tokioyoyo

Yeah, weird, I’d say? Even in Japan salary range talks come up during the first call. Or in intro emails.

7M param doesn’t do web searches for you. Otherwise, yeah?

Yeah, agreed. If Claude/ChatGPT would load and respond faster, i’d go for them. But Google’s speed is unmatched for quick queries.

Sorry, not going to take the bait here. I would argue in 2024/2025 it was dogshit, but it really isn’t nowadays. You can even see the sources for most of the summary queries, like literally on the same line as the bullet point.

My google queries are like different combos of weather, specific event questions, directly attributed to wikipedia data and etc. LLMs are way past the point of reliability when it comes to data extraction from static websites.

Yeah fair. I guess I have a very different experience, as I mentioned above, I basically never use it for any niche/hard queries that depend on multiple sources or research. For all my common queries, it's reliable.

As I mentioned above, if you’re making a niche query, maybe. If you’re making a very generic query “average temperature in Hong Kong in January”, it’s fine. You’re overestimating average complexity of a query.

Why do you think there’s a high error rate for general queries nowadays?

I never said that? But when i’n searching for a certain answer, I don’t really care about paragraphs, posts, etc. A good chunk of time it might be data auto-generated from big databases too.

Time and place for every tech/use case. When I’m in the mood for reading long pieces, i do that deliberately.

Very rare for me to use Google nowadays for something niche. For those cases, I would assume it requires multiple sources to get a definitive answer, so it goes to ChatGPT.

> I know many people are increasingly frustrated with Google

I’m not sure if this holds true nowadays. I’d say it git pretty bad, but now Gemini summary pretty much covers all of my use cases. And for any meaningful query, it’s straight to ChatGPT/Claude.

Different metrics. You measure how many users you have, whether they’re closing the app and etc. when it comes to consumer level apps. ChatGPT/Claude is heavily used, not a problem for an average user, so no need to improve that.

Kernels, performance-required stacks etc. is different, so you optimise them for that.

> It seems very unlikely they have $3-6B lying around ready to repay customer deposits

For Binance? I’d actually say very likely. They’d need to move a couple of papers around, but should be easy.

Sorry, this might be a thing in your org, and I have no idea hoe I can convince you, but it’s just not true in our environments. Maybe you might have a very non-standard set up, but frontiers do amazing job.

I’m not sure if you’re running the latest frontier models. Every single engineer (and myself included!) has mentioned numerous times how nothing beats AI in debugging. My other guess would be that you have non-standard set up that Opus 5.5 doesn’t know off the bat how to debug.

I wouldn’t say the same a year ago, but usually we get preliminary reports during incidents before someone can even auth to debug. Almost all the time, nowadays, the report is correct. We obviously verify it.

Let’s say you have an incident, and you need a resolution. Do you send out the AI for the hunt and fix it within 5 minutes, or spend 30 minutes going around, figuring out the problem yourself?

I’m on the camp of using agents as much as you can, but make sure to understand what they did. However, it will still atrophy your skills, no matter what. It’s just different times now, as on the job you’re expected to deliver faster. And obviously that means less time spending on your craft, honing your skills and etc.

At some point it clicked to me that crypto people don’t really care about the utility. Main purpose is creating new players in the world with immense amounts of power. In the end, it did produce a lot of rich people out of thin air, and I guess, served its purpose.

I’ve been interviewing a lot of candidates recently, and we have open book policy with AI usage. From what I’ve witnessed how people use AI, wrt efficiency, there seems to be some skill.

> codebase becomes too broken, messy, and bloated to run or manage

Argument is, it doesn’t matter if we can ship features and maintain them.

Imo, the problem is fragmentation, as everyone doesn’t watch the same tv/movies anymore. We don’t have shared experiences of watching Lost, Breaking Bad, Game of Thrones and etc., which used to be a global phenomenon. They do pump out quality TV (e.g. some Apple originals are top notch), it’s just it’s not like that golden era anymore.

Kinda agreed with you though, but I’m also getting old, so maybe my judgement is skewed as well.

There’s a very specific South Park episode that relates to this comment - “ You're Getting Old" (Season 15, Episode 7).

Not true. If you look at any “cheap entertainment” dashboard, you’ll see usage has been going up.

My hypothesis is, YT, Reels and TT are eating everyone’s lunch, as people spend more time on those. And for people who don’t watch much, there’s Spotify. Basically attention span degradation across all age groups.

Enterprise sales somehow convinced a lot of orgs that all AI is the same or operates on the same level. Even frontier labs have differences, and you have to use each for your own advantage.

We have had problems with people creating Apple Relay emails, registering, then deactivating their emails, then complaining that it doesn’t works. An average person doesn’t have the tech intuition that we have, so we have to work with them.

There's a neat blog write up by @patio11 - https://www.bitsaboutmoney.com/archive/optimal-amount-of-fra.... It applies to this as well, but in a larger scale.

Defending is so much harder than attacking. You don't even have to focus on one target, as you can just mass-AI-script a lot of attacks wherever it's possible, and see what sticks. Even if you did everything right, there's always the social-engineering angle, or a single compromised device, or just accidental bad upgrade through supply chain. If you can mass-spread the seed, and plant something that "calls home" constantly, you can just have an agent that would start the attack the minute it gets infected, and go deeper.

It's really a wild west right now, have no idea how we fix it. I'm assuming everything will get more concentrated in big co's hands, as having a very robust security is extremely hard.

Also, we have to agree that younger people will start companies. There will be small companies that cannot afford red teams. There will be companies that are just not equipped, and their profit margins cannot sustain constant attacks.

Anything that adds friction is a big no-go in any real commerce world.

What's the "other way" that would be better in general?

Yeah, fair point. I guess, my reasoning was pointed towards “AI will resolve the issue, so I don’t care that much about it”.

No, because expert programmers will not know what they’re building, and skew the prompts and ask for the right things.

The analogy would be math people with AI vs. non-math people with AI trying to solve some problems. I cannot solve the problems because I have no idea how to prompt it to go specific way. But it doesn’t matter how dirty your scratchpad was, because you got to the solution.

> I don't see it happening on projects that have a minimum bar for security/reliability and long-term maintenance

I don’t get this, because this is currently, right now, at this second, happening across huge projects. My point was that maintainability is less of an issue than before. Also, frankly, over years, I’ve seen bad code that AI wouldn’t even write. So there can be a case that the maintainability/code quality isn’t even going down as the models get better.

>How do you get the greatest quantity of maintainable, high quality code within a given token budget?

Doesn't matter as much. The cost of "maintenance" is going down, from what I can see internally, from other companies, and every story i'm reading from other organizations as well. Token budgets are changing, but model quality and effectiveness are going up as well.

> We've also been hearing for years that LLMs are going to automate away software development entirely, but software job postings are steadily rising.

I think this has been contested from day 1. SE jobs might be automated, but jobs might increase, if there's possibility of starting new companies / new investments / etc.