We write about frontier advances in the life sciences, machine learning, and technology, and how their fusion is reshaping the trajectory of life on Earth.
One of the best recent pieces I’ve read on landscape for developing biological foundations models. It’s comprehensive and generally follows my intuition as well when it comes to modeling real world complexity. Thank you
An incredible depth. Thank you for putting time into it. For platform builders, it's worth going through Bauer's checklist. Makes sense and is detailed.
The part that stopped me was the assumption that noise averages out. That was true enough for text, where bad examples are random. Biological measurements are not. Batch effects and assay differences are structured noise, so filtering and deduplication can remove the exact information that tells you which measurements are comparable. Curation for bio data may need to preserve provenance and assay design more than it needs more examples.
One of the best recent pieces I’ve read on landscape for developing biological foundations models. It’s comprehensive and generally follows my intuition as well when it comes to modeling real world complexity. Thank you
An incredible depth. Thank you for putting time into it. For platform builders, it's worth going through Bauer's checklist. Makes sense and is detailed.
The part that stopped me was the assumption that noise averages out. That was true enough for text, where bad examples are random. Biological measurements are not. Batch effects and assay differences are structured noise, so filtering and deduplication can remove the exact information that tells you which measurements are comparable. Curation for bio data may need to preserve provenance and assay design more than it needs more examples.