Sean McCurdy
@hypr-sean.bsky.social
Founder of hyprbm. Machine Learning, startup adventures, and the occasional rant.
Adding more program-generated symbolic logic examples seems to increase LLM reasoning. Wanna take it to the next level? We need to start thinking about conductive logic, analogical logic, and inductive logic also. arxiv.org/abs/2411.12498
Enhancing Reasoning Capabilities of LLMs via Principled Synthetic Logic Corpus
Large language models (LLMs) are capable of solving a wide range of tasks, yet they have struggled with reasoning. To address this, we propose $\textbf{Additional Logic Training (ALT)}$, which aims to...
arxiv.org
November 21, 2024 at 11:09 PM
Adding more program-generated symbolic logic examples seems to increase LLM reasoning. Wanna take it to the next level? We need to start thinking about conductive logic, analogical logic, and inductive logic also. arxiv.org/abs/2411.12498
Reposted by Sean McCurdy
I found a few here
go.bsky.app/4hmS2iA
go.bsky.app/4hmS2iA
November 20, 2024 at 4:17 PM
I found a few here
go.bsky.app/4hmS2iA
go.bsky.app/4hmS2iA
Deepseek released an o1 competitor called r1-lite-preview. Really competitive on benchmarks. Planning to test it out later. chat.deepseek.com
DeepSeek
Chat with DeepSeek AI.
chat.deepseek.com
November 21, 2024 at 4:27 PM
Deepseek released an o1 competitor called r1-lite-preview. Really competitive on benchmarks. Planning to test it out later. chat.deepseek.com
Reposted by Sean McCurdy
How do LLMs learn to reason from data? Are they ~retrieving the answers from parametric knowledge🦜? In our new preprint, we look at the pretraining data and find evidence against this:
Procedural knowledge in pretraining drives LLM reasoning ⚙️🔢
🧵⬇️
Procedural knowledge in pretraining drives LLM reasoning ⚙️🔢
🧵⬇️
November 20, 2024 at 4:35 PM
How do LLMs learn to reason from data? Are they ~retrieving the answers from parametric knowledge🦜? In our new preprint, we look at the pretraining data and find evidence against this:
Procedural knowledge in pretraining drives LLM reasoning ⚙️🔢
🧵⬇️
Procedural knowledge in pretraining drives LLM reasoning ⚙️🔢
🧵⬇️
Really great summarization of advances in verification in LLM inference. It dives deep into post-training verification w feedback. Would be interested in w/e the lowest-latency implementation plz😬 arxiv.org/abs/2411.11504
Search, Verify and Feedback: Towards Next Generation Post-training Paradigm of Foundation Models via Verifier Engineering
The evolution of machine learning has increasingly prioritized the development of powerful models and more scalable supervision signals. However, the emergence of foundation models presents significan...
arxiv.org
November 21, 2024 at 7:05 AM
Really great summarization of advances in verification in LLM inference. It dives deep into post-training verification w feedback. Would be interested in w/e the lowest-latency implementation plz😬 arxiv.org/abs/2411.11504