity in how it can be parsed, that is rightly considered a major bug. But it's almost a feature of natural languages, allowing poetry, innuendo, and other nuanced forms of communication.

I used to regularly go to my local Taco Bell, but stopped going after they rolled this out. Not mad at them or anything, it was just sometimes a frustrating experience, and overall I was not sure how I felt about it: it's more impersonal, I wondered if it meant less jobs available in my local community, etc. So without making a conscious decision, I just stopped going.

I wonder how this has affected sales and net profit at their locations using AI in this way.

There may be something I do not understand about LLMs. But it seems it is more correct to say LLMs are chaotic - in the mathematical sense of sensitive dependence on initial conditions.

The only actual nondeterminism is deliberately injected. E.g. the temperature parameter. Without that, it is deterministic but chaotic. This is the case both in training LLMs, and in using the trained models.

If I missed something, someone point it out please.

Zoom has been driving me nuts with this lately. I have to reauth with OTP 3 times a day, while logging in on the same computer on the same LAN with the same browser. I spend over $1500 a year with Zoom, and this issue is seriously making me think about how I can migrate to something else.

Different business models have different capitalization requirements; some (e.g. a grocery store chain) are nearly impossible to rapidly scale without massive outside investment.

If a business model has (a) a high enough gross margin, and (b) good-enough intrinsic economies of scale, then by the time it reaches several $100k in revenue, there is a good chance it will be crystal clear how to scale up 7 and then 8 figures. And you will be sitting on a big enough pile of cash that there is no need to take outside funding to implement that. Many B2B SaaS business models in particular can find themselves in this excellent situation.

You can still take outside funding; after all, a growing business at that point is attractive to outside investors. But then it becomes a choice about growth rate in exchange for keeping equity and strategic control. I decided in my case it was completely unnecessary, so I am not doing it.