I’ve found that sub attracts a very wide range of devs who all think their technology is the best, and that drives a lot of hate for web technologies because a lot of mindsets are stuck in the 90s and 00s. It really is quite backwards around there

Thing is I haven’t found much of an AI community to absorb myself in over here which is why I’ve been back there in recent months, but it feels like it’s impossible to get any reach for my own content there now

The updated UI framework does feel like the biggest step taken in a few years. Liquid glass isn’t even the most interesting part of iOS 26

What toppings are the best on pizza and why are they Ham and Pineapple?

Now we just need OSS review bots which remove clearly bot generated contributions to complete the cycle

Depending on country there will likely be salary benefits to having a degree or masters. I went straight to masters as a self taught dev (with career experience and unrelated Bachelors) because the UK’s R&D tax rebate made me less valuable without it.

Formal qualifications can break glass ceilings

It’s okay I can spare 4.5 seconds 👹

Similar applies to any ML, patterns will be detected during training and form a huge number of smaller highly abstract functions which combine to produce something more concrete

One of the best intuitions you can look at is visualisations of how Convolutional Neural Nets build up their abstractions, there’s an image here: stackoverflow.com/questions/38...

Combined with an understanding of the layering of maths it’s pretty powerful to see this IMO

It’s notable to me that cognition is talking about MAs which mutate a global system, while anthropic is talking about MAs which query a global system.

The former is obviously harder because the outcome is more cross-dependent and so requires constant cross-talk to produce a consistent result

And all of this leads to BSky feeling less sticky or worth opening. To compete with X they need to get really good with social algorithms and that hasn’t started to materialise yet

It also hurts smaller accounts, a good post should be able to go viral but only big accounts can do that right now

Fair enough! I’m happy to leave a suggestion to that effect then ☺️

Looks really good, I suppose since it's a new function the key can be required out the box and start to dissuade from footguns. Then TQ@6 could make that property required everywhere?

Yep, I _think_ it would work well. Should be a pretty easy union type to test out anyway

So long as the type adapts to whether a mutationKey was set I don’t see a need for enforcing it when it used to be optional.

Libraries like tRPC can ensure one is always set and user-land code gets mutationKey:undefined until it’s set so are protected from misuse

I think that’s just the typical lifecycle of a Haskell system in a business, I’ve heard it more than once

* one developer in a team gets really FP-pilled
* rest of team spends future years scared of system
* system is rewritten

Love the idea of just having a JSON framework around this concept, but it also really changed the way I think about RSCs in about 2 minutes

Great post!

Homebrew is the worst package manager out there. The way they handle the universe is very poor for both performance and compatibility.

Was trying to avoid upgrading our MacOS and Xcode to latest for a few weeks some years back, and homebrew pushed a bunch of changes that broke Xcode… why

Midnight Generation’s album is stunning, here’s the single that caught my ear:

youtu.be/2dMRxv3lbXg?...