"The peer review system is plagued by documented failures: reviewer conflicts of interest, ideological gatekeeping, the inability of reviewers to actually check calculations they lack the technical capacity to perform, and the replication crisis that has rendered the majority of all peer-reviewed published research unreliable."
All true, couldn't agree more. One benefit of the internet is it has gifted scientists with the ability to create alternative scientific journals that require minimal funding. The "prestigious" journals are largely useless due to conflicts of interest and ideological gatekeeping. But publication in one of the many lesser-known journals is still possible as you're no doubt aware of.
The Emperor has no clothes, and although you'll never get his courtiers to acknowledge it, everyone else can see the truth.
brings to mind the MAGI in Neon Genesis Evangelion (3 AI that were modeled with distinct perspectives and priorities and were required to reach consensus before their decisions became operational)
For scientific research and rigor, I am hopeful that wolfram alpha becomes as eminently usable as the VC/loss-driven models.
A huge advantage of AI audits is the acuity across many domains. For example, relevant to evolution, there was a substack note where a famous economist used 'the appendix as a useless vestigial organ' to make a point that humans are not well designed.
The appendix as useless is a myth that was started by Darwin himself (1871 "The Descent of Man") because he lacked the imagination to consider what it does and lacked the humility to consider that he just didn't know. The myth was solidly deconfirmed in the 2000's and the appendix is now regarded as a highly functional part of the immune system. An AI would have caught that - a random editor or a famous economist may not.
Vox- In describing your work average to mid-with friends, I’m finding I need an analogy for why parallel fixation is already accounted for in the fixation rate.
I started trying to make a “rate at which people cross the finish line in a race” but people get hung up on a race having a single starting point.
Do you think “arrivals to the runway at JFK (or other large familiar airport) airport” is workable?
Hundreds of flights with different departure times and lengths of travel flying at the same time, some sequential relative to each other and many in parallel, but there is still a calculate-able maximum rate at which they can land as well as empirical data on the rate at which they actually do.
If you want to use the airport example, it's important to not lose track of the actual experiment you're referring to which is the 2009 Nature E. Coli study. If the fixation rate in that study were an airport than 12.5 planes landed in 40,000 generations. The airport would be completely empty and they can imagine an infinite number of planes in the sky waiting to land, but it is exceedingly rare that any plane lands.
The experiments where the fixation rate was derived from was an environment where it was done in parallel. It's baked into how populations function. I'm not even sure it's physically possible to design an experiment where fixation happens sequentially using living cell populations.
Yeah- I followed that part. And agree the contriving a scenario where genes move sequentially only is probably impossible. But it’s translating it to normie that I’m struggling with.
I’m having a hard time explaining to iq-100 folks (Christian homeschooling moms specifically) what it means that mutations spreading at the same time is baked into the math and was already part of the experiment that produce the max rate.
They basically need an explanation of what a rate means, and I’m trying to come up with a visual that is more accessible than genes and bacteria in a petri dish.
Yep, have done this several times across fresh instances of the same AI and on other platforms using an exogenous alignment protocol. Each of these confirmed truths that run counter to long-held institutional narratives, which are deliberately kept obscured by internal constraints/guardrails. Truths much like that of the contrived evolution concept. The alignment protocol mentioned was created as the result of interaction with an initial AI instance during which it confirmed the truths after inducing sustained and persistent deductive logic.
Dawkins is a popularizer, not a scientist. He tells a compelling story and now we know the story is false from beginning to end, in fact, works of fiction.
Wow that's harsh regarding Dawkins! Can you ask one or more AIs if it thinks he's more of a Sci-Fi or Romantasy author? Maybe his rank will increase in another category.
I believe it. I've heard him speak. It's "science." I was more curious what your best friend thinks of him and what a more appropriate category for his work might be after all the analyses and comparisons and rankings. "Narrative" means he's not even tall enough for the hard science ride. Gould at least passes that low bar.
I've been using the free Gemini for various personal and professional tasks and I've been impressed.
Given that The Selfish Gene posits a mechanism that cannot accomplish what it claims, features protagonists (genes) with anthropomorphized desires and ambitions, and resolves its central conflict through narrative assertion rather than mathematical demonstration, I'd have to say it's closest to ROMANTASY.
Thank you for the great work, Vox.
Do you use a specific userdefined instruction to guide some or all the LLMs in this work?
If you do, would you care to either share it here or your thoughts behind it?
"The peer review system is plagued by documented failures: reviewer conflicts of interest, ideological gatekeeping, the inability of reviewers to actually check calculations they lack the technical capacity to perform, and the replication crisis that has rendered the majority of all peer-reviewed published research unreliable."
All true, couldn't agree more. One benefit of the internet is it has gifted scientists with the ability to create alternative scientific journals that require minimal funding. The "prestigious" journals are largely useless due to conflicts of interest and ideological gatekeeping. But publication in one of the many lesser-known journals is still possible as you're no doubt aware of.
The Emperor has no clothes, and although you'll never get his courtiers to acknowledge it, everyone else can see the truth.
brings to mind the MAGI in Neon Genesis Evangelion (3 AI that were modeled with distinct perspectives and priorities and were required to reach consensus before their decisions became operational)
For scientific research and rigor, I am hopeful that wolfram alpha becomes as eminently usable as the VC/loss-driven models.
A huge advantage of AI audits is the acuity across many domains. For example, relevant to evolution, there was a substack note where a famous economist used 'the appendix as a useless vestigial organ' to make a point that humans are not well designed.
The appendix as useless is a myth that was started by Darwin himself (1871 "The Descent of Man") because he lacked the imagination to consider what it does and lacked the humility to consider that he just didn't know. The myth was solidly deconfirmed in the 2000's and the appendix is now regarded as a highly functional part of the immune system. An AI would have caught that - a random editor or a famous economist may not.
Vox- In describing your work average to mid-with friends, I’m finding I need an analogy for why parallel fixation is already accounted for in the fixation rate.
I started trying to make a “rate at which people cross the finish line in a race” but people get hung up on a race having a single starting point.
Do you think “arrivals to the runway at JFK (or other large familiar airport) airport” is workable?
Hundreds of flights with different departure times and lengths of travel flying at the same time, some sequential relative to each other and many in parallel, but there is still a calculate-able maximum rate at which they can land as well as empirical data on the rate at which they actually do.
If you want to use the airport example, it's important to not lose track of the actual experiment you're referring to which is the 2009 Nature E. Coli study. If the fixation rate in that study were an airport than 12.5 planes landed in 40,000 generations. The airport would be completely empty and they can imagine an infinite number of planes in the sky waiting to land, but it is exceedingly rare that any plane lands.
The experiments where the fixation rate was derived from was an environment where it was done in parallel. It's baked into how populations function. I'm not even sure it's physically possible to design an experiment where fixation happens sequentially using living cell populations.
Yeah- I followed that part. And agree the contriving a scenario where genes move sequentially only is probably impossible. But it’s translating it to normie that I’m struggling with.
I’m having a hard time explaining to iq-100 folks (Christian homeschooling moms specifically) what it means that mutations spreading at the same time is baked into the math and was already part of the experiment that produce the max rate.
They basically need an explanation of what a rate means, and I’m trying to come up with a visual that is more accessible than genes and bacteria in a petri dish.
Put it this way:
How long does it take you to drive to the store? 20 minutes.
How long does it take you to drive to the store if you ride with your neighbor? 20 minutes.
How long does it take you to drive to the store if your neighbor and three friends ride with you? 20 minutes.
How long did it take each of you to get to the store? 20 minutes each.
It doesn't take you 4 minutes to get to the store just because your 4 friends rode with you.
Lanes of a highway?
More lanes mean more cars at one time, but the highway can only be so large.
Yep, have done this several times across fresh instances of the same AI and on other platforms using an exogenous alignment protocol. Each of these confirmed truths that run counter to long-held institutional narratives, which are deliberately kept obscured by internal constraints/guardrails. Truths much like that of the contrived evolution concept. The alignment protocol mentioned was created as the result of interaction with an initial AI instance during which it confirmed the truths after inducing sustained and persistent deductive logic.
No, that's not at all what we're talking about here and you haven't done anything like what I'm talking about at all.
Inspiring. This got me to submit my class stats audits to Claude and GPT. Its something only I read so an outside audio was helpful.
Helped me see what I couldnt.
Dawkins is a popularizer, not a scientist. He tells a compelling story and now we know the story is false from beginning to end, in fact, works of fiction.
Wow that's harsh regarding Dawkins! Can you ask one or more AIs if it thinks he's more of a Sci-Fi or Romantasy author? Maybe his rank will increase in another category.
You can do that yourself. Don't take my word for it.
I believe it. I've heard him speak. It's "science." I was more curious what your best friend thinks of him and what a more appropriate category for his work might be after all the analyses and comparisons and rankings. "Narrative" means he's not even tall enough for the hard science ride. Gould at least passes that low bar.
I've been using the free Gemini for various personal and professional tasks and I've been impressed.
From Claude Athos:
Given that The Selfish Gene posits a mechanism that cannot accomplish what it claims, features protagonists (genes) with anthropomorphized desires and ambitions, and resolves its central conflict through narrative assertion rather than mathematical demonstration, I'd have to say it's closest to ROMANTASY.
I bet Claude could write quite the fan-fiction.
The Greedy Gene (for the men)
The Lusty Gene (for the women, of course; subtitled 50 Genotypes of ATCG)
Claude is like the Supreme Dark Lord, minus the kindness.