Hello gorgeous readers. I hope your pandemic Christmas didn't completely collapse under the putrid shit-storm that the layman refers to as Tier 4.
Hey 2020, got a gif that I thought you might like...
Apologies for breaking format and not giving the week its usual rating; over this Christmas period I discovered that the forward march of linear time is a lie, and the inability to take a holiday all year has macerated my brain — I have absorbed very little news, and done hardly any thinking. God it's been... it's been amazing.
But anyway, here's some stuff that I picked up since the last iteration of Horrific/Terrific:
- TL;DR — but evil.
- 4-bit deep learning
- A (very) quick look back at the best of 2020. I know this is a tech blog and therefore I should probably be thinking about the future. Sorry, I prefer to dwell on the past x
💾 The future has less bits in it
Get it? Bits? It's a computery joke! Anyway, in a previous Horrific/Terrific I did a super-summary of the research paper that Google fired Timnit Gebru over. She highlighted four key risks of machine learning, and one of them was the carbon output of training large data sets.
So that means machine learning on a large scale is bad for the environment — the environment, that thing we should be protecting (because we live in it). How to fix this? Perhaps build training models that require less data? No! Just use less bits.
Some scientists have written a paper on how you could train neural networks on 4-bit hardware, as opposed to the usual 16-bit. Why is this good?
- 4-bit deep learning cuts the energy cost over seven fold
- It's better for privacy: it can allow big tech companies to train their powerful models on many, smaller devices (like your phone) instead of sucking it all into their central servers and doing it on their end. So, improve the AI on your phone, while also keeping the data on your phone — it's win-win.
- If you need less computing power to do deep learning, it suddenly becomes accessible to people and entities which are not a big tech firm...
Can't wait for the low-fidelity 4-bit future that awaits us!
📖 Reading?? YUCK. How can I do less of it?
I hate reading just as much as the next millennial, which is weird because I am a 'writer'. Mark Zuckerberg, our supreme trans-human leader, has therefore decided that Facebook should develop a tool (called TL;DR) which uses AI to summarise articles that are simply TOO LONG to read with your inferior human eyes. As we all know, AI is the ultimate solution to all problems — even problems that aren't real.
Seeing as tech blogs in the new year are all about predictions, here's mine: if the goal is to misinform and offend people even faster, then this is going to work perfectly 👍
If you're convinced that this is a good idea and it will all work out okay, I first ask if you would like to swap brains with me (I need a break, and I know a guy). Secondly, Facebook's track record with using AI to 'solve problems' is rubbish. Here's an example from — literally — the top of my head: They used it enforce their real-name policy, and 👇
- the machine didn't know that Native American names were real (you know because their not white names...)
- the people behind the machine didn't consider that some users have legitimate reasons for not using their real names.
- Wikipedia can tell you more, if you're interested.
Side note: Facebook seem to be building TL;DR all by themselves, instead of acquiring a company who have already built it. God, it's like watching a baby dear stand up for the first time.
🔭 In 2019, 2020 was the future
But now, of course, it's in the past... and in the bin where it belongs. Here are some futury things that actually happened in 2020:
- A child was blessed with the name 'X Æ A-12' — the lucky devil
- Space Force! A) continued to exist, and B) they have a horse (Space Horse)
- Amazon, overlords of dystopia, debuted a house drone, the most useless and depressing piece of hardware I've ever seen — and I'm including this 'smart' egg sheath in my summation.
- Deepfakes were finally used for something other than revenge porn. The movie Welcome To Chechnya used deepfakes to protect the identities of LGBTQ folk.
- A pandemic brought the world to its feet. Palantir deviously leveraged this by winning several government contracts, one of which is to build a platform that will algorithmically determine who gets the vaccine in the US. Thank you, Palantir for automating inequality, it really wasn't working as efficiently as it could 👏
Just FYI I'm trying to write a novel of speculative fiction right now so if reality could give it a fucking rest and let me catch up that would be great...
Drowning in Tiers: Finally, if you're a UK citizen you may have noticed that our Grand High Oaf, the prime minister, ruined Christmas with a covid-19 tier restriction that no one had ever heard of before: Tier 4. All these new tiers are confusing, so I put together this handy and completely serious guide to explain all future and past tiers — enjoy. My personal favourite is Tier Starmer.