@samuelstevens.bsky.social
it doesn't bother me that LLMs cannot reliably do multi-digit addition or multiplication. many of my labmates are worried that LLMs can't do symbolically simple tasks like association, copying, addition, etc. (see my labmate Boshi's work arxiv.org/abs/2504.01928 as an example)
Is the Reversal Curse a Binding Problem? Uncovering Limitations of...
Despite their impressive capabilities, LLMs exhibit a basic generalization failure known as the Reversal Curse, where they struggle to learn reversible factual associations. Understanding why this...
arxiv.org
April 12, 2025 at 4:08 PM
it doesn't bother me that LLMs cannot reliably do multi-digit addition or multiplication. many of my labmates are worried that LLMs can't do symbolically simple tasks like association, copying, addition, etc. (see my labmate Boshi's work arxiv.org/abs/2504.01928 as an example)
it's incredibly inspiring to get to study with Jenna and see her grow into her research community over the years. Congrats on the opportunity to share your work!
Visited Muskingum University this weekend to give a talk at the Conservation Science Symposium hosted by The Wilds. My talk, From Ohio to Kenya: Autonomous Drones in Field Ecology and Conservation, explored how edge AI aids real-time animal detection and tracking.
March 4, 2025 at 2:15 PM
it's incredibly inspiring to get to study with Jenna and see her grow into her research community over the years. Congrats on the opportunity to share your work!
Reposted
Super cool demo! I tested out the segmentation on some drone imagery of zebras I collected in Kenya a few weeks back - it worked really well!
February 26, 2025 at 3:59 PM
Super cool demo! I tested out the segmentation on some drone imagery of zebras I collected in Kenya a few weeks back - it worked really well!
What's actually different between CLIP and DINOv2? CLIP knows what "Brazil" looks like: Rio's skyline, sidewalk patterns, and soccer jerseys.
We mapped 24,576 visual features in vision models using sparse autoencoders, revealing surprising differences in what they understand.
We mapped 24,576 visual features in vision models using sparse autoencoders, revealing surprising differences in what they understand.
February 26, 2025 at 1:12 PM
What's actually different between CLIP and DINOv2? CLIP knows what "Brazil" looks like: Rio's skyline, sidewalk patterns, and soccer jerseys.
We mapped 24,576 visual features in vision models using sparse autoencoders, revealing surprising differences in what they understand.
We mapped 24,576 visual features in vision models using sparse autoencoders, revealing surprising differences in what they understand.
More and more I find myself building one-off tools for individual problems. I think aider + Claude + uv is finally making it possible to do this fluidly on-demand without serious effort.
December 24, 2024 at 8:32 PM
More and more I find myself building one-off tools for individual problems. I think aider + Claude + uv is finally making it possible to do this fluidly on-demand without serious effort.
Reposted
🚨 New Publication Alert!
We deep dive into KABR: a dataset for ungulate behavior recognition from drone footage at Mpala Research Centre, Kenya. 🦒🦓
KABR features reticulated giraffes, plains zebras, and Grévy’s zebras with 10+ hours of annotated video.
Explore: kabrdata.xyz
We deep dive into KABR: a dataset for ungulate behavior recognition from drone footage at Mpala Research Centre, Kenya. 🦒🦓
KABR features reticulated giraffes, plains zebras, and Grévy’s zebras with 10+ hours of annotated video.
Explore: kabrdata.xyz
December 21, 2024 at 6:14 PM
🚨 New Publication Alert!
We deep dive into KABR: a dataset for ungulate behavior recognition from drone footage at Mpala Research Centre, Kenya. 🦒🦓
KABR features reticulated giraffes, plains zebras, and Grévy’s zebras with 10+ hours of annotated video.
Explore: kabrdata.xyz
We deep dive into KABR: a dataset for ungulate behavior recognition from drone footage at Mpala Research Centre, Kenya. 🦒🦓
KABR features reticulated giraffes, plains zebras, and Grévy’s zebras with 10+ hours of annotated video.
Explore: kabrdata.xyz
If you haven't read The Grug Brained Developer, take a couple minutes to check it out (link below). It and "A Philosophy of Software Design" by John Ousterhout are my top two best software design guides.
December 12, 2024 at 5:29 PM
If you haven't read The Grug Brained Developer, take a couple minutes to check it out (link below). It and "A Philosophy of Software Design" by John Ousterhout are my top two best software design guides.