As long as you read the final product a time or two and make it as good as you can from a stylistic perspective, I encourage you to continue using Claude to double check your facts.
As for the problem of imperfect representation, as in the old NASA ocean map, I am reminded of Edgar Degas. Not one of his pictures is a photograph of an exact moment from a ballet— they were all painted in his studio from memory, but they are our only and best record of French ballet from the 1860s and 1870s.
Paul von Zielbauer is involved in ongoing legal proceedings involving allegations of harassment, abuse, and unwanted contact of several women at renown universities and institutions. There are concerns regarding a broader pattern of repeated and inappropriate contact involving Asian women, including women who may bear a physical resemblance to his former wife, Esther Chae https://unicourt.com/case/ca-la23-chae-esther-kyungju-vs-von-zielbauer-paul-george-795646. He reportedly had a contentious relationship with his former wife, who is Korean American, and was subsequently involved in legal proceedings initiated by her. I come from a family with several lawyers, and relevant court records are actively reviewed in California. His posts include screenshots that have been altered, cropped, and edited using image-editing software in ways that manipulate and distort the original context. Given that he derives his income solely from Substack, this raises questions about whether the presentation is primarily intended to generate engagement and ultimately to grow his own business, Aging with Strength, using others’ fame and reputation (yes, like yours @Alsassin ) without a proper degree or training in that field. Given the seriousness of these concerns, I strongly recommend exercising appropriate caution and maintaining professional distance from Paul von Zielbauer until the relevant facts and legal matters have been fully resolved. Anyone considering engaging with or publicly defending him should independently review the available documentation and understand the circumstances before associating their name or professional reputation with the matter. He is currently facing legal action from multiple women in California, including his own Ex-wife, Esther Chae, and appears to be attempting to shift responsibility onto his Substack defenders as a means of deflecting accountability. This is particularly concerning given Substack’s recent reporting, published yesterday morning, regarding a career journalist on Substack facing lawsuits through defamation and online violence. Please do not get your name and reputation contaminated.
It would still be wise to run it past a fluid dynamicist. There's a bit that makes no sense to me:
"Conventional altimeters see along a narrow track and blur out anything below roughly 150 kilometers. SWOT sweeps a wide swath at 2-kilometer native resolution". Range, horizontal, and vertical resolution seem to be in play here but the first part of the sentence is probably saying something about horizontal range and the second (SWOT) half is talking of horizontal resolution rather than range.
I stumbled about the exact same paragraph. This is something only a secondary proof-reader can spot, and coincidentally I've heard that's exactly how the best programs (in radio, TV, etc) out there are run: If any first-time reader/listener/viewer spots anything that isn't clear on the first attempt, it needs to be redone. This, of course, takes time and lots of work and access to a big team or test audience - slightly out of scope for the usual substacker. But maybe this can be simulated with AI as well (e.g. feed the final text into a new chat with clear instructions on what to do and disregarding all possible cross-session memory, so as to start with a fresh pair of digital eyes).
If the AI-enabled, yet human-assembled, narrative is closer to the truth, I’ll take that 5-days a week and twice on Sundays vs. the purely robot-generated work.
There's a lot of trust in Claude implied here. I'm not saying that's unjustified, but I think an explanation of this would round-off the point being made. Similarly, as you frequently point out, there's always expert debate on even the most niche issues, so presumably Claude is 'taking sides' now and then.
Nice post. I had 2 thoughts, one on AI and one on ocean eddies.
On AI, your experience struck me as in line with mine; AI is fantastically knowledgeable (although not entirely reliable), but only moderately, well, intelligent. A useful colleague with a mediocre intellect.
On ocean eddies, for what I understand to be your interests, do you need more than the ocean eddy, “climate”? I would have thought you need the *average* vorticity, or *average* vertical transport, or… That would tell you whether your geo-engineering, performed over long times and large areas of ocean would sequester this or that many Gtons of CO2 per year. Following individual eddies for a week might tell you how this week differs from that average, but is that important?
On your second, ocean eddie, part: What you described are not different approaches. The "climate" you mentioned (giving a long-term average sequestering value) is constructed from being able to track real individual eddies across relevant timescales.
A "climate" model that is not constructed from such real observation is just a model - with no provable link to reality, and therefore useless.
Quico- Good work here, especially checking your work with AI. AI has gotten really good at that in the last 6-10 months.
For those interested in restoring safe atmospheric CO2 levels, I'll add a few notes, which are still in peer review papers my group has submitted.
1. Atmospheric CO2 is the primary cause of global warming. Thus, for climate restoration, it's what we must measure. Ocean carbon is good scientific knowledge to have, but as you said, it's terrifically complicated, and no sane person would use it for predicting global warming--at least not for for many more decades.
2. The geologic record (Antarctic ice cores plus the Mauna Loa record, to be specific) show that 10 largest volcanic eruptions in the last 500 years, cooled the earth by typically 0.5C. Of those ten, only 3 reduced atmospheric CO2 measurably. And those three all appear to have dumped their volcanic ash in regions with frequent ocean eddies (~100-300 km diameter). The other 7 dropped their ash on land or in eddy-free ocean regions.
Conclusion (tentative, with no proposed opposition so far) is that studying atmospheric carbon, which controls the climate, shows us that downwelling eddies bury the carbon deep enough that it's secured for many centuries (or millennia). And that non-downwelling regions simply don't sequester CO2 for any useful duration--because the fish ingest and metabolize about 99% of the carbon back to the surface.
We can measure the atmospheric carbon removal accurately--far more accurately than carbon and currents in the ocean. And ocean science is important to continue for thousands of reasons.
It's funny to see this, because I remember thinking on some of your previous articles that it was clearly AI-generated writing but the ideas seemed good and you were engaging in the comments as clearly yourself which made me think "well alright, it's fine. This plankton Does seem pretty cool and he's doing good work."
I'm personally annoyed by noticeably AI-generated writing and I tend to automatically skim over it unless I put deliberate effort into actually reading what the text is saying. If I'm writing and run some AI checks (It is Very valuable to use AI in editing/fact-checking/research processes), I'll rewrite any necessary changes in my own words. This is mostly just because I want my writing to sound like me but I do think it helps me better understand whatever change I'm making. And of course, while AI has advanced a Lot it's still far from infallible.
For you, I'd say do whatever you think is best? There's definitely speedup from using its output verbatim, and that can trade off with some understanding and possibly with some readership? I'm not sure exactly how many people bounce off ai-generated writing vs like ai writing more vs don't notice, and to what extents. I totally agree that it's likely counterproductive to not use AI at all.
As long as you read the final product a time or two and make it as good as you can from a stylistic perspective, I encourage you to continue using Claude to double check your facts.
As for the problem of imperfect representation, as in the old NASA ocean map, I am reminded of Edgar Degas. Not one of his pictures is a photograph of an exact moment from a ballet— they were all painted in his studio from memory, but they are our only and best record of French ballet from the 1860s and 1870s.
Paul von Zielbauer is involved in ongoing legal proceedings involving allegations of harassment, abuse, and unwanted contact of several women at renown universities and institutions. There are concerns regarding a broader pattern of repeated and inappropriate contact involving Asian women, including women who may bear a physical resemblance to his former wife, Esther Chae https://unicourt.com/case/ca-la23-chae-esther-kyungju-vs-von-zielbauer-paul-george-795646. He reportedly had a contentious relationship with his former wife, who is Korean American, and was subsequently involved in legal proceedings initiated by her. I come from a family with several lawyers, and relevant court records are actively reviewed in California. His posts include screenshots that have been altered, cropped, and edited using image-editing software in ways that manipulate and distort the original context. Given that he derives his income solely from Substack, this raises questions about whether the presentation is primarily intended to generate engagement and ultimately to grow his own business, Aging with Strength, using others’ fame and reputation (yes, like yours @Alsassin ) without a proper degree or training in that field. Given the seriousness of these concerns, I strongly recommend exercising appropriate caution and maintaining professional distance from Paul von Zielbauer until the relevant facts and legal matters have been fully resolved. Anyone considering engaging with or publicly defending him should independently review the available documentation and understand the circumstances before associating their name or professional reputation with the matter. He is currently facing legal action from multiple women in California, including his own Ex-wife, Esther Chae, and appears to be attempting to shift responsibility onto his Substack defenders as a means of deflecting accountability. This is particularly concerning given Substack’s recent reporting, published yesterday morning, regarding a career journalist on Substack facing lawsuits through defamation and online violence. Please do not get your name and reputation contaminated.
It would still be wise to run it past a fluid dynamicist. There's a bit that makes no sense to me:
"Conventional altimeters see along a narrow track and blur out anything below roughly 150 kilometers. SWOT sweeps a wide swath at 2-kilometer native resolution". Range, horizontal, and vertical resolution seem to be in play here but the first part of the sentence is probably saying something about horizontal range and the second (SWOT) half is talking of horizontal resolution rather than range.
(I used to work in aero CFD a long time ago....).
I stumbled about the exact same paragraph. This is something only a secondary proof-reader can spot, and coincidentally I've heard that's exactly how the best programs (in radio, TV, etc) out there are run: If any first-time reader/listener/viewer spots anything that isn't clear on the first attempt, it needs to be redone. This, of course, takes time and lots of work and access to a big team or test audience - slightly out of scope for the usual substacker. But maybe this can be simulated with AI as well (e.g. feed the final text into a new chat with clear instructions on what to do and disregarding all possible cross-session memory, so as to start with a fresh pair of digital eyes).
If the AI-enabled, yet human-assembled, narrative is closer to the truth, I’ll take that 5-days a week and twice on Sundays vs. the purely robot-generated work.
There's a lot of trust in Claude implied here. I'm not saying that's unjustified, but I think an explanation of this would round-off the point being made. Similarly, as you frequently point out, there's always expert debate on even the most niche issues, so presumably Claude is 'taking sides' now and then.
Nice post. I had 2 thoughts, one on AI and one on ocean eddies.
On AI, your experience struck me as in line with mine; AI is fantastically knowledgeable (although not entirely reliable), but only moderately, well, intelligent. A useful colleague with a mediocre intellect.
On ocean eddies, for what I understand to be your interests, do you need more than the ocean eddy, “climate”? I would have thought you need the *average* vorticity, or *average* vertical transport, or… That would tell you whether your geo-engineering, performed over long times and large areas of ocean would sequester this or that many Gtons of CO2 per year. Following individual eddies for a week might tell you how this week differs from that average, but is that important?
On your second, ocean eddie, part: What you described are not different approaches. The "climate" you mentioned (giving a long-term average sequestering value) is constructed from being able to track real individual eddies across relevant timescales.
A "climate" model that is not constructed from such real observation is just a model - with no provable link to reality, and therefore useless.
The error (erring) is the best indication of the human intervention (in particular as AI hallucination is going away)
Quico- Good work here, especially checking your work with AI. AI has gotten really good at that in the last 6-10 months.
For those interested in restoring safe atmospheric CO2 levels, I'll add a few notes, which are still in peer review papers my group has submitted.
1. Atmospheric CO2 is the primary cause of global warming. Thus, for climate restoration, it's what we must measure. Ocean carbon is good scientific knowledge to have, but as you said, it's terrifically complicated, and no sane person would use it for predicting global warming--at least not for for many more decades.
2. The geologic record (Antarctic ice cores plus the Mauna Loa record, to be specific) show that 10 largest volcanic eruptions in the last 500 years, cooled the earth by typically 0.5C. Of those ten, only 3 reduced atmospheric CO2 measurably. And those three all appear to have dumped their volcanic ash in regions with frequent ocean eddies (~100-300 km diameter). The other 7 dropped their ash on land or in eddy-free ocean regions.
Conclusion (tentative, with no proposed opposition so far) is that studying atmospheric carbon, which controls the climate, shows us that downwelling eddies bury the carbon deep enough that it's secured for many centuries (or millennia). And that non-downwelling regions simply don't sequester CO2 for any useful duration--because the fish ingest and metabolize about 99% of the carbon back to the surface.
We can measure the atmospheric carbon removal accurately--far more accurately than carbon and currents in the ocean. And ocean science is important to continue for thousands of reasons.
It's funny to see this, because I remember thinking on some of your previous articles that it was clearly AI-generated writing but the ideas seemed good and you were engaging in the comments as clearly yourself which made me think "well alright, it's fine. This plankton Does seem pretty cool and he's doing good work."
I'm personally annoyed by noticeably AI-generated writing and I tend to automatically skim over it unless I put deliberate effort into actually reading what the text is saying. If I'm writing and run some AI checks (It is Very valuable to use AI in editing/fact-checking/research processes), I'll rewrite any necessary changes in my own words. This is mostly just because I want my writing to sound like me but I do think it helps me better understand whatever change I'm making. And of course, while AI has advanced a Lot it's still far from infallible.
For you, I'd say do whatever you think is best? There's definitely speedup from using its output verbatim, and that can trade off with some understanding and possibly with some readership? I'm not sure exactly how many people bounce off ai-generated writing vs like ai writing more vs don't notice, and to what extents. I totally agree that it's likely counterproductive to not use AI at all.