This has good groove, but the thing i like most about it is the sonic space. Really attentive to where these sounds fit in stereo.
2 Likes
Ok, I give up ā what are the sets going into that diagram? People with connections to Naniwa Express, people doing smooth jazz, or people who Maxhirez listen to when coding? (Iād assume the latter, but Nick Beggs pops up in a lot of places so might be missing something).
4 Likes
Sorry , was a half doz . + beers into it and feeling a bit cynical yesterday . I know that real people [ coders ] put a lot of time / work into those algorithms to try and make life easier for everyone . 
2 Likes
my absolute favourite band, came out of Nervous Conditions, who were also brilliant. Someone sign them soon please!
3 Likes
Donāt overthink it, @fredrik. There are better things to use your brainpower on!
3 Likes
You know what they say about good intentions. At the start of the decade (2011) I got really into machine learning and was very keen on recommender systems, the programming glue that looks at what you like and harnesses the power of very large data sets to show you things other people like you also like. Only recently did I recognise that these very same systems have become a powerful terror weapon, used to radicalise large subgroups by creating an ecology of extremist material that finds its way to those most susceptible to it.
This is the worst timeline.
4 Likes
Who knew that leveraging linear algebra, calculus, statistics and even sometimes geometry would become a means to create shitty endogroups in society when powered by super fast machines?
Ok, people were worried about super intellegent ai with ai safety, and they should, but ai safety is way important even on smaller scales. Hell, I have been warning people about gpt2 and itās use with hostile bot networks sharing misinformation for years. People dont realize that a somewhat human like response that is close enough, already exists, at least when that human you are making is an online troll. I have made my own transformers that I used to carry half of a conversation with people and no one suspected a thing. These were friends that knew later that I was fucking with them.
I made some joke training with gpt2 that generated sonic the hedgehog fanfiction, episodes if sienfeld, Edgar Allen Poe poems, and writing based on james oyce. This shits advanced. I even made a transformer that learned to create a hybrid if Poe and Joyce that pretty faithfully joins both authorās voices.
I didnāt dare create a transformer that joined sonic fanfiction with James Joyce. Some things are just not meant to come I to the world.
5 Likes
Well, I read that gpt3 is out (which has 175 billion parameters?! ).
2 Likes
Yes, I never bothered updating my lil scripts to gpt3. It was good enough before. Itās curious though. Things like this are amazing and are a great new tech, and itās Pandoraās box, you canāt close it back up. We just arnt prepared for it as a society.
It will take a lot of harm and exploitation by powerful people before anyone notices, and by then, who knows if we can get any movement to change things once they have more power than ever.
Having worked with these for a while, I have developed a bit of a knack and can recognize them, but I wouldnt day itās completely reliable.
2 Likes
Iām a bit worried about the fact that these massive models could be biased in ways that are not easily detected and that they will be accepted in decision making processes.
4 Likes
Yeah, using a magic machine to 100% trust is not a good idea. When you have a neural network with 1000s of hyperparameters, itās hard to pinpoint if the training you did had effects on other hyperparameters that isnāt correct.
Even good ml models are considered good when they have a 30% error rate. If your error goes to low, itās probably just overfit and canāt generalize as much.
We have means of detecting errors (rmse, r2, gradient descent, etc), but bias is not exclusively error. Bias could mean the lack of information. Without that data, the learning would be correct from itās incomplete datasets perspective, but wildly incorrect if you go bigger picture.
Add to this, that error detection can really only work on a split of your training data, and you begin to realize how evaluation of these things can be limited.
3 Likes
yeah kinda scary that the bad science fiction shit is coming true. but only if we let it ā¦
2 Likes
Some early Georgio Moroder for you all.
Enjoy.
4 Likes
Thatās Southsea promenade! Just down the road from where I grew up. Some of you might recognize the area from The Who film, Tommy, thereās the pier in the background
4 Likes
Ah the 70ās! Get signed and see the world!
4 Likes
Chicory Tip! I recall in 1972 as a Keith Emerson fan sheepishly tuning in to Top of the Pops to watch Chicory Tip, in the hope of seeing their Moog. What I got, sadly, was the sight of these guys miming with their instruments, while somebody off to the side mimed all the keyboard parts on an electric piano. My generation was very snobby about the rock/pop distinction, even when the musicians were often serious rockers trying to keep their record label happy.
2 Likes
Awe, that makes me a little melancholy today. Iād love to see the land where my ancestors came from but travel makes me sick. The sewers of Her Majestyās Realm could probably do without my barf and other liquified ejections.
Well if you ever find a way, you should come! While thereās a lot to dislike about the U.K (the gov, 300 years of being a villain to almost every country and peoples in the world), thereās more to like than dislike imo. And Iām hardly what people would call a patriot!
As for vomiting, we also binge drink a lot, and the sewers are designed to cater to this
2 Likes
Latest from the algorithm. Not quite as surprising as some other recent suggestions, but it was ages since I listened to anything from Melody A.M. (an actual song from the actual album, that is, not some fragment from it used elsewhere):
4 Likes
Throbbing Gristle live with a bit of Discipline, something to get me going before work this afternoon
6 Likes