They're tuned to target a certain customer demographic solving for certain problems. I've seen standard AI models to absolutely brilliant things sometimes. But the prompts to get it to perform like it did with GPT-3 seem to get lengthier and lengthier in time. At some point we'll probably just snip out smaller, specialized models to do certain things.
Most tech giants are one hit wonders. Microsoft is the exception, not the norm.
The playbook now is get product-market fit -> own 99% market share -> build, buy, exterminate everything that could threaten this market share -> when the disruption comes, pivot to it.
Zuck inspired the playbook. He's done it to perfection. He's been in the ring for 20 years. Yahoo dominated for about 5-7 years. By no means is he a bad CEO.
Facebook's decline was always a part of his strategy, and whatever the employees say about there being a bazillion active daily users doesn't reflect that their time is near. But Instagram still runs strong and Threads has half a billion active users.
Think that's fair. He's generally been excellent at pissing off users in ways which grow his reach without them leaving and he's been willing to spend big on nascent competitors and actually managed not to destroy their platforms, but I do get the feeling (i) real activity on Facebook actually is tailing off now and (ii) he left a money on the table going for low value ads based on click patterns rather than building events and marketplace into genuinely useful services. But that wasn't his thing and the thing he did have made him enough money to give him bragging rights over an entire room of other billionaires.
Managed not to lose it all on his stillborn metaverse or trying to outcompete AI startups too.
i) I'm sure everyone agrees with this, except maybe a few fans. The best thing he did was slowly wean the whole company off FACEBOOK, even if it's with the Meta smokescreen. Instagram is their core now.
ii) I would disagree on this. Meta is king of low value ads. Google is great at targeted ads, but sometimes you want something to reach as many eyes as possible, as cheaply as possible.
They've barely avoided many of the dark patterns like chumboxes and forced repetition of screens, just because they have enough people scrolling through it. Of course, there's other dark patterns here, but I give them a B+ for low value ads.
The marketplaces and events still provide massive value to the world, for free, as long as they lasted. In the end, it's nearly a trillion dollar company, which is a remarkable thing to build on low value ads.
Does Android Studio count as Jetbrains? It's core to mobile development still.
AI tools are not designed for code reviews for some reason. Maybe everyone is clicking "accept all". I end up reviewing in either Android Studio or even Fork.
It also does code reviews much better than GitHub. Maybe in time, Jetbrains will do a specialized tool for code review that doesn't need to use any AI.
I feel like this should be a service in itself, similar to Heroku or Supabase. Just tick which laws you want to adhere to, upload files to their buckets. Tick another box for audit logs and such and it'll ask you where you need your human in the loop and which buttons those humans need to press. So a bit like Carta or Deel in that sense.
I've had some big enterprise deals fall through because of something like this - military, insurance, fintech, etc.
I agree. I'm trying to sell to enterprise right now and this is a blocker. Basically, my product stores emails in S3, but metadata in Supabase, and caches in Redis. It's all fragmented and super difficult to keep track of. Plus Supabase isn't WORM compliant so it is hard to ensure data doesn't get deleted.
For startups, giving a shit has always been a requirement. Once it get removed from requirements, it becomes a "corporation".
It's awfully hard to find people who do give a shit and even harder to interview for it. It doesn't scale with money, which is exactly why but companies don't look for it.
The results don't lie; it's why companies like Google who hire the most expensive people in the world regularly fail to release something like Antigravity in a stable form when people are building these things solo.
Corporations end up buying these startups for massive amounts of money. It works for everyone because the startups are compensated well for giving a shit and corporations just sit on the gold mine making sure the money keeps flowing.
But it's a good point you raise here. I disagree to an extent - AI will give more of a shit than many humans. It's the main reason many people lose their jobs over it.
So it stands to reason that startups will continue to be AI-resistant. There's a lot of hype that you can build "anything". But where are the alternatives to Facebook, Jira, MS Teams, Stack Overflow, Stripe, and all these other proven business models that everyone hates but is forced to use anyway?
I work for a small start up. We often work with corporations. It does, from a distance, seem like giving a shit separates the two. I feel that. But you look inside big companies, and lots of folks inside care a great deal. I've also been through incubators and accelerators. And some of those people didn't give 1/2 a fuck.
Corporations aren't people. LLMs aren't people. Personification of these things is the source a great deal of trouble.
"Your question is stupid" can be rephrased as, "This is how you can improve your question..." SO used to do this a long time ago, but they decided that the site was being "flooded" with low quality questions and removed that. This seems like something cheap AI can do just fine.
I believe everyone understands the root cause - people are afraid to ask questions, which stems from crappy moderation, which stems from the karmic system. The founders went on to found Discourse which has the exact same problems in community, which hints heavily that it's a problem with the system.
What I would fix is the karma system.
1) Don't give mod powers and responsibilities to people with high karma; moderation skill is intangible and not related to technical competency.
2) Get rid of "bullying" habits. Punish downvoting by increasing the cost to -1. Or do what other social media has done and hide votes, especially to high karma users.
3) Reset karma to 0, but recognize their peak value. Send an email out on why. This should be enough to get rid of all the people who are in it for the karma, and pull back all the people who were in it to contribute to a knowledge base.
4) Fire all the current moderators. They've collectively done a terrible job. Add their salaries together and use it to poach dang from HN.
If there's an incentive to hit 99% non-cash rates, the app just does nothing when it would normally crash. A manager decided not to release a refactor that would fix a bug because it would have jeapordize the crash rate right near bonuses.
Efforts like R&D, clean code, and other optimizations are often punished if they don't affect the bonuses.
The QA team suddenly started making tons of tickets. Misaligned padding on top left, misaligned padding on top right, misaligned padding on the icons, colors are a little off. But they failed to catch the bug where it was not possible to enter real addresses because the address box was too short.
In the long run, the project gets 70% month on month growth and then gets deprioritized.
It's possible they don't know the actual costs. The one-time costs like hardware and training foundation models is huge. There's the ongoing costs like the PhD dudes they hire. There's the data they buy and decide not to use. There's the various offers under one payment plan - limited artifact hosting, cowork, image upload, Claude Code, Code tab on Claude. What about feature parity between the site and app? Who's working on all these?
tldr it seems really complex and by the time they've counted it they probably hired 40 new people for an unannounced feature.
Better prices? Android, especially with purchase parity
Most stable? iOS
Most flexibility and freedom? Android
Better camera or speed? This always enters the debate, but it's specific to the device, not OS
reply