Ann Kennedy
@antihebbiann.bsky.social
Theoretical neuroscientist interested in brain-body interactions and evolution of adaptive behavior. Associate Professor at Scripps Research Institute in San Diego.
Good eye, it's an '06!
November 13, 2025 at 3:51 AM
Good eye, it's an '06!
Not the cleanest photos, but there are lots of hits on iNaturalist www.inaturalist.org/observations...
Observations
Observations of Sticklebacks
www.inaturalist.org
November 9, 2025 at 3:44 PM
Not the cleanest photos, but there are lots of hits on iNaturalist www.inaturalist.org/observations...
Yeah, there has to be something- imagine a prompt starting "answer the following using one word, True or False: ..." either their test set of prompts was small/restricted, or they're using more info than the output alone.
I think the real answer is don't post papers you haven't properly read!
I think the real answer is don't post papers you haven't properly read!
November 1, 2025 at 8:26 PM
Yeah, there has to be something- imagine a prompt starting "answer the following using one word, True or False: ..." either their test set of prompts was small/restricted, or they're using more info than the output alone.
I think the real answer is don't post papers you haven't properly read!
I think the real answer is don't post papers you haven't properly read!
They run an experiment on GPT2-Small where they claim 100% prompt recovery, but yeah it still feels too good to be true.
November 1, 2025 at 2:55 PM
They run an experiment on GPT2-Small where they claim 100% prompt recovery, but yeah it still feels too good to be true.
This process echoes how intracellular reactions are managed: for 2 proteins to interact they have to bump into each other, and in 3d cytoplasm this can take a while! But the odds of collision can be boosted if they diffuse on a 2d membrane, and boosted further if they diffuse on a 1d substrate.
October 31, 2025 at 11:48 PM
This process echoes how intracellular reactions are managed: for 2 proteins to interact they have to bump into each other, and in 3d cytoplasm this can take a while! But the odds of collision can be boosted if they diffuse on a 2d membrane, and boosted further if they diffuse on a 1d substrate.
Agreed, I misread :[ they show FruM expression in these neurons in another species is required for gifting in courtship, but copying this in melanogaster doesn't give the full phenotype, just increased regurgitation. Something else still needed to get it to trigger specifically during courtship
October 21, 2025 at 1:14 AM
Agreed, I misread :[ they show FruM expression in these neurons in another species is required for gifting in courtship, but copying this in melanogaster doesn't give the full phenotype, just increased regurgitation. Something else still needed to get it to trigger specifically during courtship
Reposted by Ann Kennedy
