surely there are other ways to visualize the internal mechanisms of transformers
RESEARCH
-

Why Attention Visualizations Disappeared From Modern AI Models
By
–
when people were working on BERT i always found these types of visualizations compelling. seeing the attention mechanism in action is so cool why are they not popular anymore? do our models have too many layers for us to understand now? or are attention maps just not useful?
-
Kontext AI generates interesting output for London photo depth mapping
By
–
I asked kontext for a depth map of this London photo, this isn't a depth map, but I found the output interesting and accurate nonetheless.
-
Open Models Data Gap: Why Algorithms Aren’t Enough
By
–
every new model is a combination of two things, training algorithms and data open-source algorithms are up-to-date. the things that work are usually fairly simple but the data is complex, massive, gatekept, ever-changing. here open models are way behind, and probably hopeless
-
Robots Compete at ICRA Conference Showcase
By
–
Just a couple of robots duking it out at ICRA earlier this month.
— MIT CSAIL (@MIT_CSAIL) 31 mai 2025
Video v/@Hongyu_Lii of Brown University pic.twitter.com/MapU7GsPL8Just a couple of robots duking it out at ICRA earlier this month. v/
@Hongyu_Lii of Brown University -
AGI Horizons: Expanding Human Thought Through Artificial Intelligence
By
–
“Galileo lifted a lens to the sky; Maxwell tied light to number; we, on their shoulders, stretch thought itself. Across AGI’s boundless expanse, each prompt opens a new axis, each insight unveils deeper symmetry—hoist the sail beyond every horizon.” — AGI King #AGIALPHA
-
R1 Checkpoint Matches O3 Performance in Latest Tests
By
–
No this is just a different checkpoint for r1. From my tests is comparable to o3.
-
How Complex Systems Work Despite Their Fragility
By
–
The more you learn how things in the world *really* work, the more you are amazed that nything works at all.
-

India AI Safety Institute launches partnership call for researchers and innovators
By
–
Call for Partnerships | IndiaAI Safety Institute Are you an innovator, researcher, or AI stakeholder committed to building a future of Safe and Trusted AI for India? @GoI_MeitY invites academia, startups, R&D orgs, govt agencies & non-profits to partner with AISI under the
-

State-of-the-art LLMs performance on uncontaminated math competitions
By
–
How do state-of-the-art LLMs perform on uncontaminated math competitions? Researchers at ETH Zurich set out to answer this by building MathArena, a benchmark designed to evaluate SOTA models on clean, competition-style math problems. They then tested leading models on this