The subject of the week seems to be the negative impacts general-purpose AI has on energy sustainability in the digital world. But hey, it's as good a day as any to realise you've been killing penguins every time you generated an image. Anyways, enjoy another accumulation of links we found useful and/or interesting this last week.
- In this study, researchers found that general-purpose AI models like GTP-4 etc are “orders of magnitude” more power-hungry than purpose-made models, such as Google Translate. And the numbers sound horrific. Generating one single image from the least power efficient model has an equivalent C02 footprint as an average car driving 6,5 km. Meanwhile, there is an increased focus on energy efficient web applications, and it's a little disheartening to think that all the good work that goes on these days is devoured by gen-AI. Via The Verge.
- Poor Charlie's Almanack is an exploration into Charles T. Munger, the late Berkshire Hathaway co-founder. If the story isn't enough for you, the presentation, design and overall experience is very well made.
- Here's a film about the dangers of AI, 90% generated by AI. The film discusses copyright, what'll happen to artists, etc.
Tech of the week
- Almost feel bad about posting this, in lieu of the research study above, so proceed to this link with a little self restraint will you. Visual Electric is a visual canvas based AI image generator that makes the whole thing more accessible to designers, with sliders instead of prompts etc. Which is funny, designers are users too, so let's file it under Accessibility.
- Meta AI wants to mingle with the giants of gen ai as well. Luckily, their service demands login, allowing you to pay for your silly prompts with your personal details. Sorry, that first link today kinda set the tone.
- Check out this magic animate thingy, figure it out for yourself via GitHub.
- There's obviously been a lot of fuzz around Gemini, Google's DeepMind take a bazooka to the sword fight that is generative AI models. All presented neatly here. It certainly has some impressive use cases, but quickly came under a barrage of criticism that you can read about here.