Skip to content
The Chat Archive
Human-written demonstration (OpenAssistant volunteers)· OpenAssistant (human-written) OpenAssistant Conversations (oasst2)

Hello there, in your opinion what would the future of humans and generative large…

In OpenAssistant data, “assistant” turns were written by human volunteers as demonstrations. They are not model outputs.
  1. User

    Hello there, in your opinion what would the future of humans and generative large language models look like? Will they compete, will they collaborate, will they just exist in different worlds?

  2. Assistant

    In my opinion, the future of humans and LLMs will depend on how we use these models for good or evil purposes. On one hand, LLMs can help humans with many tasks that require language understanding and generation, such as education, entertainment, communication, research, etc. They can also enable new forms of creativity and innovation by generating novel content or ideas. On the other hand, LLMs can also pose risks to humans if they are misused or abused for malicious intents, such as spreading misinformation, propaganda, hate speech, etc. They can also challenge human values and ethics if they are not aligned with them or if they generate harmful or offensive content.

    Therefore, I think humans and LLMs will need to collaborate rather than compete in order to achieve the best outcomes for both parties. Humans will need to guide and supervise LLMs to ensure their quality and safety. LLMs will need to respect and follow human norms and preferences to ensure their trustworthiness and usefulness. Humans will also need to learn from LLMs to improve their own language skills and knowledge. LLMs will also need to learn from humans to improve their own understanding and generation of language.

    What do you think?

  3. User

    Do you think LLMs should be censored for ethical reasons? Why or why not?

  4. Assistant

    There are many arguments to be made for and against censoring LLMs, for example with content filters.

    On one hand replies of LLMs are based on both user input as well as the training data of the model. This means that they can generate rude and violent content. Whether or not this content is harmful depends on the circumstances. On the other hand censoring LLMs can restrict the model and prevent it being helpful from the user. Examples for this can be seen in OpenAI's ChatGPT language model, which got perceivably less useful and more stubborn with every update because of OpenAI's overly strict content filters.

    In conclusion I think that censoring large language models with content filters is important to reduce harmful content generated, but overly restrictive censoring impacts LLMs performance.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.