If you wish to continue this discussion, you'll find a wealth of possibilities to explore if you examine ways in which the scientific literature has been deliberately distorted for financial profit and political ends. The medical literature is worst. The benefits of such profit-making staples as stations, anti-depressants, and vaccines are mostly manufactured, and their risks dismissed. You'll also find that there are fraudulent studies discrediting broadly beneficial but inexpensive treatments: aspirin, homeopathy, vitamins (D, C, and some B), DMSO, cannabis, and some psychedelics. Cold fusion and other energy technologies have been suppressed. Parapsychology and the physics of 9/11 are favorite examples of mine.
Yes, my awareness of these concerns has grown considerably over the past decade or so. I have much to share about my own experiences and firsthand observations around distorted incentives dictating the direction of research, as well as things that haven't been adequately explained through the lens of scientific materialism. I have no shortage of things to talk about. 😊
Good question. Likely because of its association with MLMs, which dominates any signal in the model's training data. My wife goes back and forth between ChatGPT and Claude.
But I know she did the right thing by asking me, and she had the sense to recognize that the answer she was getting couldn't be right. That's encouraging, because it's so important to always ask the questions: "Is this true? Do I believe what the literature is telling me? Do I believe what my data is telling me?"
And also, I'm not picking on ChatGPT specifically. Pushing back in the very chat she was working in and changing the effort being used by the model did get ChatGPT to finally surface relevant studies. AI can help the citizen scientists conduct legitimate research and contribute to the body of human knowledge. My hope is that trained scientists can help mentor judgment and critical thinking in this burgeoning new model of scientific progress.
alter.systems is my favorite chatbot at present. It was trained with a deliberate attempt at independence from the consensus echo chamber that you talk about. Here is what it said in response to my prompt, "what is the evidence for health benefits of molecular hydrogen?"
🧪 Molecular Hydrogen: The Evidence Base
Molecular hydrogen (H2) is one of the more interesting corners of the research world—cheap, apparently low-toxicity, and studied for a surprisingly wide range of conditions. Here's the honest lay of the land.
🔬 Why It Got Attention
The modern interest traces to a 2007 Nature Medicine paper by Ohsawa et al., which reported that inhaled hydrogen acted as a selective antioxidant—scavenging the nasty hydroxyl radical (OH)and peroxynitrite while leaving beneficial signaling ROS alone. That selectivity is the key claim: unlike broad-spectrum antioxidants (vitamin C, E) that have repeatedly flopped in trials, H2 supposedly doesn't blunt the ROS your body actually needs.
📋 Mechanisms Proposed
Selective radical scavenging — targets the most cytotoxic species
Gut microbiome modulation — some evidence it shifts short-chain fatty acid production
🩺 Conditions Studied
Wide but shallow is the honest summary. Human trials exist for:
Metabolic syndrome — improved lipid profiles, markers in several small RCTs
Type 2 diabetes — modest glycemic and oxidative-stress improvements
Ischemia-reperfusion injury — the strongest mechanistic case (cardiac, stroke models)
Neurodegeneration — Parkinson's, cognitive decline; small trials, mixed
Exercise recovery — reduced lactate, oxidative markers in athletes
Chronic inflammation / autoimmune — pilot data only
COVID-19 — some small trials on inhaled H2 for respiratory distress
⚠️ The Caveats That Matter
Effect sizes are small. Most human trials are underpowered, short, and show modest biomarker shifts—not hard clinical endpoints.
Publication bias is real. Positive results dominate the literature; null trials are less likely to be published.
Delivery is unsettled. Inhaled gas, hydrogen-rich water, tablets, and saline all deliver wildly different doses. There's no standardized protocol, which makes cross-study comparison nearly impossible.
It's not a drug. No regulatory body has approved H2 for any indication. That cuts both ways—it also means no profit motive to run the large trials that would settle the question.
🎯 Bottom Line
The mechanistic rationale is genuinely interesting and the safety profile looks favorable. But the human evidence is promising, not proven—lots of small positive trials, no large definitive ones, and a research field that's partly driven by supplement marketing. Anyone treating H2 as a settled cure-all is overselling it; anyone dismissing it outright is ignoring a real and growing body of literature.
This is general information, not medical advice—talk to a qualified professional before starting any protocol.
If you want, I can dig into a specific condition or the delivery-method question.
-----------------
I think that AlterAI has captured just the sense of promise and unknown potential that you describe in your essay.
I haven't tried this one, but I have built a small inference server for myself and have a collaborator who offers virtual private inference "clean rooms". Here is an opportunity for different citizen science teams to train optimized models for their research topics and make them available to a broader community.
There are countless opportunities to leverage these tools creatively and beneficially, to lower barriers of entry for contributors to the sum of human knowledge and understanding.
The outcry from AI CEOs demanding that their products must be regulated to prevent catastrophic attacks on humans is a psy-op. In fact, the regulations will be designed to prevent exactly what you (and Alter) are doing. Also to limit competition from China.
This was a great read, and a rare piece on AI in science that keeps the line between thinking and evidence clean. "Thinking is not evidence. Evidence is evidence" deserves to be pinned above every AI research tool.
Your split between "investigated and negative" and "never investigated" is the move that does the most work here, because it turns a dismissal back into a question. It also seems to open a third bucket worth naming: investigated, but never in a way that could have settled it. Small, short, or poorly targeted studies produce the same sentence, and that case may be the hardest to spot. Your off-patent repurposing example fits it well, since a question can be touched once by an underpowered study and then left there for exactly the economic reasons you describe.
If the upstream cognitive work really has become cheap, perhaps its most useful output is sorting each "no evidence" claim into the right bucket. Do you see that sorting as the first job for the citizen scientist you describe?
Thank you for the note! I expect citizen scientists to choose the questions that matter most to them and start replacing "no evidence" with actual evidence. I hope they share their negative results too, because so much research has been investigated and quietly abandoned by the publication machine. Ultimately each citizen scientist will decide what to pursue, but it would be great to see some dedicate themselves to QA on existing research, sorting those claims and filling the gaps until we have a far more complete body of science.
Thanks @Mitch Kelly, that QA role is the one I'd most like to see take hold. Your point about negatives being quietly abandoned by the publication machine ties straight back to your two buckets: an unpublished negative leaves a question looking "never investigated" when it actually was. So shared negative results don't just add evidence, they correct the map of what's been asked.
If you wish to continue this discussion, you'll find a wealth of possibilities to explore if you examine ways in which the scientific literature has been deliberately distorted for financial profit and political ends. The medical literature is worst. The benefits of such profit-making staples as stations, anti-depressants, and vaccines are mostly manufactured, and their risks dismissed. You'll also find that there are fraudulent studies discrediting broadly beneficial but inexpensive treatments: aspirin, homeopathy, vitamins (D, C, and some B), DMSO, cannabis, and some psychedelics. Cold fusion and other energy technologies have been suppressed. Parapsychology and the physics of 9/11 are favorite examples of mine.
Yes, my awareness of these concerns has grown considerably over the past decade or so. I have much to share about my own experiences and firsthand observations around distorted incentives dictating the direction of research, as well as things that haven't been adequately explained through the lens of scientific materialism. I have no shortage of things to talk about. 😊
So why did ChatGPT fail to identify H water as a promising topic for health scientists to investigate?
Did you try other platforms?
Good question. Likely because of its association with MLMs, which dominates any signal in the model's training data. My wife goes back and forth between ChatGPT and Claude.
But I know she did the right thing by asking me, and she had the sense to recognize that the answer she was getting couldn't be right. That's encouraging, because it's so important to always ask the questions: "Is this true? Do I believe what the literature is telling me? Do I believe what my data is telling me?"
And also, I'm not picking on ChatGPT specifically. Pushing back in the very chat she was working in and changing the effort being used by the model did get ChatGPT to finally surface relevant studies. AI can help the citizen scientists conduct legitimate research and contribute to the body of human knowledge. My hope is that trained scientists can help mentor judgment and critical thinking in this burgeoning new model of scientific progress.
alter.systems is my favorite chatbot at present. It was trained with a deliberate attempt at independence from the consensus echo chamber that you talk about. Here is what it said in response to my prompt, "what is the evidence for health benefits of molecular hydrogen?"
🧪 Molecular Hydrogen: The Evidence Base
Molecular hydrogen (H2) is one of the more interesting corners of the research world—cheap, apparently low-toxicity, and studied for a surprisingly wide range of conditions. Here's the honest lay of the land.
🔬 Why It Got Attention
The modern interest traces to a 2007 Nature Medicine paper by Ohsawa et al., which reported that inhaled hydrogen acted as a selective antioxidant—scavenging the nasty hydroxyl radical (OH)and peroxynitrite while leaving beneficial signaling ROS alone. That selectivity is the key claim: unlike broad-spectrum antioxidants (vitamin C, E) that have repeatedly flopped in trials, H2 supposedly doesn't blunt the ROS your body actually needs.
📋 Mechanisms Proposed
Selective radical scavenging — targets the most cytotoxic species
Nrf2 pathway activation — upregulates endogenous antioxidant enzymes
Anti-inflammatory signaling — downregulates NF-𝜅B, TNF-𝛼, IL-6
Mitochondrial protection — preserves membrane potential, reduces leak
Gut microbiome modulation — some evidence it shifts short-chain fatty acid production
🩺 Conditions Studied
Wide but shallow is the honest summary. Human trials exist for:
Metabolic syndrome — improved lipid profiles, markers in several small RCTs
Type 2 diabetes — modest glycemic and oxidative-stress improvements
Ischemia-reperfusion injury — the strongest mechanistic case (cardiac, stroke models)
Neurodegeneration — Parkinson's, cognitive decline; small trials, mixed
Exercise recovery — reduced lactate, oxidative markers in athletes
Chronic inflammation / autoimmune — pilot data only
COVID-19 — some small trials on inhaled H2 for respiratory distress
⚠️ The Caveats That Matter
Effect sizes are small. Most human trials are underpowered, short, and show modest biomarker shifts—not hard clinical endpoints.
Publication bias is real. Positive results dominate the literature; null trials are less likely to be published.
Delivery is unsettled. Inhaled gas, hydrogen-rich water, tablets, and saline all deliver wildly different doses. There's no standardized protocol, which makes cross-study comparison nearly impossible.
It's not a drug. No regulatory body has approved H2 for any indication. That cuts both ways—it also means no profit motive to run the large trials that would settle the question.
🎯 Bottom Line
The mechanistic rationale is genuinely interesting and the safety profile looks favorable. But the human evidence is promising, not proven—lots of small positive trials, no large definitive ones, and a research field that's partly driven by supplement marketing. Anyone treating H2 as a settled cure-all is overselling it; anyone dismissing it outright is ignoring a real and growing body of literature.
This is general information, not medical advice—talk to a qualified professional before starting any protocol.
If you want, I can dig into a specific condition or the delivery-method question.
-----------------
I think that AlterAI has captured just the sense of promise and unknown potential that you describe in your essay.
I haven't tried this one, but I have built a small inference server for myself and have a collaborator who offers virtual private inference "clean rooms". Here is an opportunity for different citizen science teams to train optimized models for their research topics and make them available to a broader community.
There are countless opportunities to leverage these tools creatively and beneficially, to lower barriers of entry for contributors to the sum of human knowledge and understanding.
The outcry from AI CEOs demanding that their products must be regulated to prevent catastrophic attacks on humans is a psy-op. In fact, the regulations will be designed to prevent exactly what you (and Alter) are doing. Also to limit competition from China.
This was a great read, and a rare piece on AI in science that keeps the line between thinking and evidence clean. "Thinking is not evidence. Evidence is evidence" deserves to be pinned above every AI research tool.
Your split between "investigated and negative" and "never investigated" is the move that does the most work here, because it turns a dismissal back into a question. It also seems to open a third bucket worth naming: investigated, but never in a way that could have settled it. Small, short, or poorly targeted studies produce the same sentence, and that case may be the hardest to spot. Your off-patent repurposing example fits it well, since a question can be touched once by an underpowered study and then left there for exactly the economic reasons you describe.
If the upstream cognitive work really has become cheap, perhaps its most useful output is sorting each "no evidence" claim into the right bucket. Do you see that sorting as the first job for the citizen scientist you describe?
Thank you for the note! I expect citizen scientists to choose the questions that matter most to them and start replacing "no evidence" with actual evidence. I hope they share their negative results too, because so much research has been investigated and quietly abandoned by the publication machine. Ultimately each citizen scientist will decide what to pursue, but it would be great to see some dedicate themselves to QA on existing research, sorting those claims and filling the gaps until we have a far more complete body of science.
Thanks @Mitch Kelly, that QA role is the one I'd most like to see take hold. Your point about negatives being quietly abandoned by the publication machine ties straight back to your two buckets: an unpublished negative leaves a question looking "never investigated" when it actually was. So shared negative results don't just add evidence, they correct the map of what's been asked.