I mean, in comparison with openclaw, etc. capabilities are ofc more restricted. However I don't want to accidentally delete my entire photo album, so I do understand the direction by delivering useful, but somewhat obvious features.
Such a great project that could automate a lot vibes testing hopefully! A pity that the dataset only contains 55 questions. I'd like to see this number in the thousands.
I was going to ask, are all other variables accounted for? Are we really comparing apples to apples here? Still worth doing obviously, as it serves a good e2e evaluations, just for curiosity's sake.
Assuming a large contributing factor is all the coal plants now running to sustain Germany's independence from nuclear? Berlin's air quality has also tanked a lot since the energy crisis started.
Slightly off topic, but now that long context machine translation is roughly on-par with humans: are there any official efforts from Wikipedia, to translate the "best" or "most complete" language version of each article to all other languages? I'd imagine that the effort of getting all languages up to the same standards are just an impossible one and people from "lower-resource" languages would benefit a lot.
>It's just too fiddly, requiring way more "IT people" running around configuring Samba shares and printer drivers.
IIRC several German states went with nextcloud to make the transition a bit smoother. No idea about the effort when it comes to print servers, but on the other hand Denmark is quite digitized at this point and I would imagine printing papers is less of a thing there than in Germany.
>And LibreOffice is many many years behind MS Office, and it'll continue to be that way.
I believe you, but could you elaborate on what's missing?
I can't even imagine the amount of submissions coming in the next years, reviewers will have to start using AI tools as well. In the end (computer) science is just AI reviewers filtering AI submissions.
Kudos to Hugging face for going the extra mile on open sourcing knowledge. They could have "just" created a hub for model downloads, but the fact that they keep publishing top-tier tutorials, materials, experiments and code that benefits the whole community is such a blessing. IMO they're one of the most important players advancing the open source community effort. Keep it up!
Fun and potentially useful project, love it! When I tried it though, it was quite often hard to see whether the bounding box is "really" correct, as it hides what's underneath. Maybe some slight opaqueness could help.
Also, my first image had no bounding box at all. Being met by "Swipe right if the red shape is correctly outlining a building. If not swipe left", it felt like the wording or the UX could be improved by filtering for images that are guaranteed to have such a box.
I'd challenge some of these criticisms and give my 2c on this.
I've spent the last 6 months working on a rather complex chat with routes, agents, bells and whistles sort of system. Initially, time to POC was short, so I picked it to get quick at my feet. Eventually, I thought. The code base isn't enormous, I can easily rewrite it, but I'd like to see what people mean with "abstraction limiting progress" kind of statements. I've now kept building this project for another 6 months and I must say the more I work with it and understand its philosophy.
It's not that complicated. The philosophy is just different from many other python projects. The LCEL pipes for example is a really nice way to think of modularity. Want to switch out one model for another? Well just import another model and replace the old. Want to parse it more strictly, exchange the parser. The fact that everything is an instance of `RunnableSerializable` is a really convenient way of making things truly modular. Want to test your pipe syncronously? Easy just use `.stream()` instead of `.astream()` and get on with it.
I think my biggest hurdle was understanding how to debug and pipe components, but once I got familiarized with it, I must say it made me grow as a python dev and appreciate the structure and thought behind it.
Where complexity arise is when you have a multi-step setup, some sync and some async. I've had to break some of these steps up in code, but otherwise it gives me tons of flexibility to pick and chose components.
My only real complaint would be lack of documentation and outdated documentation, I'm hardly the only one, but it really is frustrating sometimes to understand what some niche module can and cannot do.
Shout out to another project by the same creator: https://www.derivative-calculator.net/
Both of these helped me through my undergrad physics degree, especially because of the step-by-step solutions. Without it I wouldn't have learned nearly as much about calculus.
On the other hand, I read stories online where people claim that it's hard/impossible to get promoted as remote employees. I wonder if these constraints will loosen as well.
In Germany: this was my experience late last fall as well. It's not that I got no offers, it's just that the ones I got (2) were at least 30% below what I had asked for. It's definitely employer's market right now.