DeepSeek introduced CodeI/O, a method that helps AI learn reasoning patterns hidden in code.

Models train to predict inputs/outputs of given code, all while explaining its reasoning with Chain-of-Thought (CoT) in natural language.

This improves models' general reasoning skills

Here's how:

SwiftKV promises to make open source AI up to 75% cheaper, enhancing its efficiency

We spoke to Yuxiong He and Samyam Rajbhandari, Snowflake’s AI research leads, who explained:
- How SwiftKV works
- Additional methods, limitations
- Why they decided to open source it
www.youtube.com/watch?v=9x1k...

10 recent advancements in math reasoning:

▪️ AceMath from NVIDIA
▪️ Qwen2.5-Math-PRM and PROCESSBENCH evaluation
▪️ rStar-Math from @msftresearch.bsky.social
▪️ BoostStep
▪️ URSA
▪️ U-MATH
▪️ SVE-Math
...

It's a very interesting shift in AI! Check this out for more info: huggingface.co/posts/Ksenia...

New work from Alibaba_Qwen🔥

Qwen2.5-Math-PRM 7B & 72B 🔢 Process Reward Models for enhanced process supervision in the mathematical reasoning of LLMs.

Paper:
huggingface.co/papers/2501....
Model:
huggingface.co/Qwen/Qwen2.5...
huggingface.co/Qwen/Qwen2.5...

AIs beat humans in basically all creativity tests like the Torrence Test, but it isn't clear what exactly that means compared to humans.

Yet more evidence that people can’t accurately detect well-prompted AI writing (and AI can’t accurately detect well-prompted AI writing, either). arxiv.org/pdf/2407.08853

AIs in the world can create complex systems with unexpected & risky feedback loops.

Each output affects the world, affecting inputs, affecting outputs... Even simple loops can drive optimization behavior, making systems turn extreme. No training needed, just interaction. arxiv.org/pdf/2402.06627