Note:
This topic has been translated from a Chinese forum by GPT and might contain errors.
Original topic: 【TiDBer 唠嗑茶话会 78】聊聊人工智能时代 :你都用过哪些 AI 大模型工具?
If we talk about the tech highlights of 2023, ChatGPT is definitely the hottest topic. With the advent of ChatGPT, AI chatbot programs have come into the public eye, and more and more similar programs are emerging. I believe all TiDBers have experienced it and noticed some minor issues with these programs.
For example, I found that sometimes my questions would receive an answer that looks “correct” but is actually fabricated. This makes me wish that the answers could include sources or links to the information!
So today, let’s talk about the shortcomings of these AI chatbot programs that you’ve noticed in practice during this era of booming AI chatbots. Or are there any aspects that can be optimized?
This Topic:
What AI large model tools have you used?
What aspects do you think AI chatbots (like ChatGPT) can still be optimized?
What fun things do you think can be created based on ChatGPT/AI large models?
Activity Rewards
Participate in this chat session to receive 30 points reward~
Activity Time:
2023.7.21 - 2023.7.28
Used cousin’s AI assistance
Which AI large model tools have you used?
GPT-3, GPT-4, BERT, Transformer
Do you think there are any aspects of AI chatbots (similar to ChatGPT) that can still be optimized?
Understanding context is still not very good.
What other fun things do you think can be made based on ChatGPT/AI large models?
Language learning assistant
- ChatGPT 3.5, 4.0
- TiDB Assistant
- Alibaba’s Tongyi Qianwen
- Baidu’s Wenxin Yiyan
It feels like a lot of work can be handed over to GPT to complete, mostly assistant-related tasks, which indeed can improve efficiency.
For example:
TiDB Assistant can reduce the time you spend searching for documents and content, allowing you to find answers more quickly and accurately. There is a prerequisite: knowing how to ask questions, which makes it easier for GPT to recognize and understand…
Xiao Ai and Xiao Bing classmates
ChatGPT
Baidu’s Wenyan Yixin
Discord Midjourney
I’ve used ChatGPT, Wenxin Yiyan, iFlytek Spark, and Tongyi Qianwen. Overall, the models launched by domestic giants have a better understanding of Chinese, while OpenAI’s is somewhat inferior. However, when it comes to maintaining context over longer conversations, they tend to give irrelevant answers. Moreover, many times the responses seem serious but are actually off-topic.
This is a common issue with large models at the current stage. They require very professional and powerful data annotation to feed valuable data; otherwise, they can mislead the AI. In terms of image generation and article generation, each has its own strengths, which also depend on the underlying computing power.
I’ve used Wenxin Yiyan and ChatGPT. ChatGPT feels amazing, but Wenxin doesn’t.
Which AI large model tools have you used?
Wenxin Yiyan, it feels pretty good.
What aspects do you think AI chatbots (similar to ChatGPT) can still be optimized?
I hope they can be customized to provide a unique chatbot for each user.
What fun things do you think can be made based on ChatGPT/AI large models?
Automatically analyze business issues based on logs and act as an intelligent DBA.
I have used ChatGPT, and it has been somewhat helpful for me. However, as of now, I can only trust certain facts or answers with specific rules, such as historical facts, translations, simple Excel formula writing, converting text to tables, and so on. For scenarios that require a certain level of judgment, the accuracy of AI responses is still quite low. For example, when dealing with issues related to online components or finding technical solutions, AI’s ability to fabricate answers is too strong, and it takes a lot of effort to verify the correctness of the answers. After asking a few times, I reverted to my previous method of searching and summarizing on my own.
Additionally, I feel that another direction, such as generating images or videos based on text, has broader application scenarios and does not require high precision, making it a better use case for AI.
It feels like this is similar to mobile internet, a technological revolution that could potentially affect all aspects of life. We are witnessing history.
ChatGPT
Bard
Claude seems to be called this
The answers from Wenxin Yiyan and ChatGPT are sometimes incomprehensible, and there is still a lot of room for improvement.