4 ms·
Ask HN: Multiple LLM work together to give higher than gpt4o performance
I am building a ChatGPT-like platform that, instead of asking one LLM a question, it will ask multiple LLMs all at the same time and create a final output by corroborate output from each mode. Model will work together to come up with the final answer and we see greatly improve performance and reduce hallucinations.
Notably, on AlpacaEval 2.0, using solely open-source models, we achieved a margin of 7.6% absolute improvement from 57.5% (GPT-4 Omni) to 65.1% (Mix Model) and 65.7% using close source model.
Is this something that people find to be useful? Not sure what are the use case, or is working with one LLM enough
- ricopags 2y agoYou mean like this? https://www.together.ai/blog/together-moa https://www.together.ai/blog/together-moa
- VictorPenJust 2y agoyeah, but it will be like a chatgpt like interface, where you can see all the output, the source, and the final answer, and you can ask further questions
- p1esk 2y agowe achieved a margin of 7.6% Are you one of the original contributors of the work done at together.ai [1], or are you trying to take credit for it? [1] https://arxiv.org/abs/2406.04692 https://arxiv.org/abs/2406.04692
- altdataseller 2y agoI would not find it useful, because that would entail 5X the costs, and the cost is not worth the extra benefit of less hallucinations