SuccToes
Firewall Cracker
2
MONTHS
2 2 MONTHS OF SERVICE
LEVEL 1
300 XP
https://openreview.net/pdf?id=_3ELRdg2sgI]"We show that STaR (Self-Taught Reasoner) significantly improves
performance on multiple datasets compared to a model fine-tuned
to directly predict final answers, and performs comparably to fine-tuning
a 30× larger state-of-the-art language model on CommensenseQA.
Thus, STaR lets a model improve itself by learning from its own generated reasoning."
Well, it's finally here. Self-improving and self-teaching AI has hit a major inflection point. We saw glimpses of things like this with previous generative pre-trained models https://arxiv.org/abs/1706.03762. But we've never seen it to this degree and especially at this scale. A big critique of AI for a long time has been it lacks common sense or real-world understanding. It appears this is the solution. As the paper explains it is utilizing what are basically "in-between steps" before outputting the final answer -- much like human reasoning. This is also how neural-networks operate in general, leveraging a high number of "hidden neuron layers" between input and output. It seems we've discovered something pretty major here and it's obvious with the new improvements in
it had some ability to reason before, we've stumbled onto something new that is not only capable of exponentially more powerful and accurate reasoning it also is capable of logarithmically improving using self-generated "synthetic data."
https://openreview.net/pdf?id=_3ELRdg2sgI]"In this paper, we adopt a different approach: by leveraging the LLM’s pre-existing reasoning ability,
we iteratively bootstrap the ability to generate high-quality rationales. Specifically, we few-shot
prompt a large language model to self-generate rationales and refine the model’s ability further by
fine-tuning on those rationales that lead to correct answers. We repeat this procedure, using the
improved model to generate the next training set each time. This is a synergistic process, where
improvements in rationale generation improve the training data, and improvements in training data
further improve rationale generation."
So where do we go from here? Is this the point when AI becomes a superintelligence? Is Terminator coming in 2025? Are we going to cure cancer and solve world hunger?
What do you think?
Either way, seems we're in for a wild ride.
source: https://openreview.net/pdf?id=_3ELRdg2sgI
performance on multiple datasets compared to a model fine-tuned
to directly predict final answers, and performs comparably to fine-tuning
a 30× larger state-of-the-art language model on CommensenseQA.
Thus, STaR lets a model improve itself by learning from its own generated reasoning."
Well, it's finally here. Self-improving and self-teaching AI has hit a major inflection point. We saw glimpses of things like this with previous generative pre-trained models https://arxiv.org/abs/1706.03762. But we've never seen it to this degree and especially at this scale. A big critique of AI for a long time has been it lacks common sense or real-world understanding. It appears this is the solution. As the paper explains it is utilizing what are basically "in-between steps" before outputting the final answer -- much like human reasoning. This is also how neural-networks operate in general, leveraging a high number of "hidden neuron layers" between input and output. It seems we've discovered something pretty major here and it's obvious with the new improvements in
https://openreview.net/pdf?id=_3ELRdg2sgI]"In this paper, we adopt a different approach: by leveraging the LLM’s pre-existing reasoning ability,
we iteratively bootstrap the ability to generate high-quality rationales. Specifically, we few-shot
prompt a large language model to self-generate rationales and refine the model’s ability further by
fine-tuning on those rationales that lead to correct answers. We repeat this procedure, using the
improved model to generate the next training set each time. This is a synergistic process, where
improvements in rationale generation improve the training data, and improvements in training data
further improve rationale generation."
So where do we go from here? Is this the point when AI becomes a superintelligence? Is Terminator coming in 2025? Are we going to cure cancer and solve world hunger?
What do you think?
Either way, seems we're in for a wild ride.
source: https://openreview.net/pdf?id=_3ELRdg2sgI