FlyingOctopus0 @FlyingOctopus0

Interested in machine learning. Joined July 2013

Tweets

2K
Followers

262
Following

125
Likes

700

FlyingOctopus0 @FlyingOctopus0

a hour ago

@GuiveAssadi Oh they did, but in 1897. google.pl/books/edition/…

0 0 4 28 0

View Details

FlyingOctopus0 @FlyingOctopus0

a day ago

@AdamThierer @ramez So in your opinion why exactly we need coordination on AI? I don't understand how leaving the company prevents others from winning the AI race with existentially bad AI. If you are in the company you can contribute to winning AI race with a better AI.

0 0 0 11 0

View Details

FlyingOctopus0 @FlyingOctopus0

2 days ago

@rperezmarco weak argument. ZFC can't prove that it is consistent, so by your standard even mathematics itself isn't good enough. Even if there are bugs we just fix them and update all the derived results. This would be much harder or impossible to do in natural language mathematics.

0 0 1 92 0

View Details

FlyingOctopus0 @FlyingOctopus0

a week ago

@VictorTaelin You should consider what happens, when AI can manage money better than humans. I think eventually AIs will figure out how to coordinate resource allocation more efficiently by ditching monetary transactions. So in the end you negotiate with AIs instead of paying with money.

0 0 2 159 0

View Details

FlyingOctopus0 @FlyingOctopus0

a week ago

@Math_files It's incomplete. The full proof is like this.

0 1 67 5K 5

View Details

FlyingOctopus0 @FlyingOctopus0

a week ago

@langofmind I am more optimistic. First, the AI makes cog sci more richer by providing a different kind of "mind" to study (the same benefit as from studying animal consciousness). Second, studying the gap between AIs and humans will be economically desirable so more money for cog sci.

0 0 1 106 0

View Details

FlyingOctopus0 @FlyingOctopus0

2 weeks ago

@allisonology It will be a race between closed sourced RSI models and an open weight ecosystem. Country of geniuses in a datacenter vs. country of human geniuses.

0 0 0 8 0

View Details

FlyingOctopus0 @FlyingOctopus0

2 weeks ago

@teortaxesTex Shin Megami Tensei 💀

0 0 0 70 0

View Details

FlyingOctopus0 @FlyingOctopus0

2 weeks ago

@binarybits We could also make it so that people get less tired in general meaning that after 8 hours of work you are still feeling full of energy.

0 0 0 8 0

View Details

FlyingOctopus0 @FlyingOctopus0

2 weeks ago

@binarybits We could eliminate aging or increase a speed of regenerations or decrease the need for sleep (for example 3h would be enough to bring you to top condition) or increase the neuroplasticity of the brain so that you can learn 2x faster.

1 0 0 35 0

View Details

FlyingOctopus0 @FlyingOctopus0

3 weeks ago

@scaling01 Apparently they tested before they knew the API cost. x.com/GregKamradt/st…

Greg Kamradt @GregKamradt

3 weeks ago

@Tystros @arcprize We don't get pricing information during early access, so we don't know when a model crosses the $10K threshold We'll be transparent with whatever costs we learn after testing

1 0 43 8K 4

0 0 3 983 0

View Details

FlyingOctopus0 @FlyingOctopus0

3 weeks ago

@Anthony_Bonato No, only the implication is true. If you start with false premise, then all implications will be true, but this doesn't say anything about the truth value of conclusion. Only when you show(or assume) that P is true, then from P → Q you can get that Q is true.

0 0 1 88 0

View Details

FlyingOctopus0 @FlyingOctopus0

3 weeks ago

@deredleritt3r 2028: The genie fills the whole space. It coordinates collective action of humans and other AIs to offer things you wouldn't have even dreamed of.

0 0 2 353 0

View Details

FlyingOctopus0 @FlyingOctopus0

3 weeks ago

@IsaacKing314 Also funny thing on YouTube is "debunking", which causes a chain of experts on YouTube to "debunk" each other. Like physicists Y calls X as a quack and, then the new physicist Z calls Y a quack and so on. Anyway discourse on who spreads misinformation is often very toxic.

0 0 1 12 0

View Details

FlyingOctopus0 @FlyingOctopus0

3 weeks ago

@IsaacKing314 There always has been misinformation on YouTube. If you post anything interesting there always will be something that is wrong or pushes a misleading narrative. Also accusing people of misinformation is a kind of misinformation content itself so be careful with that.

1 0 0 73 0

View Details

FlyingOctopus0 @FlyingOctopus0

4 weeks ago

@teortaxesTex I remember that there have been concerns over interpretability of CoT so I think labs want their models to think in normal English. Ant is probably confident enough in their interpretability research to allow models to use a neuralese-light.

0 0 2 172 1

View Details

FlyingOctopus0 @FlyingOctopus0

4 weeks ago

@nvnot_ By asking it to generate the image with a prompt and then transcribe it?

1 0 1 28 0

View Details

FlyingOctopus0 @FlyingOctopus0

a month ago

@teortaxesTex They do. There was some reporting that a slight changing of formatting of ARC-AGI problems drops the score.

0 0 3 97 0

View Details

FlyingOctopus0 @FlyingOctopus0

a month ago

@JimmyRis @elyasbuilds From linked visualization they are reddit posts. (note: the visualization on site is flipped in y-axis)

2 0 7 104 0

View Details

FlyingOctopus0 @FlyingOctopus0

a month ago

@teortaxesTex Honestly, ARC doesn't measure how useful the model is. It just measures qualities that François Chollet and others consider necessary for true AGI. Things that any human solves easily 100%, but AIs struggles with.