Marco D'Agostini
@madacol.com
madacol.com
I love to automate stuffs, mostly using web pages nowadays
I love to automate stuffs, mostly using web pages nowadays
and yet it still is a win-win scenario.
Most venezuelans would gladly give all the oil if it meant this criminals left power
Most venezuelans would gladly give all the oil if it meant this criminals left power
December 17, 2025 at 4:35 PM
and yet it still is a win-win scenario.
Most venezuelans would gladly give all the oil if it meant this criminals left power
Most venezuelans would gladly give all the oil if it meant this criminals left power
I know, but I still can't see the difference
December 9, 2025 at 6:22 PM
I know, but I still can't see the difference
I fail to differentiate rounding-to-false vs oversimplified
December 9, 2025 at 6:15 PM
I fail to differentiate rounding-to-false vs oversimplified
if you click the share button, there's an tick box to add the time to the URL
November 14, 2025 at 5:27 PM
if you click the share button, there's an tick box to add the time to the URL
I find it strange that there was no number for how much safer is vaping vs smoking
is it 10% safer?
100%?
10x safer?
100x safer?
What are the orders of magnitudes we are dealing with?
is it 10% safer?
100%?
10x safer?
100x safer?
What are the orders of magnitudes we are dealing with?
November 4, 2025 at 3:21 PM
I find it strange that there was no number for how much safer is vaping vs smoking
is it 10% safer?
100%?
10x safer?
100x safer?
What are the orders of magnitudes we are dealing with?
is it 10% safer?
100%?
10x safer?
100x safer?
What are the orders of magnitudes we are dealing with?
our ability to use AI will rely on our ability to verify.
And to verify you need to know some stuffs
And to verify you need to know some stuffs
November 2, 2025 at 12:09 PM
our ability to use AI will rely on our ability to verify.
And to verify you need to know some stuffs
And to verify you need to know some stuffs
Although I agree with that sentiment in general, I am not sure that this is a good representing case of that corruption.
Anti money laundering laws usually look like a medicine that is worst than the disease
Anti money laundering laws usually look like a medicine that is worst than the disease
October 23, 2025 at 3:52 PM
Although I agree with that sentiment in general, I am not sure that this is a good representing case of that corruption.
Anti money laundering laws usually look like a medicine that is worst than the disease
Anti money laundering laws usually look like a medicine that is worst than the disease
October 19, 2025 at 12:46 PM
It seems It just needs debouncing
October 18, 2025 at 5:56 PM
It seems It just needs debouncing
Simon you only need one skill.
the skill to read your blog and extract skills from each one of your posts
the skill to read your blog and extract skills from each one of your posts
October 17, 2025 at 7:23 AM
Simon you only need one skill.
the skill to read your blog and extract skills from each one of your posts
the skill to read your blog and extract skills from each one of your posts
@simonwillison.net I'm sure I'm not the first person to try this, but running your pelican test on some of the horny/abliterated models is horrifyingly hilarious. Some of the comments on the SVG are perfect. gist.github.com/kdkd/84c0453...
October 16, 2025 at 11:38 AM
that pelican looks disturbingly similar to the one generated by the horny model
"welcoming ass chick posture"
"welcoming ass chick posture"
October 16, 2025 at 11:33 AM
that pelican looks disturbingly similar to the one generated by the horny model
"welcoming ass chick posture"
"welcoming ass chick posture"
do you have anything written about that setup and how you talk to it?.
Seems like something I want to replicate
Seems like something I want to replicate
October 12, 2025 at 12:47 PM
do you have anything written about that setup and how you talk to it?.
Seems like something I want to replicate
Seems like something I want to replicate
I'm having a hard time understanding what's going on. What is the purpose of SKILLS.md?
What is the tool doing?
It seems like a bunch of prompts to encourage Claude to follow a specific type of process, but is it doing more than that? or am I failing to see the potential in that?
What is the tool doing?
It seems like a bunch of prompts to encourage Claude to follow a specific type of process, but is it doing more than that? or am I failing to see the potential in that?
SKILLS.md
October 11, 2025 at 4:53 PM
I'm having a hard time understanding what's going on. What is the purpose of SKILLS.md?
What is the tool doing?
It seems like a bunch of prompts to encourage Claude to follow a specific type of process, but is it doing more than that? or am I failing to see the potential in that?
What is the tool doing?
It seems like a bunch of prompts to encourage Claude to follow a specific type of process, but is it doing more than that? or am I failing to see the potential in that?
Just to be clear, this is not running lean, the engine, in the browser.
It's forwarding the code to a server where Lean runs.
BTW, can lean 4 run on the browser via WASM? Is there any plan for it?
It's forwarding the code to a server where Lean runs.
BTW, can lean 4 run on the browser via WASM? Is there any plan for it?
October 10, 2025 at 7:53 PM
Just to be clear, this is not running lean, the engine, in the browser.
It's forwarding the code to a server where Lean runs.
BTW, can lean 4 run on the browser via WASM? Is there any plan for it?
It's forwarding the code to a server where Lean runs.
BTW, can lean 4 run on the browser via WASM? Is there any plan for it?
Reposted by Marco D'Agostini
(The same thing extends to prompt injection. Human employees are often tricked, scammed or phished, and that’s the baseline that computers need to beat.)
October 9, 2025 at 12:24 PM
(The same thing extends to prompt injection. Human employees are often tricked, scammed or phished, and that’s the baseline that computers need to beat.)
I suppose you have already tried by giving more weight to the likes of your likes/follow.
I find it a bit counterintuitive that this ☝️strategy performs worse 🫤
I find it a bit counterintuitive that this ☝️strategy performs worse 🫤
October 9, 2025 at 10:54 AM
I suppose you have already tried by giving more weight to the likes of your likes/follow.
I find it a bit counterintuitive that this ☝️strategy performs worse 🫤
I find it a bit counterintuitive that this ☝️strategy performs worse 🫤
> When you like a reply - you
increase your "trust score" toward people who liked that reply.
I don't think that's a good idea
people like things motivated by different reasons, and sometimes there's misunderstandings and lack of context, which means the signal-to-noise ratio of that signal is low
increase your "trust score" toward people who liked that reply.
I don't think that's a good idea
people like things motivated by different reasons, and sometimes there's misunderstandings and lack of context, which means the signal-to-noise ratio of that signal is low
October 9, 2025 at 9:13 AM
> When you like a reply - you
increase your "trust score" toward people who liked that reply.
I don't think that's a good idea
people like things motivated by different reasons, and sometimes there's misunderstandings and lack of context, which means the signal-to-noise ratio of that signal is low
increase your "trust score" toward people who liked that reply.
I don't think that's a good idea
people like things motivated by different reasons, and sometimes there's misunderstandings and lack of context, which means the signal-to-noise ratio of that signal is low
Reposted by Marco D'Agostini
Both these are true.
October 5, 2025 at 8:43 PM
Both these are true.
I think Rob miles has said that misalignments can survive post training processes
I assume that includes RL
Of course, that doesn't rule out that it could be reduced to negligible amounts and make it work in practice, is this what you mean?
I assume that includes RL
Of course, that doesn't rule out that it could be reduced to negligible amounts and make it work in practice, is this what you mean?
October 3, 2025 at 7:45 AM
I think Rob miles has said that misalignments can survive post training processes
I assume that includes RL
Of course, that doesn't rule out that it could be reduced to negligible amounts and make it work in practice, is this what you mean?
I assume that includes RL
Of course, that doesn't rule out that it could be reduced to negligible amounts and make it work in practice, is this what you mean?