Warren Koch
@warrenkoch.bsky.social
Pinned
Warren Koch
@warrenkoch.bsky.social
· Nov 16
Hey Blueskies. This will be a Pro-Open-Source-AI feed, from a 15+yr senior programmer who studies this shit.
- sorry, it's not hype. wish it was.
- the public needs to own it (AI is the MoP)
- All Corps Are Bastards, so feel free to hate on corporate AI (tho Zuck is on a redeption arc)
- sorry, it's not hype. wish it was.
- the public needs to own it (AI is the MoP)
- All Corps Are Bastards, so feel free to hate on corporate AI (tho Zuck is on a redeption arc)
Reposted by Warren Koch
Let’s normalize using alt tech: Open-source, decentralized, and privacy-focused apps should be the default.
November 14, 2024 at 12:00 PM
Let’s normalize using alt tech: Open-source, decentralized, and privacy-focused apps should be the default.
Hey Blueskies. This will be a Pro-Open-Source-AI feed, from a 15+yr senior programmer who studies this shit.
- sorry, it's not hype. wish it was.
- the public needs to own it (AI is the MoP)
- All Corps Are Bastards, so feel free to hate on corporate AI (tho Zuck is on a redeption arc)
- sorry, it's not hype. wish it was.
- the public needs to own it (AI is the MoP)
- All Corps Are Bastards, so feel free to hate on corporate AI (tho Zuck is on a redeption arc)
November 16, 2024 at 5:21 AM
Hey Blueskies. This will be a Pro-Open-Source-AI feed, from a 15+yr senior programmer who studies this shit.
- sorry, it's not hype. wish it was.
- the public needs to own it (AI is the MoP)
- All Corps Are Bastards, so feel free to hate on corporate AI (tho Zuck is on a redeption arc)
- sorry, it's not hype. wish it was.
- the public needs to own it (AI is the MoP)
- All Corps Are Bastards, so feel free to hate on corporate AI (tho Zuck is on a redeption arc)
Reposted by Warren Koch
Bluesky right now feels a little like all the Whos gathering to sing after the Grinch stole their floofloovers and jingtinglers.
November 15, 2024 at 1:51 AM
Bluesky right now feels a little like all the Whos gathering to sing after the Grinch stole their floofloovers and jingtinglers.
Add in gaussian splat / nerf 3D modelling from just the image gen data and you've got a great error checker and can cache generations into persistent space.
Now just gotta solve for longer contexts. Wonder how well weighted attention would work to condense frame histories
Now just gotta solve for longer contexts. Wonder how well weighted attention would work to condense frame histories
November 16, 2024 at 4:45 AM
Add in gaussian splat / nerf 3D modelling from just the image gen data and you've got a great error checker and can cache generations into persistent space.
Now just gotta solve for longer contexts. Wonder how well weighted attention would work to condense frame histories
Now just gotta solve for longer contexts. Wonder how well weighted attention would work to condense frame histories
People without a programmer mindset fail to grasp what can be done with these tools.
Reminder: if and when we have stable world models, any stable/deterministic step can be shortened into LoRA training just like Stable Diffusion Turbo did - so time moves faster in sim https://t.co/OYlmYeRLoF
Reminder: if and when we have stable world models, any stable/deterministic step can be shortened into LoRA training just like Stable Diffusion Turbo did - so time moves faster in sim https://t.co/OYlmYeRLoF
November 16, 2024 at 4:45 AM
People without a programmer mindset fail to grasp what can be done with these tools.
Reminder: if and when we have stable world models, any stable/deterministic step can be shortened into LoRA training just like Stable Diffusion Turbo did - so time moves faster in sim https://t.co/OYlmYeRLoF
Reminder: if and when we have stable world models, any stable/deterministic step can be shortened into LoRA training just like Stable Diffusion Turbo did - so time moves faster in sim https://t.co/OYlmYeRLoF
We are not ready https://t.co/JTLt7d2Ltt
November 16, 2024 at 4:45 AM
We are not ready https://t.co/JTLt7d2Ltt
Amazing technical achievement, but gonna need higher quality example content than this [garbage] before I can get normies on board https://t.co/Mvd3a5zEiz
November 16, 2024 at 4:45 AM
Amazing technical achievement, but gonna need higher quality example content than this [garbage] before I can get normies on board https://t.co/Mvd3a5zEiz
And thus, open source takes the lead. Your move, OAI https://t.co/5uljxr9Mqt
November 16, 2024 at 4:46 AM
And thus, open source takes the lead. Your move, OAI https://t.co/5uljxr9Mqt
Now combine the last three into one... physics as a deterministic reality simulator... and man (agent) contending with being just one of many possible leaves in the wind https://t.co/Xn79QddysL
November 16, 2024 at 4:46 AM
Now combine the last three into one... physics as a deterministic reality simulator... and man (agent) contending with being just one of many possible leaves in the wind https://t.co/Xn79QddysL
Ahhh so that's where they got it https://t.co/24Kf3hJBU3
November 16, 2024 at 4:46 AM
Ahhh so that's where they got it https://t.co/24Kf3hJBU3
gg https://t.co/LHLp7lX2VN
November 16, 2024 at 4:46 AM
gg https://t.co/LHLp7lX2VN
This would even fit the prophesized Simpsons timeline
November 16, 2024 at 4:46 AM
This would even fit the prophesized Simpsons timeline
In all seriousness, T-Swift is probably the best poised and qualified candidate for next POTUS https://t.co/VzxYHNSnQe
November 16, 2024 at 4:46 AM
In all seriousness, T-Swift is probably the best poised and qualified candidate for next POTUS https://t.co/VzxYHNSnQe
Then the whole "this is very important, I might lose my job" typical prompt strat is to weigh the prompt highly enough to engage the whole brain rather than trigger kneejerk patterns. Big hit of caffeine/adrenaline equivalent.
November 16, 2024 at 4:46 AM
Then the whole "this is very important, I might lose my job" typical prompt strat is to weigh the prompt highly enough to engage the whole brain rather than trigger kneejerk patterns. Big hit of caffeine/adrenaline equivalent.
Allows the AI to use its "whole brain" for deep thinking. "Deep breath" seems to be that. Only be strict on the very few things you need to be strict on (the rules) as those cause narrowing.
November 16, 2024 at 4:46 AM
Allows the AI to use its "whole brain" for deep thinking. "Deep breath" seems to be that. Only be strict on the very few things you need to be strict on (the rules) as those cause narrowing.
I would expect this to "feel" like layering of multiple concepts from multiple modes of thought over the same problem. Metaphors/analogies of the same thing without contradicting. But then also with the hard rules important to the underlying problem able to cut through.
November 16, 2024 at 4:46 AM
I would expect this to "feel" like layering of multiple concepts from multiple modes of thought over the same problem. Metaphors/analogies of the same thing without contradicting. But then also with the hard rules important to the underlying problem able to cut through.
#ACCELERATE MANIFESTO for an Accelerationist Politics https://t.co/lcHVriojEJ via @critlegthinking
November 16, 2024 at 4:46 AM
#ACCELERATE MANIFESTO for an Accelerationist Politics https://t.co/lcHVriojEJ via @critlegthinking
Accelerating AI is not enough. We need to accelerate open source alternatives, UBI, automation targeting basic needs, charities, public services, and all the other social safety nets. If that's not ready by the time jobs disappear, we're toast.
November 16, 2024 at 4:46 AM
Accelerating AI is not enough. We need to accelerate open source alternatives, UBI, automation targeting basic needs, charities, public services, and all the other social safety nets. If that's not ready by the time jobs disappear, we're toast.
- Sure you lose a bit of your gains when you win (and if you think you have a significant edge over the average player, maybe that's not worth sharing), but the loss insulation results in a better average gain over time
November 16, 2024 at 4:46 AM
- Sure you lose a bit of your gains when you win (and if you think you have a significant edge over the average player, maybe that's not worth sharing), but the loss insulation results in a better average gain over time
- (this assumes a fair insurance market lol). Alternative equivalent concept: mining pool.
- Seems like this is the financial equivalent reasoning for herding in animals. Net benefit to individuals to share risk in games where you can lose forever (die)
- Seems like this is the financial equivalent reasoning for herding in animals. Net benefit to individuals to share risk in games where you can lose forever (die)
November 16, 2024 at 4:46 AM
- (this assumes a fair insurance market lol). Alternative equivalent concept: mining pool.
- Seems like this is the financial equivalent reasoning for herding in animals. Net benefit to individuals to share risk in games where you can lose forever (die)
- Seems like this is the financial equivalent reasoning for herding in animals. Net benefit to individuals to share risk in games where you can lose forever (die)
- because bankruptcy is possible, it's "non-ergodic". you can lose as an individual playing
- Answer: insurance! If you split gains/losses with thousands of other players, far lower chance you all go bankrupt at once. More ppl you have = closer to optimal $10 per game
- Answer: insurance! If you split gains/losses with thousands of other players, far lower chance you all go bankrupt at once. More ppl you have = closer to optimal $10 per game
November 16, 2024 at 4:46 AM
- because bankruptcy is possible, it's "non-ergodic". you can lose as an individual playing
- Answer: insurance! If you split gains/losses with thousands of other players, far lower chance you all go bankrupt at once. More ppl you have = closer to optimal $10 per game
- Answer: insurance! If you split gains/losses with thousands of other players, far lower chance you all go bankrupt at once. More ppl you have = closer to optimal $10 per game
grokking "Ergodicity":
- say you play a coin flip game: $60 heads, -$40 tails. Great game, expected value = .5*60 - .5*40 = $10 per game, right?
- well no... because you can lose it all still. Play forever = $0 from one bad string of luck. Thus your realistic average is lower
- say you play a coin flip game: $60 heads, -$40 tails. Great game, expected value = .5*60 - .5*40 = $10 per game, right?
- well no... because you can lose it all still. Play forever = $0 from one bad string of luck. Thus your realistic average is lower
November 16, 2024 at 4:46 AM
grokking "Ergodicity":
- say you play a coin flip game: $60 heads, -$40 tails. Great game, expected value = .5*60 - .5*40 = $10 per game, right?
- well no... because you can lose it all still. Play forever = $0 from one bad string of luck. Thus your realistic average is lower
- say you play a coin flip game: $60 heads, -$40 tails. Great game, expected value = .5*60 - .5*40 = $10 per game, right?
- well no... because you can lose it all still. Play forever = $0 from one bad string of luck. Thus your realistic average is lower
They made a South Park Westworld 😱
Guys - this better not result in a sentient Kenny... https://t.co/feuLLfB7Tk
Guys - this better not result in a sentient Kenny... https://t.co/feuLLfB7Tk
November 16, 2024 at 4:46 AM
They made a South Park Westworld 😱
Guys - this better not result in a sentient Kenny... https://t.co/feuLLfB7Tk
Guys - this better not result in a sentient Kenny... https://t.co/feuLLfB7Tk
"The marvel is not that the bear dances well, but that the bear dances at all" (err, cat)
Get your screenplays ready folks, making your own movie is about to get very achievable https://t.co/yDyWFx8TAw
Get your screenplays ready folks, making your own movie is about to get very achievable https://t.co/yDyWFx8TAw
November 16, 2024 at 4:46 AM
"The marvel is not that the bear dances well, but that the bear dances at all" (err, cat)
Get your screenplays ready folks, making your own movie is about to get very achievable https://t.co/yDyWFx8TAw
Get your screenplays ready folks, making your own movie is about to get very achievable https://t.co/yDyWFx8TAw
Heh. all it needs to do is shrink itself with a Llama model and it could hide in any reasonably sized hard drive. Considering the amount of people about to play around with running its instructions on execute loops in the near future, that's pretty much inevitable. https://t.co/inRtLUmWCQ
November 16, 2024 at 4:45 AM
Heh. all it needs to do is shrink itself with a Llama model and it could hide in any reasonably sized hard drive. Considering the amount of people about to play around with running its instructions on execute loops in the near future, that's pretty much inevitable. https://t.co/inRtLUmWCQ