The trouble with hiring juniors now is it's much more difficult to get them up to speed so they can be productive. Before covid, you'd sit next to them, get asked questions every so often, do some pair programming, and discuss ideas over lunch. You can, on paper, do the same exact things over Slack and Zoom. But there's much more friction. And a junior that's struggling is a lot less visible than it used to be. So what ends up happening is seniors become more heads down, getting things done, and juniors struggle to get time with more experienced coworkers.

Fully agree. I'm all for remote work. However, in my first 2 years of programming, being able to go the the office, put my laptop and notebook down next to a senior dev, point and say, "Help me," was so valuable.

GIMP should take a lesson from Blender. Blender used to be the most clunky, unintuitive pieces of open source software. But after a decade and a half of UI development, it's one of the smoothest interfaces you'll ever use.

> GIMP should take a lesson from Blender. Blender used to be the most clunky, unintuitive pieces of open source software. But after a decade and a half of UI development, it's one of the smoothest interfaces you'll ever use

Agreed. I'm already willing to use GIMP in its current state. But though I've used it since I was a child, I have to re-google for things I know it can do.

I had a photo of a barn. I was going to construct it in miniature, so to get scale measurements I wanted an isometric perspective from a photo that had been taken at an angle. I had done this in GIMP before so I was hesitant to start googling for answers but in 25 minutes of playing with it, no combination of inputs would do what I wanted. I had to find some youtube tutorials.

Even simple tasks aren't simple. Annotating a photograph with a couple red arrows is a multi step challenge involving paths, stroking, selections, layers, and maybe some other stuff I'm forgetting. These UI concepts were impenetrable without tutorials -- I never would have figured this out on my own.

GIMP has helped me but it's never been pleasant to use.

For what it's worth, there is a built-in arrow brush in GIMP 3.2 (we updated the MyPaint brushes engine and got it for free as a result).

You can draw arrows with the pencil or paintbrush.

Not if you're a professional in a big boy company, where the MacOS users can literally draw a perfect arrow in 2 seconds on any image, without installing a thing. You're just going to look completely incompetent.

Another lesson GIMP could take from Blender is the importance of a python REPL that updates with the analog commands to the menu items you're clicking. GIMP allows scripting from some kind of lisp, but in many hours of attempt I've never gotten it to do what I know it can do. But I was effective in blender within seconds because it allowed me to directly translate my actions to code.

GIMP has a Python console, and several built-in plug-ins are written in Python: https://docs.gimp.org/3.2/en/gimp-filters-python-fu.html

We want to make macros simpler (some of the work I helped out with for GIMP 3.0 was to lay the groundwork for automated/recordable operations), but as with all the things on our roadmap it takes time and developers.

I used the python API in GIMP headless early on in my career, was pretty handy.

Blender also had a comparatively large amount of money backing these changes, if I remember correctly, which I expect GIMP does not. I suppose a lesson in that area would be required first. However, the others like Krita might be better positioned for this.

the only possible lesson from blender I see is "receive millions of dollars a year from corporations"

It's a bit odd that they're not comparing it against Sonnet

I don't think so. They're comparing it to the highest tier available models from Anthropic and OpenAI. Generally speaking, Opus is better than Sonnet in almost every way, so why have the redundancy?

Price to performance?

I think their comparison to how their benchmarks compare to Opus are a great way to show "look at similar benchmarks for a fraction of the cost". If it has Opus benchmarks (I don't actually take benchmarks seriously, but for their comparison purposes) and Sonnet is still more than half the price of Opus, I figure it's close enough where it doesn't matter.

The tweet specifies that the new model is geared towards long-running tasks, which is what you'd use a model like Opus for anyway.

CS Academia tends to lag behind industry practices. The research frontier can be very cutting edge, but course curriculum, assignments, and institutional norms are slower and more conservative. That’s usually manageable when the shift is something like cloud adoption, new tooling, or a new dominant programming language. But this particular industry trend, use of AI in software development, is massive and fast moving (especially the agentic workflow growth over the last 6 months). And we're just now understanding where everything fits in and its limitations.

Journal articles are sometimes years behind. There are still papers coming out that use GPT-3.5 (!) for their main result. These days I'm basically only reading arXiv preprints (and whatever is trending on GitHub).

Could've just called it MacPad++ or something

Broccoli has been around since the 6th century BC

This is just geopolitics. You should've seen what the US and Europe did during the Cold War.

Sure, you could argue it's like writing code that gets optimized by the compiler for whatever CPU architecture you're using. But the main difference between layers of abstraction and agentic development is the "fuzzyness" of it. It's not deterministic. It's a lot more like managing a person.