Andreas Grivas
LightNews LightNews
banner
andreasgrv.bsky.social
Andreas Grivas
@andreasgrv.bsky.social
1K followers 540 following 38 posts
Postdoc in ML/NLP at the University of Edinburgh.

Interested in Bottlenecks in Neural Networks; Unargmaxable Outputs.

https://grv.unargmaxable.ai/
Posts Replies Media Videos
andreasgrv.bsky.social
github.com/pytorch/pyto...

Also related and an interesting read, thanks for sharing!
cure unnecessary NaN values that arise from ±∞ arguments · Issue #49724 · pytorch/pytorch
🐛 Bug First problem logaddexp (⊕) and its generalization logsumexp are supposed to be associative operations. Indeed ((-∞ ⊕ -∞) ⊕ 2) returns the same forward value as (-∞ ⊕ (-∞ ⊕ 2)) or logsumexp(-...
github.com
November 4, 2025 at 9:46 AM
andreasgrv.bsky.social
I would grab this opportunity and go full VC; I would rephrase the above as multi-label classification with labels: possible / sensical / dystopian and require that the candidates shatter the AI startups
October 28, 2025 at 9:30 PM
Reposted by Andreas Grivas
olil.bsky.social
Servers are back online
A clear view of Edinburgh castle
October 20, 2025 at 11:39 AM
andreasgrv.bsky.social
The one I haven't decoded yet is why it would say bee in Italian, maybe it was a warning.
September 25, 2025 at 12:57 PM
andreasgrv.bsky.social
Then, 5 minutes before the ICLR deadline, it surprised me with motivation in Greek (Ela! means Come on!).
September 25, 2025 at 12:57 PM