ChatGPT (GPT-3.5) was not mentioned in the LLaMA paper. Either these tests were conducted prior to ChatGPT’s release, or it was intentionally left out of the comparison study.
@aibreakfast
-
Offline customized model that doesn’t record your inputs, valuable to some
By
–
Offline, customized model that doesn't record your inputs. Not necessary for the vast majority of users, but valuable to some.
-
LLMs could have Stable Diffusion moment, run locally soon
By
–
Decent LLMs could have their "Stable Diffusion" moment within a year – and be able to run locally on your own machine. They are getting smaller and more efficient: Meta's LLaMA 13B model beat GPT-3 in 5:7 zero-shot common sense reasoning tests, with ~92% fewer parameters.
-
Information about LLaMA, open language model
By
–
More info on LLaMA here: https://research.facebook.com/publications/llama-open-and-efficient-foundation-language-models/ …
-
LLaMA omits GPT’s results from MMLU test chart due to size.
By
–
This was the Massive Multitask Language Understanding test (Humanities, STEM, Social Sciences, etc) When finetuned for the test, LLaMA omitted GPT's superior results from the chart because it is not of "moderate size"
-
Meta’s new LLaMA model paper highlights PaLM’s capabilities
By
–
Meta's new AI model LLaMA came out today. The paper was a gentle reminder of how capable Google's 540-billion parameter model PaLM will be.
-
Self-custodying your LLM: The future of AI in 2025
By
–
2025: "You don't self-custody your own LLM?"
-
NVIDIA CEO predicts AI models 1 million times more powerful than ChatGPT
By
–
Moore's Law in action: NVIDIA CEO Jensen Huang predicts AI models "one million times more powerful than ChatGPT" within 10 years. NVIDIA currently controls an estimated 95% of the AI/ML market with its chips.
-
US Copyright Office denies copyright for AI images
By
–
US Copyright Office announces AI-generated images are not eligible for copyright protection, per Reuters
