ben
@clearnoise.bsky.social
Scientist
Formerly at Pitt Med, HHMI Janelia and Umass Med, but have jumped head first from the ivory tower
cryo-Electron Microscopy &&
HPC aficionado
OpenAccess (mostly) Free Software
Formerly at Pitt Med, HHMI Janelia and Umass Med, but have jumped head first from the ivory tower
cryo-Electron Microscopy &&
HPC aficionado
OpenAccess (mostly) Free Software
One surprise I’ve encountered while watching the internet drown in #LLM garbage, is a renewed appreciation for books.
I’ve always maintained a fairly well stocked library; but now I’m noticing that I take more time to relish the flow of thought, unique and flavored by the given author.
1/2
I’ve always maintained a fairly well stocked library; but now I’m noticing that I take more time to relish the flow of thought, unique and flavored by the given author.
1/2
May 17, 2024 at 8:48 PM
One surprise I’ve encountered while watching the internet drown in #LLM garbage, is a renewed appreciation for books.
I’ve always maintained a fairly well stocked library; but now I’m noticing that I take more time to relish the flow of thought, unique and flavored by the given author.
1/2
I’ve always maintained a fairly well stocked library; but now I’m noticing that I take more time to relish the flow of thought, unique and flavored by the given author.
1/2
Reposted by ben
Reminder the switch in terminology from "refereeing" to "peer review" in the 1970s was partly due to concern about government interference journals.plos.org/plosbiology/...
April 17, 2024 at 10:26 PM
Reminder the switch in terminology from "refereeing" to "peer review" in the 1970s was partly due to concern about government interference journals.plos.org/plosbiology/...
Can we start a rumor that chatGPT uses “elucidate” at a high rate and it is a trigger for review of a.i. gen content?
#academia #academicchatter
#academia #academicchatter
April 11, 2024 at 1:28 AM
Can we start a rumor that chatGPT uses “elucidate” at a high rate and it is a trigger for review of a.i. gen content?
#academia #academicchatter
#academia #academicchatter
#cryoEM scientist 10 or more years ago - “I have to write an entire software package to do my research, okay then!”
Kids these days - “I have to install a new package and follow the simple il instructions someone took time to share on a mailing list?! How unfair!”
Kids these days - “I have to install a new package and follow the simple il instructions someone took time to share on a mailing list?! How unfair!”
April 9, 2024 at 2:53 PM
#cryoEM scientist 10 or more years ago - “I have to write an entire software package to do my research, okay then!”
Kids these days - “I have to install a new package and follow the simple il instructions someone took time to share on a mailing list?! How unfair!”
Kids these days - “I have to install a new package and follow the simple il instructions someone took time to share on a mailing list?! How unfair!”
Brains do not come with gdb stubs. All we can do is printf debugging (aka talk about stuff).
HT: derf
HT: derf
April 4, 2024 at 10:23 AM
Brains do not come with gdb stubs. All we can do is printf debugging (aka talk about stuff).
HT: derf
HT: derf
Diving Grand Cayman in a week and can’t wait to listen a little closer : )
HT Albert Cardona
#ScienceOutreach #Fish #scuba
youtu.be/POITH02VVrw?...
HT Albert Cardona
#ScienceOutreach #Fish #scuba
youtu.be/POITH02VVrw?...
Fish Sounds: Do fish talk to each other? | Earth Unplugged
Did you know fish chirp, crackle and whoop?! Hear why fish make these surprising sounds as Professor Steve Simpson reveals how technology is uncovering a hidden world of communication on our coral reefs. Subscribe: http://bit.ly/SubscribeToEarthUnplugged Animal Slow Motion: http://bit.ly/EarthUnpluggedSlowMotionTimelapse Expeditions: http://bit.ly/EarthUnpluggedExpeditionsAndLocationFilms Big Questions with Maddie: http://bit.ly/BigQuestionsWithMaddieMoate Wilderness Sessions: http://bit.ly/WildernessSessionsFilms Ocean X Media are a team of scientists, explorers and filmmakers driven to discover what lies beneath the waves and to document untold ocean stories. You can find out more here: http://www.oceanx.org Welcome to Earth Unplugged! We make films about the incredible natural world, we investigate the conundrums, quirks and beautiful science of our amazing planet. Want to share your views with the team behind BBC Earth and win prizes? Join our BBC Studios Voice: https://www.bbcstudiosvoice.com/register This is a channel from BBC Studios who help fund new BBC programmes. Service information and feedback: http://bbcworldwide.com/vod-feedback--contact-details.aspx
youtu.be
March 29, 2024 at 3:16 PM
Diving Grand Cayman in a week and can’t wait to listen a little closer : )
HT Albert Cardona
#ScienceOutreach #Fish #scuba
youtu.be/POITH02VVrw?...
HT Albert Cardona
#ScienceOutreach #Fish #scuba
youtu.be/POITH02VVrw?...
FWIW, i tried to short DJT but there were no available shares at $70 on IBKR
March 27, 2024 at 4:02 PM
FWIW, i tried to short DJT but there were no available shares at $70 on IBKR
An important reminder for all of us freelancers:
If I fix your #gpu #code in an hour, that’s because I’ve spent 13 years learning how to do it in an hour.
You owe me for the years, not the hours.
#consulting
If I fix your #gpu #code in an hour, that’s because I’ve spent 13 years learning how to do it in an hour.
You owe me for the years, not the hours.
#consulting
March 26, 2024 at 3:39 PM
An important reminder for all of us freelancers:
If I fix your #gpu #code in an hour, that’s because I’ve spent 13 years learning how to do it in an hour.
You owe me for the years, not the hours.
#consulting
If I fix your #gpu #code in an hour, that’s because I’ve spent 13 years learning how to do it in an hour.
You owe me for the years, not the hours.
#consulting
When doing subtomo PCA in #cryoET, the MW bias can be seen in the latent space mapping
If you can see the bias in the eigenvectors, it sure seems like a directional FSC with the e-vectors could allow you to ignore or regularize them prior to projecting feature vectors.
Anyone in #TeamTomo tried?
If you can see the bias in the eigenvectors, it sure seems like a directional FSC with the e-vectors could allow you to ignore or regularize them prior to projecting feature vectors.
Anyone in #TeamTomo tried?
March 25, 2024 at 12:57 PM
In explaining my research/funding model to some tenured professors recently, one of the more interesting replies was -
“Oh, so you’re kinda like a research mercenary?”
In so far as I left a broken #academic system to do what I’m *really* good at
AND 👏 get 👏 paid 👏
I guess so
#PostdocExodus
“Oh, so you’re kinda like a research mercenary?”
In so far as I left a broken #academic system to do what I’m *really* good at
AND 👏 get 👏 paid 👏
I guess so
#PostdocExodus
March 25, 2024 at 12:43 PM
In explaining my research/funding model to some tenured professors recently, one of the more interesting replies was -
“Oh, so you’re kinda like a research mercenary?”
In so far as I left a broken #academic system to do what I’m *really* good at
AND 👏 get 👏 paid 👏
I guess so
#PostdocExodus
“Oh, so you’re kinda like a research mercenary?”
In so far as I left a broken #academic system to do what I’m *really* good at
AND 👏 get 👏 paid 👏
I guess so
#PostdocExodus
1/3
Some server math, thinking out loud.
A #gpu running for 50% of a year, with average electricity cost of $0.23/kWh* costs about the same to run as the number of watts used.
I limit my cards to 300W because it cost only 5% performance vs saving 25% elec.
So that is ~$300/ gpu year.
Some server math, thinking out loud.
A #gpu running for 50% of a year, with average electricity cost of $0.23/kWh* costs about the same to run as the number of watts used.
I limit my cards to 300W because it cost only 5% performance vs saving 25% elec.
So that is ~$300/ gpu year.
March 22, 2024 at 11:53 AM
1/3
Some server math, thinking out loud.
A #gpu running for 50% of a year, with average electricity cost of $0.23/kWh* costs about the same to run as the number of watts used.
I limit my cards to 300W because it cost only 5% performance vs saving 25% elec.
So that is ~$300/ gpu year.
Some server math, thinking out loud.
A #gpu running for 50% of a year, with average electricity cost of $0.23/kWh* costs about the same to run as the number of watts used.
I limit my cards to 300W because it cost only 5% performance vs saving 25% elec.
So that is ~$300/ gpu year.
Reposted by ben
Please note that this is not new: it basically is Bayesian Polishing from #RELION. Academics invent these algorithms, which get picked up by commercial companies. This is great, as #OpenSoftwareAcceleratesScience, but don't forget where it all came from.
pubmed.ncbi.nlm.nih.gov/30713699/
pubmed.ncbi.nlm.nih.gov/30713699/
The new reference-based motion correction in cryoSPARC is great! Processing for 110 kDa protein, difficult dataset to process due to the small particle with no defining features, so was hard to align. Left is before RBMC, Right is after.
December 15, 2023 at 2:01 PM
Please note that this is not new: it basically is Bayesian Polishing from #RELION. Academics invent these algorithms, which get picked up by commercial companies. This is great, as #OpenSoftwareAcceleratesScience, but don't forget where it all came from.
pubmed.ncbi.nlm.nih.gov/30713699/
pubmed.ncbi.nlm.nih.gov/30713699/
Gets a new #cryoET data set from collaborators -
“wow, this ice looks great…I can even see the receptors in the surface proximal membranes.”
Measures thickness - 320 nm
It’s no wonder the OG data wasn’t getting past 1 nm resolution. Musta been pushing 1/2 μm
#IDontLikeThickIceAndICannotLie 🎶
“wow, this ice looks great…I can even see the receptors in the surface proximal membranes.”
Measures thickness - 320 nm
It’s no wonder the OG data wasn’t getting past 1 nm resolution. Musta been pushing 1/2 μm
#IDontLikeThickIceAndICannotLie 🎶
March 21, 2024 at 11:39 AM
Gets a new #cryoET data set from collaborators -
“wow, this ice looks great…I can even see the receptors in the surface proximal membranes.”
Measures thickness - 320 nm
It’s no wonder the OG data wasn’t getting past 1 nm resolution. Musta been pushing 1/2 μm
#IDontLikeThickIceAndICannotLie 🎶
“wow, this ice looks great…I can even see the receptors in the surface proximal membranes.”
Measures thickness - 320 nm
It’s no wonder the OG data wasn’t getting past 1 nm resolution. Musta been pushing 1/2 μm
#IDontLikeThickIceAndICannotLie 🎶
Is your code in the stack?
Several of my repos are…for now. They do provide an opt-out via issue creation on #GitHub
Based on what they are showing, they only scraped my more permissively licensed #opensource code and not any #gpl ’ed work
#machineLearning
huggingface.co/spaces/bigco...
Several of my repos are…for now. They do provide an opt-out via issue creation on #GitHub
Based on what they are showing, they only scraped my more permissively licensed #opensource code and not any #gpl ’ed work
#machineLearning
huggingface.co/spaces/bigco...
Am I in The Stack? - a Hugging Face Space by bigcode
Discover amazing ML apps made by the community
huggingface.co
March 21, 2024 at 10:55 AM
Is your code in the stack?
Several of my repos are…for now. They do provide an opt-out via issue creation on #GitHub
Based on what they are showing, they only scraped my more permissively licensed #opensource code and not any #gpl ’ed work
#machineLearning
huggingface.co/spaces/bigco...
Several of my repos are…for now. They do provide an opt-out via issue creation on #GitHub
Based on what they are showing, they only scraped my more permissively licensed #opensource code and not any #gpl ’ed work
#machineLearning
huggingface.co/spaces/bigco...
"Eight years ago Hinton suggested that it was no longer worth training radiologists, (….) His error was not in his assessment of the way AI would develop, but rather (…) to understand the way a profession evolves as certain tasks are automated.”
#AI #machineLearning
www.lrb.co.uk/the-paper/v4...
#AI #machineLearning
www.lrb.co.uk/the-paper/v4...
Paul Taylor · Llamas, Pizzas, Mandolins: AI Doomerism
A great deal of importance seems to be attached to preventing models from generating harmful content if used...
www.lrb.co.uk
March 20, 2024 at 11:23 AM
"Eight years ago Hinton suggested that it was no longer worth training radiologists, (….) His error was not in his assessment of the way AI would develop, but rather (…) to understand the way a profession evolves as certain tasks are automated.”
#AI #machineLearning
www.lrb.co.uk/the-paper/v4...
#AI #machineLearning
www.lrb.co.uk/the-paper/v4...
Just getting setup on bsky - I’ve been hiding out on mastodon since elno took over there.
Looking forward to connecting back to a larger #cryoEM #cryoET and #opensource community here.
Looking forward to connecting back to a larger #cryoEM #cryoET and #opensource community here.
March 19, 2024 at 2:20 PM
Just getting setup on bsky - I’ve been hiding out on mastodon since elno took over there.
Looking forward to connecting back to a larger #cryoEM #cryoET and #opensource community here.
Looking forward to connecting back to a larger #cryoEM #cryoET and #opensource community here.