certificate down again

[dobkeratops]

the certificate for the site is down again,
curious if you still want to maintain this or not .. I realise you pay for hosting of this.

I think it's still a handy resource.. fine-tuning and testing vision models, fine tuning image generators. .. and keeping contributions to an open training corpus in case the companies try to close everything up.
I've been messing with VLM's and image generators myself quite a bit lately.
I could imagine an evolved workflow where one provides corrections to AI generated captions.

[bbernhard]

Thanks for letting me know - I've fixed the issue ~2 weeks ago, but forgot to let you know. Sorry!

I also have bad news for you. I got a mail from Hetzner (the company where the server is rented) and they are raising the prices up to 30-40% beginning of April. I always allocated ~100 EUR/month to keep the server up and running, but with the price increase it exceeds the allocated budget for this hobby project by much :/. Since a rented server at Hetzner was already the cheapest option out there back then (way cheaper than renting a cloud instance), I think there is nothing cheaper out there to move to.

Reflecting on the recent changes in AI (OpenClaw, Claude, ChatGPT, etc) and how it is used by big companies also made me realize that I kinda lost the interest in the technology (for now). While I do think that there might be valid use cases out there for AI (in the future), I have the feeling that every company out there at the moment tries to slap the AI label on its product just to ride the wave (reminds me a bit about the early days of Bitcoin, where everybody wanted to solve problems with a blockchain).

Personally I do feel the impact of AI - however, right now mostly the negative aspects of it. RAM and HDD prices are skyrocketing, electricity is costing more and more, some excellent senior developer friends got laid off and replaced by Junior devs with AI access and no experience in building sustainable software, Open Source projects are flooded with low quality AI generated pull requests, etc.

Of course, not all of those problems are entirely the AI's fault, but I think its a driving factor. I sincerely think (hope), that things are getting better in the future and that the hype wears off and we end up with a more realistic perception of what AI is capable off and what's just not possible.

I hoped that ImageMonkey will still be still around until then, but with the price increases that will come in April I think that's not possible in the long run. I'll aim to keep the service running until at least summer and latest until the end of the year. Should things improve and prices come down, I am definitely revisiting the decision. Of course I'll create a backup of the data and keep the domain registered - so if things are improving there might be a reboot. Sorry - I wished I had better news!

[dobkeratops]

Right makes sense . I'm enthusiastic about AI itself (the breakthrough of LLMs and diffusion models) - but it is being abused , the massive datacentre buildout is probably going too far and rushing to replace human input will backfire.

I see things like imagemonkey (curated open datasets) as essential for keeping 'grass-roots aligned' AI . labelled data can cut down the actual amount of data needed to train. And we can use this kind of data to train LoRAs on the smaller nets (e.g. google gemma3-4b-q4 will run on laptops phones etc and it can do pretty good image descriptions .. you could have a bunch of LoRAs on the device to adapt it with specialised knowledge. An extention of the original Imagemonkey idea where you could automatically spit out a fine-tune of the VGG net would be ideal.

There's a real risk right now of the open internet (and even open computing ecosystem) dying out .. if most users just go to ChatGPT, Grok etc for everything .. write all their code through Claude (then opensource dies out) .. so I'm continuing to try and help with efforts like this.. keep the open ecosystem alive instead of just taking things for granted.

I figured eventually you might give up on hosting this with just one user.. if I grab a refresh of the data then I can keep it around.

I do also have a little VPS myself now but dont run anything serious on it (I dont even have a domain name) . I use it to host a little web-based 3d engine & game and for sharing data.. i'll throw a mirror of this dataset there.
thanks for having kept this going so long.

[bbernhard]

There's a real risk right now of the open internet (and even open computing ecosystem) dying out .. if most users just go to ChatGPT, Grok etc for everything .. write all their code through Claude (then opensource dies out) .. so I'm continuing to try and help with efforts like this.. keep the open ecosystem alive instead of just taking things for granted.

Totally agreed! I really hope that this huge AI push won't kill the open source ecosystem by flooding the scene with sloppy developed programs and purposely hiding backdoors in some innocently looking code. I think this could become a big problem and supply chain attacks will definitely be on the rise in the future :/

I figured eventually you might give up on hosting this with just one user.. if I grab a refresh of the data then I can keep it around.

I do also have a little VPS myself now but dont run anything serious on it (I dont even have a domain name) . I use it to host a little web-based 3d engine & game and for sharing data.. i'll throw a mirror of this dataset there.
thanks for having kept this going so long.

That would be really nice!

Currently there are about 11k images which aren't unlocked yet. I think that's do-able for me to go through them in the next couple of months to either unlock them or delete them. I really want to go all of those images manually, as some people in the past uploaded inappropriate pictures and if I make the data publicly available, I want to be sure that there's nothing offending in them. Once that is done I am planning to upload a backup of the dataset to archive.org. Of course, I could also sync the data to your server too :)