jonathan-roberts1/zerobench
Viewer
•
Updated
•
434
•
1k
•
16
Endre Stølsvik
stolsvik
·
AI & ML interests
LLMs, generative
Recent Activity
liked
a dataset
25 days ago
jonathan-roberts1/zerobench
liked
a dataset
9 months ago
HuggingFaceFW/fineweb
reacted
to
santiviquez's
post
with ❤️
about 1 year ago
Confidence * may be * all you need.
A simple average of the log probabilities of the output tokens from an LLM might be all it takes to tell if the model is hallucinating.🫨
The idea is that if a model is not confident (low output token probabilities), the model may be inventing random stuff.
In these two papers:
1. https://aclanthology.org/2023.eacl-main.75/
2. https://arxiv.org/abs/2303.08896
The authors claim that this simple method is the best heuristic for detecting hallucinations. The beauty is that it only uses the generated token probabilities, so it can be implemented at inference time ⚡
Organizations
None yet
stolsvik's activity
liked
a
dataset
25 days ago
liked
a
dataset
9 months ago
reacted to
santiviquez's
post with ❤️
about 1 year ago
Post
Confidence * may be * all you need.
A simple average of the log probabilities of the output tokens from an LLM might be all it takes to tell if the model is hallucinating.🫨
The idea is that if a model is not confident (low output token probabilities), the model may be inventing random stuff.
In these two papers:
1. https://aclanthology.org/2023.eacl-main.75/
2. https://arxiv.org/abs/2303.08896
The authors claim that this simple method is the best heuristic for detecting hallucinations. The beauty is that it only uses the generated token probabilities, so it can be implemented at inference time ⚡
A simple average of the log probabilities of the output tokens from an LLM might be all it takes to tell if the model is hallucinating.🫨
The idea is that if a model is not confident (low output token probabilities), the model may be inventing random stuff.
In these two papers:
1. https://aclanthology.org/2023.eacl-main.75/
2. https://arxiv.org/abs/2303.08896
The authors claim that this simple method is the best heuristic for detecting hallucinations. The beauty is that it only uses the generated token probabilities, so it can be implemented at inference time ⚡
- 4 replies
liked
a
model
about 1 year ago
liked
a
Space
about 1 year ago
TheBloke's quants?
#5 opened about 1 year ago
by
stolsvik
Should not be called mixtral, the models made into the moe are yi based
9
#2 opened about 1 year ago
by
teknium