Not a lot going on lately (mostly centered on work these days), but still a couple cool links and some shocking news. Let’s start:
Watering your watch?
Cute study that really made me think. A groupp of researchers designed simple wearables that, instead of being just made of electronics, required a living organism to properly function (a slime mold). If the user does not feed the mold, the device just stops working.
This seems to generate a link between the user and the device (people even started calling the mold “my little pet mold friend”), which some think could be a way for reducing electronic waste. Of course, I agree with the fact that users might be more careful about their gadgets this way, and maybe even consider not replacing them so fast or just throwing them away after a few uses. However, I have always considered gadgets as a tool which would be ideally invisible, without any attachment on my side. I just want things to work seamlessly, without bugs or failure, and of course for as long as possible. Add to that the option for repairing, which we allowed companies to abandon in the last two decades, and that would make a perfect device to me.
What If Your Day-To-Day Devices Were Alive?, on Hackaday
The week on AI
Almost ran out of headings for AI-related news, as it seems that everyday I see something that I find worth sharing, so I will resort to naming them with a default AI ~weekly recap (I know, a weekly recap inside a weekly recap, get over it).
This week the people at Facebook published a preprint on BioRxiv describing a new model that was used to predict the shape of +600 million proteins. What I found super interesting is the fact that the technique is based on the use of a large language model, which can fill in the gaps in amino acid chains. Then they trained an algorithm to infer protein structure from its amino acid sequences, similar to what the people at AlphaFold did before (in that case, they were able to predict +200 million proteins).
Large language models are also being used to help with more mundane tasks, by the way. I found this article talking about how researchers have started using them to help write scientific papers. It seems that they input some general ideas or a rough draft of the information they want to put on the manuscript, and they mingle with the results of the model until they get a text they like. Not sure about how I feel about it: is it cheating? or is it a wonderful tool for non-native English speakers to write good and accessible texts? On a side note, there are other programs that help researchers find articles or summarize them to speed discovery. I hope those spread soon to the whole community so I can stop reading things like “to the best of our knowledge, this is the first time…” just before seeing someone showing literally the same results others have published before.
AlphaFold’s new rival? Meta AI predicts shape of 600 million proteins, on Nature
Could AI help you to write your next paper?, on Nature
This one hit hard because I was not expecting it at all. Nibel (formerly nibellion@twitter) just stopped covering video game news on Twitter and erased his account. The reasons he gave are multiple. First and foremost, it seems that he never really managed to earn a fair living for its work. This is very shocking, because Nibel was one of the top journalists on the platform, airing news faster than most giant media outlets. For an industry that generates billions on revenue each year, being among the best journalists out there should assure you more than enough to make a living.
On his farewell letter, he also stated that Twitter as a platform never really worked for him. He never managed to get verified (though now it seems that everyone with 8 bucks will be able to do it), which caused many people to make alt-accounts impersonating him and spreading fake news (being a journalist, that sucks big time). He tried to build a Patreon, but all those moves always seem fruitless to me. If your audience is on Twitter, you need to monetize your content there, as the chances of someone going to a different platform to give you money are close to zero (moreso when you offer your content for free).