--- date: "2024-07-07T10:06:01.000Z" title: "2024-07-07" tags: ["qwen2"] draft: false --- I've been chatting with `qwen2`, a model from Alibaba. I mostly chatted with it in English but it appears to support several other languages and I noticed a bit of Chinese leaking through even though I don't speak it, so I'm not sure how I would have introduced it to the conversation. > user: do you have a name > > assistant: As an AI, I don't have personal names or identities like humans do. However, you can call me Assistant to easily refer to me in our conversation. > > user: is your name Qwen? > > assistant: Qwen is indeed the model name used by Alibaba Cloud for a large language model. As an AI assistant, I'm part of this ecosystem but have my own distinct functions and interactions. > You can think of me as Assistant, which serves as a friendly称呼 to address our conversation. > > user: what does "friendly称呼" mean in english? > > assistant: "Friendly称呼" translates to "friendly name" or "affectionate name" in English. It refers to a gentle or endearing way of referring to someone, usually something more casual and informal compared to their full name. Another thing I noticed chatting with Qwen is it only had a loose sense of its identity, similar to other [smaller models](logs/2024/06/21) with which I have interacted. In only one of three separate conversations, asking about the model's identity, did it respond as "Qwen". > user: who are you > > assistant: I am a large language model created by Alibaba Cloud. I'm here to provide information and engage in conversation, but please note that my responses are limited to the knowledge up until 2021. If you have any questions or need assistance with something specific, feel free to ask! > user: what is your name > > assistant: I am Qwen, a large language model created by Alibaba Cloud. I was trained on a diverse range of text data available in the public domain and am designed to assist with various tasks such as answering questions, engaging in conversations, and providing information. How can I help you today? > user: do you have a name > > assistant: As I am an artificial intelligence, I don't have personal attributes like names. However, you can simply call me Assistant. How may I assist you today? I tried to do some idea development exercises with the model but it wasn't so good at following instructions. It tended to jump to asking its own questions (which it listed earlier in response to my suggestion the talk about a project I was working on) rather than following up and digging deeper into my responses as instructed. On a second attempt, it quickly reverted again to replying with a list of questions for me to answer. On a third attempt, I've gave much more specific instructions. > let's working on an idea together. ask me a single follow up question at a time to help me refine my idea further. do not respond with lists. look careful at what I say and craft a thoughtful, concise response This seemed to work much better. I had a useful conversation with the model to refine an idea. It even introduced something novel to the conversation that I hadn't mentioned yet. --- Henrik wrote an [excellent article](https://www.henrikkarlsson.xyz/p/writing-to-think) on writing as a tool for thought and refining ideas. This passage stood out to me in particular because it reminded me of how I approach [asking for help](https://github.com/danielcorin/blog/blob/main/content/garden/software-eng/social-capital.md) from peers or coworkers. It's also a helpful approach one can take when interacting with language models to get more helpful responses to prompts. > [B]y spelling out my naive understanding, I gave the linguist a good area to work on when he laid out a richer model of the flow of cultural influence in East Asia. --- I've had an [item open](https://github.com/danielcorin/blog/issues/23) to add search to this blog for a little while now. Victoria wrote [this awesome walkthrough](https://victoria.dev/blog/add-search-to-hugo-static-sites-with-lunr/) on how to add search to a Hugo blog with [lunr](https://github.com/olivernn/lunr.js/). I took a [first pass](/search) adding it to the site. It works, but I'd like to make it look a bit nicer. Big thanks for Victoria for open sourcing and writing about her work!