AI/ML
AI and machine learning news
Dear readers and customers,
it’s an issue of our time and can be observed from AI bots to politics: sycophancy.
But what is it and why is it here?
First of all, the complexity of our times are so exceptionally high we probably won’t be able to find a single root cause here. All we know is it’s here, it’s a problem which needs to be addressed and there’s not really much we can do individually. What’s important here is to develop a kind of awareness. As well as keeping it in mind on a daily basis, e.g. when talking to colleagues and friends, consuming media, etc.
In short, what it is:
“Sycophancy is insincere flattery given to gain advantage from a superior. A user of sycophancy is referred to as a sycophant or a “yes-man.” [via]
It may sound like a harmless negative character trait but unfortunately it isn’t. Keywords to take away here are “insincere”, “gain advantage from a superior” plus “yes-man”. We don’t live in times for “yes-men”, nothing outside currently is agreeable per se, it’s actually the other way round. The female component is missing here too. Furthermore, if someone or something is “insincere”, you can as well state it’s a joke. And neither AI topics nor politics are a joke or should be treated as such. This planet won’t exist 50-100 years from now on and there are polls which even state an exact date for Armaggeddon, which is in itself a remarkable statement of our times.
Whenever “dark patterns” arise, like seen with LLMs and AI models here, the process of eridacting a failure for the future failed. As such, we can’t turn back time and we wouldn’t advise even if the possibility existed. However, we can try to avoid it for the future. Be it for ourselves, AI bots, politics or whatever. The main problem here is people and individuals will be tricked and this is like gas lighting an entire society into oblivion. It will be a mind state and a world situation where human beings won’t be able to think about anything relevant anymore. If such a “dark pattern” already made it into a machine learning algorithm, we took the wrong turn somewhere in the past. With this post we want to sharpen our senses so that it might be the only dark pattern we found in AI bots five to ten years from now on.
Thank you for your help here and always keep in mind: we still hold it in our own hands. We can be the change, please spread this awareness and try to communicate it in a positive way. Humans are worth it. And always were.
Keep your head up and best wishes,
the aethyx staff
Dear readers and customers,
starting this month, we added a new feature to all our web projects: basically we are blocking all AI crawlers. Or at least we try to.
It’s not said that it works or the crawlers will respect our manually integrated rules. However, at least we tried our best.
We do this for two main reasons:
1) you should be in control of your posts and thus your data. If your individual posts are used for alteration, you should know beforehand. Currently this is not given with the methodologies machine learning tools are trained. These just use what they can find on the open web
2) if your work helps in any way for monetisation of individual companies, you should get your portion. Our idea is let’s be fair: 50/50. For every Euro earned with your hard work, you should get at least 50 Cents
Here is the current list of crawlers we try to block as of now:
| AI2Bot |
Explores sites for web content that is used to train open language models |
More Info |
|---|---|---|
| AmazonBot |
Used by Amazon’s Alexa AI to provide AI answers. |
More Info |
| AppleBot |
Used by Apple for generative AI features across Apple products, including Apple Intelligence, Services, and Developer Tools. |
More Info |
| Bytespider |
Used by TikTok for AI training. |
More Info |
| Cohere |
Used by Cohere to scrape data for AI training. |
More Info |
| ChatGPT |
Used by OpenAI to power ChatGPT. |
More Info |
| ClaudeBot and Claude-Web |
Used by Anthropic’s Claude. |
More Info |
| CommonCrawl |
Compiles datasets used to train AI models. |
More Info |
| Diffbot |
Used by Diffbot to scrape data for AI training. |
More Info |
| FacebookBot |
Used by Meta (Facebook) for their AI. |
More Info |
| Friendly Crawler |
Crawls websites to build datasets for machine learning experiments. |
More Info |
| Google Extended |
Used by Google to power Gemini (formerly known as Bard). |
More Info |
| ImagesiftBot |
Used by Hive’s Imagesift tool that scrapes images. This may be used for the company’s generative AI product. |
More Info |
| Kangaroo Bot |
Used to power the Australia-focused Kangaroo LLM. |
More Info |
| Meta-ExternalAgent / Meta-ExternalFetcher |
Used by Meta (Facebook) to train AI products. |
More Info |
| OAI-SearchBot |
Used by OpenAI for their SearchGPT product. |
More Info |
| Omgilibot |
Used by Omigili to scrape data for AI training. |
More Info |
| PerplexityBot |
Used by Perplexity for their AI products. |
More Info |
| Scrapy |
Blocks the Scrapy bot (used for scraping websites). |
More Info |
| SentiBot |
Blocks SentiOne’s AI-powered social media listening and analysis tools. |
More Info |
| Timpibot |
Used by Timpi; likely for their Wilson AI Product. |
More Info |
| Webzio |
Used by Webz.io for their social listening and intelligence platforms. |
More Info |
| Webzio-Extended |
Used by Webz.io for AI training. |
More Info |
| YouBot |
Used by You.com to train AI products. |
More Info |
If you are already a customer (thank you!), we activated it automatically on your website for free. There is no additional cost and there never will be.
If you want to join as a new happy customer, the feature is added automatically when we set up your site. The information is already up to date on the product overview page: https://aethyx.eu/eshop/.
Sorry for the inconvenience. In an ideal world this would never have happened. However we are far from ideal at the moment. Let’s look into the future as things can only get better from here.
Enjoy fall and best wishes,
the aethyx staff
Dear readers and customers,
by the end of this month, we are reaching the end of our very first year of AI art generation. Okay, technically it’s not AI but strict machine learning. But nobody gives a damn out there anyways.
So without further ado, let’s celebrate our most popular creations for the first time and tell you some details about them (click to enlarge):
#5 Atlantis
This is really one of the first ones we ever did, end of August 2022 with the open source tool Stable Diffusion on one of our laptops. The command how we generated it can even be seen publicly on the artwork detail page on Deviantart. We are proud we made it and it aquired 6 fans so far. (Buy it as an adoptable for 9$)
#4 8K Space Odyssey
We were experimenting with the built-in generator tool from Deviantart called “DreamUp”. The biggest difference from what we could create last year is the resolution: 1280x1280px instead of 512x512px here. We didn’t see much of a difference in details or quality. We guess “DreamUp” is built around Stable Diffusion itself as it’s freely available as an open source artwork generator tool. This space odyssey attracted 7 fans so far.
#3 Japanese concept cars in a Japanese cherry tree spring setting I
Another one created with “DeviantArt”s own tool. The prompt is there as well, 7 fans so far.
#2 Japanese concept cars in a Japanese cherry tree spring setting II
Except the cherry tree alley, we think this one of the ugliest ones we ever created, haha! Nevertheless, it’s almost our most popular, attracting 9 fans so far. Same as above, from our five most popular AI art creations, three are currently done by “DreamUp”.
#1 a Neuromancer quote
It honours us that the most popular AI art creation was done by us with our local machine! It’s the visualisation of an original Neuromancer quote: “His eyes were eggs of unstable crystal, vibrating with a frequency whose name was rain and the sound of trains, suddenly sprouting a humming forest of hair-fine glass spines” We and 8 others think this must have been a masterpiece for its prompt. And we won’t complain, as it’s also the only entry here with a comment (Thank you soulcreator789!).
Currently, we are testing out SDXL 1.0 on the exact same machine. Our first experiments look promising, although we want to be honest and admit that a hardware piece from 2019 with 6GB of RAM shows us our current limits very fast. We can’t do hi-res creations, as an example. And one creation with a minimum stepping of 30 iterations takes over 10 minutes for us. So it seems, before we can invest 3,000€-4,000€ to get that 4090 laptop on the market, we have to sell a gazillion of adoptables to really be able to keep up pace with the current industry. Yes, being indie is hard, no doubt about that.
We hope you enjoyed our little artwork excursion today and who knows, if we are still in the AI art creation space next year, we publish another AETHYX MEDIAE artwork ranking entry. 🙂
So long and keep counting those s/it,
your aethyx staff