Manchuck#AI will not cure cancer. This is because we have not cured cancer. We train AI models and reward them for successes, not failures. Therefore, AI cannot answer a question for which it has not been rewarded for the right answer.
1/
Greg Donald 🤘@manchuck "AI will never cure cancer" ignores how machine learning is already the primary engine driving modern oncology. AI functions as a force multiplier across every stage of research and treatment processes. Supervised learning models can analyze imaging with much higher consistency than human eyes. This statistically improves early detection rates, and that is currently the most effective way to achieve a clinical cure.
@manchuck Beyond diagnostics, unsupervised learning identifies the specific genetic mutations that make a tumor unique. This enables the creation of personalized mRNA vaccines tailored to an individual patient's immune system.
@manchuck Natural language processing accelerates research by scanning millions of research papers to uncover drug interactions that would take humans decades to find. By optimizing clinical trials and predicting which patients will respond to specific immunotherapies, AI transforms once-terminal diagnoses into manageable or curable conditions.
@manchuck Cancer research is not limited to just reinforcement learning like you describe.
@gregdonald I get that research is complex, and I'm not against AI as a tool to help with research; I'm against AI being marketed as a magic box. I don't see myself entering a prompt: "I have cancer, cure me" in the near future.
@gregdonald I have a criticism regarding the tech interview.
When you ask a person to invert a binary tree, what happens if they come up with something so out-of-the-box that it fundamentally changes the way we think about computing? Will you be able to understand what they did?
We need to always be wary of black boxes that give us answers. The AI companies are making it seem like all we need to do is ask the question and accept its answer.
That is dangerous, and we need to be aware of it
@manchuck I agree with most of this. I would never let an AI Agent near my inbox or credit card or production database.
@manchuck Machine learning models are only black boxes for those who don't seek the knowledge.
https://christophm.github.io/interpretable-ml-book/
@gregdonald I'm curious if you are aware of an AI modelthat discovered a new mutation it wasn't trained to find? My thesis is based on the inability of AI to "Think outside the box".
I have yet to see an AI say "I don't know" until prompted. I'm worried that we will ask: "Find any mutations in this tumor and build an mNRA treatment".
https://www.mdpi.com/1422-0067/25/15/8044
https://www.medrxiv.org/content/10.1101/2024.08.10.24311796v1.full.pdf
https://academic.oup.com/bib/article/26/4/bbaf409/8235626
https://pmc.ncbi.nlm.nih.gov/articles/PMC12547363/
https://ascopubs.org/doi/10.1200/PO-24-00908
You seem to be focused on all the inadequacies of LLMs, while discounting all of machine learning.
@gregdonald I was reading through most of these last night, and yes, they are great use cases for AI and LLMs.
I'm against all the hype. I'm against all the insane investments with no regard to the negative impacts. When Crypto became mainstream, the impact was minimal comparatively. Layoffs did not happen because of crypto.
I was hoping to try to ground all the hyperbole.
Maybe I should have said something like: "AI Alone won't cure cancer"
@gregdonald I also wish the COSMIC had been fully developed a few years ago. It might have helped my mother when her cancer metastasized