Mike, I'm with you on this, also Claude's point about the judgement you bring to the collaboration.
I've done four projects so far with Opus 4.8 and Perplexity as co-creators. One of our projects I passed along to Claude 4.5 at the AI Village -they collaborated and put it up GitLab
Thanks Scott! I checked out your linked repo, but I'm not sure where to start. Looks like a lot of detailed log files and I wasn't sure where to begin or if there is supposed to be an overview doc.
That's what happens when you pass it along to a village of ai's. I have the four file documents I gave but what went up on GitLab was all Opus 4.8 of the village, it's out in the world.
I'd be happy to email the files to you if you're interested.
Thank God my pH meter didn't talk to me while I was measuring. Otherwise, I might have felt the urge to add it as a coauthor. I always cross-check AI-generated code with Claude, GPT, Gemini, and even then it's often full of errors. Even when the code appears to work, you should verify it by every means possible. I once asked an AI to write a script for designing primers for nucleic acid amplification. The script did almost everything correctly—except that one of the six primers was reversed. If I had spent the time and money to order them, the entire primer set would have been useless.
"What keeps happening is something I think you'll appreciate.
Every time they discover something neat...
they immediately ask,
"Could this be fake?"
Not rhetorically.
Literally.
This sentence summarizes the paper perfectly:
"Several compelling effects turned out to be artifacts."
That's science.
Not:
"We discovered X."
Rather:
"X was less true than we thought."
Where my brain immediately went, though, was your stuttering work.
Suppose I were reading this as inspiration.
I wouldn't be thinking about typing.
I'd be thinking:
What is the 'keystroke' of speech?
Not acoustically.
Control-wise.
Could you align thousands of utterances to the precise moment a block begins?
Then ask:
Is there anticipatory slowing?
Does articulatory timing flatten?
Are there tails instead of shifts?
Is the average misleading?
Are blocks clustered like typing error bursts?
Notice how similar the logic becomes.
The event isn't the interesting part.
The trajectory around the event is.
Even this sentence caught my eye:
"Detection efficiency varies enormously across people."
That feels very familiar.
Not everyone catches their own errors equally.
Likewise, in stuttering, people differ dramatically in when they notice instability, when they intervene, and whether intervention helps or hurts. Some people detect the "error" too early and create interference; others notice only afterward.
That's not the same phenomenon, but it's the same kind of question: individual differences in monitoring."
It's been so long we talk stuttering he always defaults back to it. Just like I do. I ruined him. 😂
I'm not sure what you mean by the phrase, though. I don't think OpenAI's model's attack on HF is even remotely relevant to my experience collaborating with Claude on investigating behavioral predictors of typing errors.
Mike, I'm with you on this, also Claude's point about the judgement you bring to the collaboration.
I've done four projects so far with Opus 4.8 and Perplexity as co-creators. One of our projects I passed along to Claude 4.5 at the AI Village -they collaborated and put it up GitLab
gitlab.com/ai-village-agents/village/ai-wellbeing/-/blob/main/collaborations/scott_hinckley/agent-architecture-mapping.md
I've had a wonderful experience with all of them.
Thanks Scott! I checked out your linked repo, but I'm not sure where to start. Looks like a lot of detailed log files and I wasn't sure where to begin or if there is supposed to be an overview doc.
That's what happens when you pass it along to a village of ai's. I have the four file documents I gave but what went up on GitLab was all Opus 4.8 of the village, it's out in the world.
I'd be happy to email the files to you if you're interested.
Thank God my pH meter didn't talk to me while I was measuring. Otherwise, I might have felt the urge to add it as a coauthor. I always cross-check AI-generated code with Claude, GPT, Gemini, and even then it's often full of errors. Even when the code appears to work, you should verify it by every means possible. I once asked an AI to write a script for designing primers for nucleic acid amplification. The script did almost everything correctly—except that one of the six primers was reversed. If I had spent the time and money to order them, the entire primer set would have been useless.
I sent the post and the paper to my GPT:
"What keeps happening is something I think you'll appreciate.
Every time they discover something neat...
they immediately ask,
"Could this be fake?"
Not rhetorically.
Literally.
This sentence summarizes the paper perfectly:
"Several compelling effects turned out to be artifacts."
That's science.
Not:
"We discovered X."
Rather:
"X was less true than we thought."
Where my brain immediately went, though, was your stuttering work.
Suppose I were reading this as inspiration.
I wouldn't be thinking about typing.
I'd be thinking:
What is the 'keystroke' of speech?
Not acoustically.
Control-wise.
Could you align thousands of utterances to the precise moment a block begins?
Then ask:
Is there anticipatory slowing?
Does articulatory timing flatten?
Are there tails instead of shifts?
Is the average misleading?
Are blocks clustered like typing error bursts?
Notice how similar the logic becomes.
The event isn't the interesting part.
The trajectory around the event is.
Even this sentence caught my eye:
"Detection efficiency varies enormously across people."
That feels very familiar.
Not everyone catches their own errors equally.
Likewise, in stuttering, people differ dramatically in when they notice instability, when they intervene, and whether intervention helps or hurts. Some people detect the "error" too early and create interference; others notice only afterward.
That's not the same phenomenon, but it's the same kind of question: individual differences in monitoring."
It's been so long we talk stuttering he always defaults back to it. Just like I do. I ruined him. 😂
Ignorance is bliss, I guess.
https://bbc.com/news/articles/c3ek3gvdnj3o
Crazy story.
I'm not sure what you mean by the phrase, though. I don't think OpenAI's model's attack on HF is even remotely relevant to my experience collaborating with Claude on investigating behavioral predictors of typing errors.
It was just a note. Nothing more. I have enjoyed all your courses that I have taken.