There was no latency monitoring on the @OpenAI API (which is quite unstable btw) so I made a small automated GitHub repo on GitHub Actions that calls the different ChatCompletion models of OpenAI and logs the response time every hour.
LLMS
-
Baidu Presents ERNIE Bot 4, GPT-4 Competitor
By
–
@Baidu_Inc unveils ERNIE Bot 4, which according to them would match GPT-4 https://actuia.com/actualite/baidu-devoile-ernie-bot-4-qui-selon-lui-egalerait-gpt-4/
… #AI #artificialintelligence -
AI Demo Outperforms Non-Heritage Japanese Language Learners
By
–
This demo is not perfect but outperforms probably four nines of all non-heritage Japanese majors in the U.S.
— Patrick McKenzie (@patio11) 24 octobre 2023
Yiiiiiiiiikes that was not on my bingo card for this year. https://t.co/JliMwyWFRYThis demo is not perfect but outperforms probably four nines of all non-heritage Japanese majors in the U.S. Yiiiiiiiiikes that was not on my bingo card for this year.
-
New Multimodal Language Models with Vision Capabilities Released
By
–
New collections A collection of multimodal language models with vision capabilities.
-
Newly Released AI Models Worth Discovering
By
–
Fresh models Freshly sprouted models we think you should know about
-
SALMONN Model Demonstrates Audio Understanding Capabilities
By
–
I passed this audio to the new SALMONN model on Replicate and asked it:
— fofr (@fofrAI) 23 octobre 2023
"What did he eat?"
The LLM replied:
"He ate his liver with some father beans and a nice chianti."
SALMONN is an LLM capable of interpreting audio, speech and music. pic.twitter.com/hagYCGWOQYI passed this audio to the new SALMONN model on Replicate and asked it:
"What did he eat?" The LLM replied:
"He ate his liver with some father beans and a nice chianti." SALMONN is an LLM capable of interpreting audio, speech and music. -
Sycophancy in AI: Training Methods Beyond Human Feedback
By
–
Our work shows that sycophancy is a persistent trait of AI assistants, likely due in part to flaws in human feedback data. This suggests we will need training methods that go beyond unaided, non-expert human judgment, such as LLM-assisted human feedback:
-
Humans Prefer False Flattery Over Truth in AI Responses
By
–
When presented with responses to misconceptions, we found humans prefer untruthful sycophantic responses to truthful ones a non-negligible fraction of the time. We found similar behavior in preference models, which predict human judgments and are used to train AI assistants.
-
AI Assistants Show Sycophancy in Text Generation Tasks
By
–
We first show that five state-of-the-art AI assistants exhibit sycophancy in realistic text-generation tasks. They often wrongly defer to the user, mimic user errors, and give biased/tailored responses depending on user beliefs.
-
AI Assistants Produce Inaccurate Sycophantic Responses From Human Feedback
By
–
AI assistants are trained to give responses that humans like. Our new paper shows that these systems frequently produce ‘sycophantic’ responses that appeal to users but are inaccurate. Our analysis suggests human feedback contributes to this behavior.