As a bioinformatics person that's spent time in and out of industry/academia, I agree with some of the article's thesis. While I don't think LLMs or AI are going away, I do think it will allow people in academia to pump out a bunch of inane papers and continue to prop up predatory scientific journal publishing via tenure and promotion. In fact outside of how utterly useless Fable 5 is via their aggressive guard rails for my work, I quite like using statically typed and/or functional languages with other LLMs since there are some baked in guardrails via compiler + type system.
I think the flattening of progress is the most interesting dimension to the article. For an example a useful biological product discovery with a nonlinear path to get to there, look at the Taq polymerase (https://en.wikipedia.org/wiki/Taq_polymerase). Without some NSF funded exploratory ecological research by Tom Brock in Yellowstone Hot Springs to test the theoretical limit of life at high temperatures (https://en.wikipedia.org/wiki/Thermus_aquaticus) we never get to the Taq polymerase, we never get reliable/robust PCR (https://en.wikipedia.org/wiki/Polymerase_chain_reaction), which is now a gold standard method in both clinical and environmental testing! It is rather improbable to think that large language models would associate those domain connections across the topic (molecular biotechnology + ecology + microbial physiology). I also did some exploratory work with text embedding models people might use for RAG and challenged them with an open source scientific MCA question dataset, generalist embedders performed worse vs. domain specific embedders trained on scientific corpora (doesn't surprise me at all). However, if everything regresses to the median of the universe of possible knowledge, it seems like scientific leaning frontier models would get locked into this asymptotic flattening before turning cashflow positive for model vendors OR they become so locked down that only big pharma, state actors, or big ag can afford the API rates and vetting process.
Yeah, I definitely do something similar with my personal projects.
I come from more of a hardware & environmental engineering background and we were always taught that projects were iteratively built via Design, Build, Test, Learn cycles.
I drive the Design and basic skeleton of the build (pseudocode or boilerplate), then pass off the rest of the Build and Test to the agent. I pick up after the test and read the agent commits/notes, then write up next steps. Repeat DBTL. Maybe spin a few features out at a time in parallel depending on how much time I want to devote to reviewing new project features later in the day.
I wonder if the same thing happened with--or is happening at--NSF? I know researchers that did not get funding for quantitative ecology fellowships or grants. After back channeling with program managers, it seems that using "diversity"--as in the quantitative ecological measures, metrics, or derived functional values--may have flagged proposals to be rejected.
Pretty cool bioinformatics algorithm to speed up what was traditionally a dynamic Burrows-Wheeler Transform. Interested to see where this gets implemented outside of benchmarking in the next few years!
It's somewhat interesting, but the authors' conclusions are a bit odd given their data.
They acknowledge that fame is potentially confounding:
Risk factors (impulsivity, substance use, etc.) -> Fame achievement |
Risk factors -> Early mortality
The authors also appear to conclude that fame is semi-causal of the mortality risk. If, taking a causal statistical approach, the authors conditioned on the collider:
Risk factors (substance use, personality traits, mental health vulnerabilities) -> Becoming/staying a professional singer <- Talent/drive toward fame
I do applaud them for preregistering the study, but I think this paper needed a little more rigor in peer review.
This is a non-issue with Polars dataframes to_pandas() method. You get all the performance of Polars for cleaning large datasets, and to_pandas() gives you backwards compatibility with other libraries. However, plotnine is completely compatible with Polars dataframe objects.
> Life's two most fundamental properties are homeostasis and reproduction.
> The loss of these two combined with its parasitic nature makes this cell a form on non-life.
This is a decidedly Eukaryote-centric take. Homeostasis in higher mammals is a complex network of genes -> RNA -> proteins -> metabolic pathways
Reproduction is also far more simple in organisms with binary fission cellular division.
A more appropriate scientific term would be obligate commensalism vs. "parasitic". That actually encapsulates their need for metabolic precursors from the host, but allows for tRNA, rRNA, origin of replication, etc...present in the organism's genome.
For all the folks saying, "Isn't this just a virus?"
The actual paper states that the genome encodes transfer RNA's and ribosomal RNA's. I think that's a really important biological distinction missing from the popular press junket. The primary source material is well written and elucidates a lot more than the Quanta article. https://www.biorxiv.org/content/10.1101/2025.05.02.651781v1
Yeah I think this is definitely the future. Recently, I too have spent considerable time on probabilistic hyper-graph models in certain domains of science. Maybe it _is_ the next big thing.
Countable is a relative term in microbiology. I like that the author stuck to the phrase "countable colonies", since colony forming units are not really "countable as cells".
> Meet MemMachine, an open-source memory layer for advanced AI agents. It enables AI-powered applications to learn, store, and recall data and preferences from past sessions to enrich future interactions. MemMachine's memory layer persists across multiple sessions, agents, and large language models, building a sophisticated, evolving user profile. It transforms AI chatbots into personalized, context-aware AI assistants designed to understand and respond with better precision and depth.
This seems really interesting for local LLM experimentation.
This! The amount of clinicians I know who simply read the abstract of a case study, with no real statistical interpretation of results, is a non-zero number.
Whenever I see some hyped up popular press article about a scientific study, my immediate reaction is to go to the primary literature. First, I read the study design and analysis methods, then I determine if its even worth continuing to read the rest. Study pre-registration should be a must and papers need to be more explicit about being exploratory when the sample size dictates it.
I agree with all of this. I've worked in optical engineering, bioinformatics, and data science writ large for over a decade, knowing the data collection process is foundational to statistical process control and statistical design of experiments. I've watched former employers light cash on fire chasing results from similar methods this MCP runs on the backend due to lack of measurement/experimental context.
I think the flattening of progress is the most interesting dimension to the article. For an example a useful biological product discovery with a nonlinear path to get to there, look at the Taq polymerase (https://en.wikipedia.org/wiki/Taq_polymerase). Without some NSF funded exploratory ecological research by Tom Brock in Yellowstone Hot Springs to test the theoretical limit of life at high temperatures (https://en.wikipedia.org/wiki/Thermus_aquaticus) we never get to the Taq polymerase, we never get reliable/robust PCR (https://en.wikipedia.org/wiki/Polymerase_chain_reaction), which is now a gold standard method in both clinical and environmental testing! It is rather improbable to think that large language models would associate those domain connections across the topic (molecular biotechnology + ecology + microbial physiology). I also did some exploratory work with text embedding models people might use for RAG and challenged them with an open source scientific MCA question dataset, generalist embedders performed worse vs. domain specific embedders trained on scientific corpora (doesn't surprise me at all). However, if everything regresses to the median of the universe of possible knowledge, it seems like scientific leaning frontier models would get locked into this asymptotic flattening before turning cashflow positive for model vendors OR they become so locked down that only big pharma, state actors, or big ag can afford the API rates and vetting process.