AaronParisi @AlwaysParisi

ai research engineer (safety, hcai) terrified of the AI world I previously contributed to Joined September 2014

Tweets

7
Followers

11
Following

15
Likes

4

AaronParisi @AlwaysParisi

5 days ago

I remain optimistic that we can still course-correct on AI safety and ensuring good outcomes for all humans. My concern is that the profit motive is wildly misaligned (if you'll forgive the pun) with truly safe and equitable outcomes. How do we disentangle the two?

1 0 0 11 0

View Details

AaronParisi @AlwaysParisi

6 days ago

Congrats to @mlbileschi_pub and team for making such insane steps towards automating discovery of life saving medicine! deepmind.google/blog/accelerat…

0 0 1 11 0

View Details

AaronParisi @AlwaysParisi

a month ago

A tremendous amount of media attention on the newfound concept of AI Risk (which has been a concern for researchers since pre-LLM ages) is focused on existential risk instead of more salient short term risks like mass unemployment, manipulation, and surveillance...how convenient

0 0 1 20 0

View Details

AaronParisi @AlwaysParisi

2 months ago

All the AI researchers now panicking about the repercussions of AI could have been contributing to researching safety, or alignment, or human-AI effects, instead of contributing to capabilities research. The work to ensure good outcomes unglamorous but necessary.

0 0 2 41 0

View Details

AaronParisi @AlwaysParisi

2 months ago

Idk what I'm doing on here other then saying I'm still around, and very much trying to make sure my tiny contribution to this massive era change is a positive one focused on equitable outcomes for all... aclanthology.org/2025.emnlp-mai… arxiv.org/abs/2605.14097

0 0 1 52 0

View Details

Max Bileschi @mlbileschi_pub

3 years ago

2+2=5? “LLMs are not Robust to Adversarial Arithmetic” a new paper from our team @GoogleDeepMind with @bucketofkets, @culpla, @AlwaysParisi, @gamaleldinfe, @jaschasd, Noah Fiedel TLDR: We ask an LLM to attack itself and find this works extremely well.

2 11 49 11K 13

View Details

nfiedel @nfiedel

4 years ago

Tool Augmented Language Models. abs: arxiv.org/abs/2205.12255 Smaller tool augmented models outperform larger non-augmented models in two domains (thus far), and on out-of-distribution examples. Great collaboration w/@AlwaysParisi and @YaoZhaoAI!