1. The default settings aren’t well optimized. So the pattern a lot of people fall into is: start a project and everything’s fast, then add more code and dependencies and everything gets super slow. Then you Google how to fix it, change some compile settings and restructure your project and it speeds up again. Rust requires knowledge and effort to keep compile times reasonable - it’s not set up that way by default.
2. Rust Analyzer / cargo check is inefficient and throws a lot of data away each time instead of caching it. So in a larger project you’re waiting on it to catch up. Waiting on Rust Analyzer to check my code is easily the largest amount of time wasted, more so than actually compiling it.
> This Rust rewrite would've taken a team of engineers with full-context on the codebase a year of work. With 1 engineer using Fable & closely monitoring Claude Code, we went from start to 100% of the test suite passing on all platforms in 11 days.
This is impressive from a technological standpoint, but it does gloss over the fact that it would have cost $165k in tokens were Bun not part of Anthropic.
The comparison here isn’t completely fair - it would take a small team a year to port it if they spent $0 extra on it.
I’d be interested to see a comparison between spending $165k in 11 days on Claude vs splitting that between 50 people over 11 days for a line-by-line rewrite of the Zig code. I suspect Claude might be faster and therefore cheaper, but maybe not by a lot.
The article focused on image editing but it’s happening on the video side too.
The free version of DaVinci Resolve is extremely capable and the Studio upgrade is a very reasonable one-time purchase. I really don’t know how Blackmagic makes money from it - it may be the case that they’re subsidizing it with hardware purchases.
A Mac Studio is a much better buy in terms of memory bandwidth, but impossible to buy in a 128 GB configuration. Honestly there aren’t great options right now and it’s probably better to wait for the market to be less insane.
What an absolutely terrifying situation for that woman. It reminded me of a story from the UK where a police officer pulled over a woman, arrested her (for no reason) and raped and murdered her. I don’t know what you’re supposed to do in that situation because the law says you’re supposed to submit to police officers, not resist arrest, etc.
I think the people who do this fall into two categories:
1. The genuinely naive. This is new technology and the social expectations haven’t been figured out yet. Some people still think this is helpful / useful.
2. People who don’t care as long as they generate traffic to their site. AI-written articles are the new zero-effort promotion and SEO strategy.
I’ve been locking my phone in a timed lockbox for the past few weeks and it has been transformative. There’s something psychological about knowing my phone isn’t available for the next 1-2 hrs and I don’t need to think about it. I think it’s better than Brick because it can’t be bypassed.
I have been considerably more focused and productive as a result. The problem I had is that my phone controls a lot of other stuff that I might need while it’s locked away (e.g. my smart lights), but I am working on getting single-purpose physical equivalent devices, and maybe a burner phone with no apps or web browser installed for everything else.
Well the company statement is that they say they can do this in their terms of service. It seems very plausible that Meta is doing what their TOS says they’re allowed to do.
Why are there only three major memory companies? Because memory is a really hard market for new entrants.
Firstly, IP is heavily patented. CXMT only got around this by using a different manufacturing strategy they bought the patent for from another company.
Secondly, scale and yield is really hard. CXMT’s production costs per unit are higher than the established companies because they’re still struggling with yield. They only made a profit last year because memory is massively overpriced. Most startups do not have the runway to spend a decade improving their yield. CXMT only survived because the Chinese government had an incentive to bankroll them long-term.
Companies like Apple and NVIDIA could get into the memory game for their own needs if they wanted to, but it would take a long time and they’re only incentivized to do so if they think memory will be high forever. If it’s only high for the next 2-3 years it’s not worth it.
Taste is relative - what’s tasteful to one person may be distasteful to another. It may differ among cultures too.
I think it’s wrong to say AI doesn’t have taste; it’s more accurate to say it doesn’t match your own taste and it can be hard to get it to match because it’s intangible.
This is more evident with Claude Design than Claude Code, where it does incredibly dull and uninspiring designs and then writes little puff pieces about how “bold” and “edgy” they are. For some situations that is the correct design, but I found it difficult to get better results with additional prompting.
Sure, but this administration doesn’t behave logically and big AI companies are already pushing the “AI is dangerous and only we should be trusted to wield it” angle.
I believe the open model party will eventually end. Perhaps because companies realize it’s too much of a commercial advantage, countries don’t want to give other countries commercial or military help, or maybe even an outright ban after someone uses an open model to guide them through how to make a bomb.
> The numbers are great: 194 impressions, 6 clicks, but zero conversions.
You can expect only a small percentage of people who click to subscribe. A sample size of 6 is too small to draw any conclusions. I would say that you have not paid enough money for advertising to know if it works or not.
But the reason people say to use your network isn’t really for finding customers - it’s to get feedback from potential customers. That is way more valuable than any money they might give you. It’s pretty uncommon to have instant product-market fit. A lot of startups have to make significant changes to the product or pivot entirely to become successful.
One of the things I learned from real customers is that not everyone uses a computer the same way I do, and things that I assumed everyone did are not as common as I thought. It changed how I design software and such a lesson is difficult to learn when you’re just in your own bubble.
A lot of providers have basic DNS for free but charge for more sophisticated uses like geo routing. AWS charges for every single thing:
https://aws.amazon.com/route53/pricing/
I refuse to use Cloudflare for two reasons:
1) It’s free until it’s not and they don’t tell you where that limit is.
2) If something goes wrong you cannot contact anyone unless you have an enterprise agreement. Just paying them money is not enough to be able to speak to a person. I had serious issues with R2 and had no-one to contact. They have a Discord and you see people posting the same issues over and over again that never get fixed.
If my comment is more than a few sentences, I’ll often type it up in my Notes app and paste it in. It means I don’t lose it if there’s some kind of technical issue, and I don’t run the risk of accidentally fat-fingering the Submit or Reload buttons. I do it with text messages too.
1. The default settings aren’t well optimized. So the pattern a lot of people fall into is: start a project and everything’s fast, then add more code and dependencies and everything gets super slow. Then you Google how to fix it, change some compile settings and restructure your project and it speeds up again. Rust requires knowledge and effort to keep compile times reasonable - it’s not set up that way by default.
2. Rust Analyzer / cargo check is inefficient and throws a lot of data away each time instead of caching it. So in a larger project you’re waiting on it to catch up. Waiting on Rust Analyzer to check my code is easily the largest amount of time wasted, more so than actually compiling it.
But it should be said that this is being worked on: https://rust-lang.github.io/rust-project-goals/2025h2/relink...