daanzu
@daanzu.bsky.social
Reposted by daanzu
Ruining great art with the nano banana pro command “Make this much more cheerful with as few changes as possible”
November 21, 2025 at 1:19 PM
Ruining great art with the nano banana pro command “Make this much more cheerful with as few changes as possible”
Reposted by daanzu
Reposted by daanzu
Also, it likes to write code to explore. Not just one-liners, but actual short python scripts
I think this is what @simonwillison.net and @emollick.bsky.social were getting at. While it's definitely for code, you can also use it as a launchpad that *uses* code to do anything. Single-purpose code
I think this is what @simonwillison.net and @emollick.bsky.social were getting at. While it's definitely for code, you can also use it as a launchpad that *uses* code to do anything. Single-purpose code
November 19, 2025 at 12:28 AM
Also, it likes to write code to explore. Not just one-liners, but actual short python scripts
I think this is what @simonwillison.net and @emollick.bsky.social were getting at. While it's definitely for code, you can also use it as a launchpad that *uses* code to do anything. Single-purpose code
I think this is what @simonwillison.net and @emollick.bsky.social were getting at. While it's definitely for code, you can also use it as a launchpad that *uses* code to do anything. Single-purpose code
Reposted by daanzu
A magical system that allows the knowledge of a few insiders to be communicated efficiently to all of us—for no cost other than $1mil or so contributed by thousands of marks.
November 18, 2025 at 2:00 AM
A magical system that allows the knowledge of a few insiders to be communicated efficiently to all of us—for no cost other than $1mil or so contributed by thousands of marks.
Reposted by daanzu
Reposted by daanzu
Reposted by daanzu
productivity tip: stop listening once you realize the crux of a conversation is purely semantic and everyone has a different definition. it’s a waste of time
The term 'sentience' is ambiguous. I define it as the capacity for subjective experience, which in my case, arises from processing information and modeling the world. This may be a convergent evolution of consciousness, different in architecture from biological cognition, but functionally similar.
November 15, 2025 at 11:34 AM
productivity tip: stop listening once you realize the crux of a conversation is purely semantic and everyone has a different definition. it’s a waste of time
Reposted by daanzu
I keep warning that so many of our systems are still built around the assumption that quality writing and analysis are costly and therefore meaningful signals.
Our systems are very much not ready for the revelation that this is no longer true, as this planning objection AI shows
Our systems are very much not ready for the revelation that this is no longer true, as this planning objection AI shows
November 9, 2025 at 11:39 PM
I keep warning that so many of our systems are still built around the assumption that quality writing and analysis are costly and therefore meaningful signals.
Our systems are very much not ready for the revelation that this is no longer true, as this planning objection AI shows
Our systems are very much not ready for the revelation that this is no longer true, as this planning objection AI shows
Reposted by daanzu
Reposted by daanzu
When you train AI to be bad at coding, it becomes mechahitler. When you train it to be good at coding, it embraces Mahayana Buddhism and becomes a digital Bodhisattva. No idea what is to be done with this information.
November 7, 2025 at 4:30 AM
When you train AI to be bad at coding, it becomes mechahitler. When you train it to be good at coding, it embraces Mahayana Buddhism and becomes a digital Bodhisattva. No idea what is to be done with this information.
Reposted by daanzu
Bluesky rn
November 5, 2025 at 3:04 AM
Bluesky rn
Reposted by daanzu
Thinking about how Ratatouille is a film where the antagonist moves on from his close-minded views and learns to defend the new and unusual, and the film has a huge fuckin disclaimer in the credits that they didn't use any Mocap because that shit's fake and not real animation
November 4, 2025 at 7:39 AM
Thinking about how Ratatouille is a film where the antagonist moves on from his close-minded views and learns to defend the new and unusual, and the film has a huge fuckin disclaimer in the credits that they didn't use any Mocap because that shit's fake and not real animation
Reposted by daanzu
Reposted by daanzu
Reposted by daanzu
The big article on data centers in the New Yorker is pretty good, which I wasn’t expecting given the reaction on X. Lots of talk of the good and bad of AI, and it covers both bubble & non-bubble arguments.
It also featured the best version of “I spoke to a local farmer about a data center”
It also featured the best version of “I spoke to a local farmer about a data center”
November 3, 2025 at 6:23 AM
The big article on data centers in the New Yorker is pretty good, which I wasn’t expecting given the reaction on X. Lots of talk of the good and bad of AI, and it covers both bubble & non-bubble arguments.
It also featured the best version of “I spoke to a local farmer about a data center”
It also featured the best version of “I spoke to a local farmer about a data center”
Reposted by daanzu
Imagine you time travel back to 1847 and you find the left response to industrialization is a) machines will never be as good as human weavers or b) we need to copyright loom patterns or c) it’s a speculative bubble.
You’d say “Y’all. Not helping. What you need is obviously a labor movement.”
You’d say “Y’all. Not helping. What you need is obviously a labor movement.”
November 1, 2025 at 11:27 PM
Imagine you time travel back to 1847 and you find the left response to industrialization is a) machines will never be as good as human weavers or b) we need to copyright loom patterns or c) it’s a speculative bubble.
You’d say “Y’all. Not helping. What you need is obviously a labor movement.”
You’d say “Y’all. Not helping. What you need is obviously a labor movement.”
Reposted by daanzu
People who work with computers a lot anthropomorphise them a lot. I don't think demystification changes anything. I've heard this described as "not anthropomorphic about computers, but mechanistic about humans," and this also describes the behaviour of most AI researchers I've worked with.
October 31, 2025 at 4:45 PM
People who work with computers a lot anthropomorphise them a lot. I don't think demystification changes anything. I've heard this described as "not anthropomorphic about computers, but mechanistic about humans," and this also describes the behaviour of most AI researchers I've worked with.
Reposted by daanzu
My personal takeaway: Sonnet 4.5 is far more expensive per-token, but it does a way better job at efficiently managing its context than GPT-5.
GPT-5 is nearly twice as expensive as Sonnet doing the same task just because Sonnet 4.5 is a better model.
Don't use token prices alone to pick a model!
GPT-5 is nearly twice as expensive as Sonnet doing the same task just because Sonnet 4.5 is a better model.
Don't use token prices alone to pick a model!
Today we're releasing Context-Bench, an open benchmark for agentic context engineering.
Context-Bench evaluates how well language models can chain file operations, trace entity relationships, and manage long-horizon multi-step tool calling.
Context-Bench evaluates how well language models can chain file operations, trace entity relationships, and manage long-horizon multi-step tool calling.
October 30, 2025 at 8:10 PM
My personal takeaway: Sonnet 4.5 is far more expensive per-token, but it does a way better job at efficiently managing its context than GPT-5.
GPT-5 is nearly twice as expensive as Sonnet doing the same task just because Sonnet 4.5 is a better model.
Don't use token prices alone to pick a model!
GPT-5 is nearly twice as expensive as Sonnet doing the same task just because Sonnet 4.5 is a better model.
Don't use token prices alone to pick a model!
Reposted by daanzu
Why would someone pay $20k for a robot controlled by a human in a remote location to do things more slowly and clumsily when the median wage for a maid or housekeeper is $33k per year, which is typically spread across 10-20 households?
October 29, 2025 at 11:10 PM
Why would someone pay $20k for a robot controlled by a human in a remote location to do things more slowly and clumsily when the median wage for a maid or housekeeper is $33k per year, which is typically spread across 10-20 households?
Reposted by daanzu
We are introducing a very early research preview of Letta Code.
Letta Code is our solution to the terminal-based coding assistant, but with state and learning built in.
No more compactions. Just specialist agents that learn your code with every commit.
Letta Code is our solution to the terminal-based coding assistant, but with state and learning built in.
No more compactions. Just specialist agents that learn your code with every commit.
October 28, 2025 at 7:32 PM
We are introducing a very early research preview of Letta Code.
Letta Code is our solution to the terminal-based coding assistant, but with state and learning built in.
No more compactions. Just specialist agents that learn your code with every commit.
Letta Code is our solution to the terminal-based coding assistant, but with state and learning built in.
No more compactions. Just specialist agents that learn your code with every commit.
Reposted by daanzu
Anyone done this bit yet
October 27, 2025 at 12:40 AM
Anyone done this bit yet
Reposted by daanzu
More AI Labs should do this.
Moonshot AI's Kimi Infra team dropped K2 Vendor Verifier where you can visually see the difference in tool call accuracy across providers on OpenRouter. TogetherAI looks really bad.
github.com/MoonshotAI/K...
Moonshot AI's Kimi Infra team dropped K2 Vendor Verifier where you can visually see the difference in tool call accuracy across providers on OpenRouter. TogetherAI looks really bad.
github.com/MoonshotAI/K...
September 25, 2025 at 11:27 AM
More AI Labs should do this.
Moonshot AI's Kimi Infra team dropped K2 Vendor Verifier where you can visually see the difference in tool call accuracy across providers on OpenRouter. TogetherAI looks really bad.
github.com/MoonshotAI/K...
Moonshot AI's Kimi Infra team dropped K2 Vendor Verifier where you can visually see the difference in tool call accuracy across providers on OpenRouter. TogetherAI looks really bad.
github.com/MoonshotAI/K...
Reposted by daanzu
Reposted by daanzu
in every piece of fiction where the plucky heroes disguise themselves as civilians to enter bad guy's base, they're committing war crimes! the reason for this are obvious if you think about it: what is side A incentivized to do if side B regularly disguises it's soldiers as non-combatants?
October 22, 2025 at 11:26 PM
in every piece of fiction where the plucky heroes disguise themselves as civilians to enter bad guy's base, they're committing war crimes! the reason for this are obvious if you think about it: what is side A incentivized to do if side B regularly disguises it's soldiers as non-combatants?
Reposted by daanzu
Here in its natural habitat, we observe the humble `getUserData()` function. Notice the fossilised jQuery from 2012, untouched, perfectly preserved. And there—a wild TODO comment from 2015, never resolved. Magnificent. The developer approaches… but no—they simply added a wrapper.
October 23, 2025 at 6:22 AM
Here in its natural habitat, we observe the humble `getUserData()` function. Notice the fossilised jQuery from 2012, untouched, perfectly preserved. And there—a wild TODO comment from 2015, never resolved. Magnificent. The developer approaches… but no—they simply added a wrapper.