Hi, agreed with Daniel's comment above. And just a note, this problem can be readily resolved with an appropriate system prompt, even something simple like: "You are Granite, an AI language model developed by IBM in 2024."
Aside from that, the model dev team will also take a thorough look at it in 3.1 release to thoroughly address identity issues like this. Thanks for flagging the feedback !!
------------------------------
HUI WU
------------------------------
Original Message:
Sent: Thu November 14, 2024 09:24 AM
From: Daniel Toczala
Subject: Granite3 itself claimed it is created by Anthropic?
LLMs need to be tuned, and they need to be managed. A "rew" LLM is not always going to answer correctly, and the hallucinations can be quite wild, noteworthy, and somewhat embarrassing. So before you use Granite (or any LLM for that matter) it is best to remember this incident, and test and validate what you are deploying. Have you tuned it correctly? Are the proper guiderails in place?
P.S. I'm actually a little surprised that it mentioned Anthropic - most of the "chatter" on the internet is about Chat-GPT, so I would have expected it to claim it was created by OpenAI. Or IBM (since the foundation was trained with IBM curated data). Like I said above: tune, test, and monitor.
------------------------------
Daniel Toczala
Community Leader and Customer Success Manager - Watson
dtoczala@us.ibm.com
Original Message:
Sent: Tue November 12, 2024 09:42 PM
From: SK Tay
Subject: Granite3 itself claimed it is created by Anthropic?
I downloaded and run granite3 with ollama and chatted with it.
It claimed it is created by Anthropic, not IBM ...
Any thoughts?
------------------------------
SK Tay
------------------------------