The internet was a mistake.

Our modems used to scream to warn us.

@yokofakun "Don't believe everything you read on the Internet"

Abraham Lincoln

I feel like technology, especially software, have become what derivatives are to the actual economy.

It has moved on to create its own universe that theoretically has roots in the real world but in reality is completely detached from it.

People are no longer thinking in terms of real problems and solutions to them. Instead, they are making stuff up just to participate in the ecosystem.

And everyone has convinced themselves that this is progress.

PhD students, come join us! CAFA is the premier community ML competition for protein function prediction models. If you are interested in applying machine learning to protein function prediction, we are the place to be. http://friedberglab.net for more.

PhD students, come join us! CAFA is the premier community ML competition for protein function prediction models. If you are interested in applying machine learning to protein function prediction, we are the place to be. http://friedberglab.net for more.

parents PLEASE check your kid's candy this year. i just found the critically acclaimed MMORPG Final Fantasy XIV in a Snickers bar. With an expanded free trial which you can play through the entirety of A Realm Reborn and the award-winning Heavensward expansion up to level 60 for free with no restrictions on playtime.

reformulating some dynamic programming as a terrible dream

An Ode To Vim https://bokwoon.com/posts/1khtfep-an-ode-to-vim/

• Notably, some proteins, like PARK7 and TP53, exhibit rapid bursts of annotation growth, but many others appear to have reached a “knowledge saturation” point in the GO database, limiting future annotation gains.

• The study underscores the need for reallocation of research efforts toward understudied proteins to reduce annotation disparities and improve our overall understanding of human biology.

https://www.biorxiv.org/content/10.1101/2024.10.18.619148v1

• Analysis shows a Gini coefficient above 0.6 for annotation inequality, highlighting that the uneven distribution of function knowledge has not improved significantly between 2013 and 2022.

• Another critical observation is that high-throughput experiments generate less informative annotations, while low-throughput studies contribute more meaningful, specific insights into protein function.

• Despite various initiatives, including the “Illuminating the Druggable Genome” project, biases continue to favor well-known proteins, with new annotations primarily assigned to proteins that were already heavily studied.

• The study demonstrates a disconnect between research interest and functional knowledge. Proteins with high publication counts do not always gain corresponding new annotations, pointing to a curation bottleneck.

• Using metrics like term count, unique term count, and information content from Gene Ontology (GO) annotations, the authors quantified the “knowledge wealth” of proteins over the last decade.

• A key finding is the persistently high inequality: a small fraction of proteins receive the majority of research attention, while many remain poorly annotated or entirely unstudied.

New study from our lab!

(Source: https://twitter.com/BiologyAIDaily/status/1848909775649845289)

A Longitudinal Analysis of Function Annotations of the Human Proteome Reveals Consistently High Biases

• This study introduces an economic framework to analyze disparities in protein function annotations, comparing the human proteome’s knowledge distribution to wealth inequality.

https://www.biorxiv.org/content/10.1101/2024.10.18.619148v1

https://simonwillison.net/2024/Oct/15/chatgpt-horoscopes/

The seminar on procrastination has been postponed.

xargs is my favourite recent command line discovery. You can use xargs to avoid having to write a loop to call a command on a large number of files. So let's say you want to convert a folder of TTFs to WOFF2. Will shell globbing work?

$ woff2_compress *.ttf
One argument, the input filename, must be provided.

Nope. It must be called with just one file path. With xargs, that's as easy as:

$ ls *.ttf | xargs -n1 woff2_compress

Boom!
#commandline #unix #linux

It's really effing obvious LLMs are a con trick:

If LLMs were actually intelligent, they would be able to just learn from each other and would get better all the time. But what actually happens if LLMs only learn from each other is their models collapse and they start spouting gibberish.

LLMs depend entirely on copying what humans write because they have no ability to create anything themselves. That's why they collapse when you remove their access to humans.

There is no intelligence in LLMs, it's just repackaging what humans have written without their permission. It's stolen human labour.

#LLM #LLMs #AI #AIs #ChatGPT

Practically no talk on how expertsourcing / crowdsourcing enabled the Nobel prize in chemistry. Over the past 30 years, CASP has been the vehicle driving the advances in protein structure prediction, and which enabled both AF and Rosetta.

https://predictioncenter.org/

Me, celebrating a grant funded after 4 years of submissions.

«if a software team had the earnest desire to make every single user happy by implementing every single feature request, motivated not by profit but entirely by love for their users, they would create a user experience of forbidding complexity, no matter how useful each individual feature would be in its own right.»

I feel like I know some interfaces that have that origin story 😀😂
https://theluddite.org/#!post/rationing