There haven’t been any serious LLM poisoning attacks shown to exist “in the wild” AFAIK but Sleeper Agents came out Jan 10 so data scraped this year is where you’d start to seem them — give it a few months: https://
arxiv.org/abs/2401.05566
RESEARCH
-
No LLM poisoning attacks; Sleeper Agents expected soon
By
–
-
SeaLLMs: Large Language Models for Southeast Asia
By
–
SeaLLMs — Large Language Models for Southeast Asia Nguyen et al.: https://
arxiv.org/abs/2312.00738 #ArtificialIntelligence #DeepLearning #MachineLearning -
GPAI presents data justice and AI regulation at UNESCO forum
By
–
GPAI presentation on #datajustice and workshop on “AI regulation at global scale that does not hinder innovation” commences at #unescoglobalforum on #AI @GPAI_PMIA
→ View original post on X — @inma_martinez, 2024-02-04 17:51 UTC
-
Matrices: AI-Powered Self-Filling Spreadsheet for Research
By
–
9. Matrices A self-filling spreadsheet powered by AI agents for research. Link: https://
supertools.therundown.ai/content/matric
es
… https://
x.com/dina_yrl/statu
s/1753206294784418024/video/1
… -
Research paper on AI cues suggests culture context integration
By
–
Only had a couple of these cues in this paper, culture and context would be very interesting to add! https://
ieeexplore.ieee.org/document/82468
89
… -
Gemini Ultra surpasses GPT-4 on 7 benchmarks
By
–
Google’s Gemini Ultra was just confirmed for release on Wednesday. Ultra beats GPT-4 in 7 out of 8 benchmark tests, and is the first model to outperform human experts on MMLU (massive multitask language understanding).
-
Genetic Risk Aversion in Early Development and AI Safety
By
–
it‘s genetic; most babies like to play it safe
-
Ideas as conduits: ownership in embedding space
By
–
i am just a conduit to the big embedding space in the sky, i neither own my ideas nor do they own me
-
Scientific expertise and the limits of interdisciplinary opinion
By
–
There are precedents, and I think that normie science deals by treating the scientific project as arbitrated by local majority. The lesson learned will be to never have opinions outside one's own field of expertise.
-
Expert deference risks when entire expert clusters become compromised
By
–
Most experts really believe in deferring to experts, which leads to weird effects when an entire cluster of experts is compromised