What is Meta’s generative AI Tool, and what are its benefits?
Meta is trying to tackle the problem of generative AI tools producing inaccurate or misleading responses. Hence, Meta is using AI via a new process. People at Meta are calling it “Shepherd.”
Meta’s new Shepherd LLM is designed to review model responses and suggest refinements. It will help a lot in powering more accurate generative AI outputs.
Meta further states that the core of our approach is a high-quality feedback dataset that they are curating from community feedback and human annotations.
Even though Shepherd is miniature (7B parameters), its reviews are either equivalent or preferred to those from established models, including ChatGPT.
Using GPT-4 for evaluation, Shepherd is reaching an average win rate of 53-87% compared to all the other competitive alternatives.
In human evaluation, Shepherd strictly outperforms all other models and closely ties with ChatGPT on average.
So it is improving at providing automated feedback on why generative AI outputs are wrong. It helps to guide users to probe for more information or to clarify the details.
Why not build this into the primary AI model and produce better results without this middle step?
Most of the users are not coding geniuses; hence, they will not pretend to understand whether this is even possible at this stage.
Although, the end goal is to facilitate better responses by forcing generative AI systems to re-assess their incorrect or incomplete answers to pump out better replies to your queries.
Indeed, OpenAI says that its GPT-4 model is already producing far better results than the current commercially available GPT systems.
It is like those used in the current version of ChatGPT, while some platforms are also seeing good results by using GPT-4 as the code base for moderation tasks, and it often rivals all the human moderators in performance.
It will lead to significant advances in AI usage by social media platforms. Meanwhile, all these systems will likely never be as good as humans at detecting nuance and meaning.
It could soon be subject to more automated moderation within our posts.
And for general queries, having additional checks and balances like Shepherd’s will also help refine the results provided or help developers build better models to meet demand.
Ultimately, the push will see these tools getting more innovative and better at understanding every user query. While generative AI is impressive in what it can provide now, it is getting closer to being more reliable as an assistive tool and likely a more significant part of your workflow.
Do you want to read more about the Meta’s Shepherd system?
You can read about Meta’s Shepherd system here.
Do you want to know about more updates about Meta?
Click here to learn more about Meta and its daily updates.