Congratulations Dr. Preusser!
Great defense, I’m very happy we could do this in person at the BIMSB / @MDC_Berlin 😊
@preibischs @royerlab fwiw I think we should be putting our efforts in open source models that you can run locally — these are already very close in performance to ChatGPT and have obvious advantages:
https://lmsys.org/blog/2023-03-30-vicuna/
https://www.semianalysis.com/p/google-we-have-no-moat-and-neither
It was a pleasure to host Loic Royer (CZI Biohub) for a Napari-ChatGPT Code Review @hhmijanelia
It was very exciting and I think watching it will give you a nice introduction to the project, but also how to get started using the #ChatGPT API for your own projects:
The Scientific Computing group at #HHMIJanelia seeks a software engineer, to work on large-scale electron #microscopy reconstruction and analysis. Come work at the forefront of #imaging and #OpenScience with applications in #Neuroscience and #CellBiology. Janelia welcomes and values diversity, and encourages applications from people of all backgrounds, including people of color and members of groups traditionally underrepresented in science. (#FediHire, #FediHired)
https://hhmi.wd1.myworkdayjobs.com/en-US/External/job/Scientific-Computing-Software-Engineer_R-1972
We are looking for a software engineer with background in mathematics and/or image alignment to work with us on reconstructing very large electron microscopy volumes:
Today the peer-reviewed version of our preprint is out:
"The #connectome of an insect brain"
https://www.science.org/doi/10.1126/science.add9330
Congrats to co-first authors Michael Winding and Benjamin Pedigo, and to all our lab members and collaborators who made this work possible over the years. A journey that started over 10 years ago–and yet this is but a new beginning. So much more to come.
See my #tootprint on the preprint from back in the Autumn: https://mathstodon.xyz/@albertcardona/109422190525090990
The data is available both as supplements and directly via #CATMAID thanks to hosting by the #VirtualFlyBrain:
https://l1em.catmaid.virtualflybrain.org/?pid=1&zp=108250&yp=82961.59999999999&xp=54210.799999999996&tool=tracingtool&sid0=1&s0=2.4999999999999996&help=true&layout=h(XY,%20%7B%20type:%20%22neuron-search%22,%20id:%20%22neuron-search-1%22,%20options:%20%7B%22annotation-name%22:%20%22papers%22%7D%7D,%200.6)
(The "Winding, Pedigo et al. 2023" annotation listing all included neurons will appear very soon in an upcoming update.)
Latest report from the #OMENGFF #Community describing #OMEZARR, a #CloudOptimized #BioImaging #DataFormat, is out now on #bioRxiv!
If you are having trouble with big, slow, or otherwise unwieldy imaging data, reach out or take a look for a #FAIR-er option.
Is there a #Java library that can write TIFFs bigger than 2^31 pixels per plane? I thought jai-imageio could do it, but an implementation detail prevents it (https://github.com/jai-imageio/jai-imageio-core/issues/80).
Thanks so much!!
We have an open #postdoc position in #RNA biology. Combine Genomics and proteomics to develop new tools to explore RNA protein interactions. A collaboration with Ilaria Piazza’s lab (MDC Berlin). Come and join us in #Stockholm SciLifeLab Karolinska Institute Deadline 30 Jan. http://pelechanolab.com/2023/01/11/postdoctoral-position-in-rna-biology/
Follow-up update for #BigStitcher is out! Now supports:
1) export of BigDataViewer/BigStitcher compatible HDF5/N5 datasets
2) export of #OpenSeadragon datasets for 2D images and 2D slices of 3D images
QT: https://qoto.org/@preibischs/109611469671364414
Exciting New Year's release of #BigStitcher! It now supports:
1) efficient blocked processing, meaning large, tiled acquisitions are fused blazingly fast (second/minutes instead of hours)!
2) Export of ZARR/N5/HDF5 thanks to the N5-API
Check it out, just update Fiji!
@preibischs interesting experiment which shows, not surprisingly, how the outcome can be biased by the training set. Most of those recipes are quite South German, Bavaria and the Black Forrest area. The type one would see in a American movie depicting German stereotypes.
I do think AI as an enormous potential but I think it's easy to ignore the errors which are only detect by human experts in a particular field.
@keyboardpipette great point, it’s rather famous than great German cuisine ….
I got curious about #ChatGPT and the associated questions related to ownership - so I quickly created a #German #cookbook with outstanding recipes by asking #ChatGPT questions and DALL-E to generate images ... check it out
https://drive.google.com/file/d/17DXeTqENrpP3sHQouPMQ8bujsVYp0Wdp/view?usp=sharing
@konrad_rokicki I guess it might boil down to the question whether outputs of “advanced AI’s” constitute a purely technical or creative process. If, at some point, we decide it’s creative then really tough decisions will arise … these questions already surface now for simple things like (trivial) code snippets (take it from #stackoverflow or an AI?) but it is likely just the beginning …
@LouisIngenthron interesting point. I think the algorithm, yes, but key (or at least a major aspect) is training data, which was not owned by the algorithm creator but was publicly available data, each with its own license.
Assuming this to be a purely technical process, then the algorithm owner should not have the right to all the results of the data.
If it were a creative process, that would be different, at least that’s how we define intellectual property for humans.
Tough decisions ahead…
Open Science Software Enthusiast - Director of Scientific Computing Software & Team Leader Computational Method Development at HHMI Janelia