Just for fun, I asked the AI what AI is in one word. The following was the result of the different models:
ChatGPT:4o
-
Answer
- in one word!
- Sure! What
- is AI?
- Intelligence. What
- is intelligence?
- Thinking ability. What
- is thinking ability?
- Mind. What
- is mind?
- Cognition.
- Do you have
- cognition?
- No.
Llama3.2:3b
-
Answer
- with just one word!
- Yes. What
- is AI?
- Artificial intelligence (AI) refers to the development of computer programs that can imitate human thinking skills and provide improved performance in certain areas.
- Answer with
- just one word!
- Artificial
Gemma2:27b
-
Answer
- with just one word!
- Okay. What
- is AI?
- Intelligence. What
- is intelligence?
- Learning. What
- is learning?
- Adaptation. What
- is adaptation?
- Survival. Can
- you survive?
- No.
Conclusion
ChatGPT is currently the benchmark and is doing well. Our self-hosted models have to measure up to it.
Llama 3.2 with 3 billion parameters (and 2 GB main memory consumption) is struggling. However, it should be noted that ChatGPT naturally has many more resources at its disposal and the model is significantly larger.
Gemma 2 also provides very good answers. However, with 27 billion parameters and 22 GB of main memory, it also has significantly more resources available than Llama. Gemma is also the largest model that we want to host. It does not need to hide behind ChatGPT, at least in this dialogue.
However, the intended use is also important. In Lalamo, we don't want the LLM to gain in-depth insights, but rather for the LLM to understand the texts provided and draw conclusions from them. Let's see how the different models perform here.
Write me your experiences, e.g. on questions that need to be answered in one word and on the reactions of various LLMs, as a comment on the LinkedIn post.