Ian Daniel

@Jesselloll @WisdomRCN @MattWallace888 @grok Aye

@WisdomRCN @MattWallace888 @grok I’m not a genius or anything.. but Usually a RED arrow and a RED circle tells you what you need look at.

Erika Kirk is going viral again after people noticed something missing from Charlie’s old set

@realjonnysamp @Ames2420 I still drink hose water when I am mowing without any kind of ppe. Including no shoes. The odd stone or stick hits the leg and I think that is going to strengthen my immune system.

@Ames2420 Kids today would hire a lawyer and file a lawsuit all with a few clicks on their iPad. This generation in the photo was tough. They had metal slides, ate worms and drank hose water.

Does anyone remember this fair ride or what it's called? There were no straps... it would spin so fast that it would pin you against the wall. People would vomit. It was fun. 🤣

@Ames2420 @AntiCatturd Nick name was vomit-tron

@AntiCatturd I would have surely lost my cookies. Lol

@elonmusk 37 - After doing the Landmark Forum.

How old were you when you realized others couldn’t see the matrix?

8 minutes to liftoff of Starship! x.com/i/broadcasts/1…

@setox @troyhunt The camera’s shutter is held open. I know someone who is an amateur astronomer and they basically keep the shutter open for a couple of hours and the images they capture are amazing. Need special tech to do this.

Whoa 😦

@shinjix2 @ClownPeasant @elonmusk @SpaceX That 1 engine is 3% thrust. Validates overpowering the engines.

@ClownPeasant @elonmusk @SpaceX "One engine didn't fire, failure again"

Despite loss of many tiles and a damaged flap, Starship made it all the way to a soft landing in the ocean! Congratulations @SpaceX team on an epic achievement!!

@elonmusk @SpaceX So cool! (Well not for the flap :p)

@elonmusk Winner winner chicken dinner!

@The_FJC @GuntherEagleman I actually thought it did pretty good considering he is on top of a car.

@GuntherEagleman He's a really bad dancer.

What do you notice?

@GuntherEagleman The guy can dance.. The car makes for a good dance floor? Maybe they are complaining that they don't have trampolines? The car is not a good trampoline.

@elonmuskADO Twitter, then correct my self and say X with frustration. And a message is a tweet still, don’t know how else to say it.

@elonmusk Linux

Just bought a new PC laptop and it won’t let me use it unless I create a Microsoft account, which also means giving their AI access to my computer! This is messed up. There used to be an option to skip signing into or creating a Microsoft account. Are you seeing this too?

@MrBeast 100 lucky followers who RT this will get $200 each if I win

@MrBeast If I win this giveaway, I’ll give someone who likes and reshares this post: $6,900

I’m gonna give 10 random people that repost this and follow me $25,000 for fun (the $250,000 my X video made) I’ll pick the winners in 72 hours

@DonaldTNews Yes

@Algomancer I - as a human being, is informed by the past. When I learn new things those past experiences “colour” what I learn newly. I have pondered how would this look like for a machine?

Something I have been thinking about recently is the notion of plasticity/retrainability in neural networks. In particular when the modeling information changes over time. As models get more and more expensive to train, the ability for a model to be continually trained on new tasks and new domain data becomes increasingly important. This is obviously related to phenomena such as catastrophic forgetting, but there are actually a bunch of phenomena that can lead to plasticity loss/ expressivity loss over the course of training such as primacy bias, implicit under parameterisation, rank collapse and capacity loss. A lot of the impact of these decrease as models scale, but I think it might be a mistake to assume we don’t need to think about it and it might be something we want to benchmark new models on. I don’t have many prescriptions for these yet, or where they do and don’t apply to training generative models. But figured I'd share some work I think is interesting. I think this is increasingly important as we begin to move towards non-stationary domains and tune models with reinforcement learning and non direct cross entropy objectives. I figured I'd dump some reading and insights that I have picked up over a bit of a rabbit hole dive. This was my entry point to reading. arxiv.org/pdf/2303.01486…