the deeper point here connects to something the field keeps rediscovering. we trained reasoning models to think longer. then we discovered longer doesn't mean better. now this paper shows the models themselves already know that. they're generating stop signals that our inference
OPEN SOURCE
-
DeepSeek and Qwen3 performance improvements
By
–
specific numbers worth sitting with: > DeepSeek-R1-7B on MATH-500: 93% accuracy (up from 91.6%), tokens cut from 3,871 to 2,141 > DeepSeek-R1-1.5B on AIME 2025: accuracy jumps 6.2 percentage points > Qwen3-8B: response length halved from 18,342 to 9,183 tokens with no accuracy
-
SAGE: Efficient Reasoning with Confidence Checks
By
–
their solution: SAGE (Self-Aware Guided Efficient Reasoning). instead of generating token by token, SAGE extends chains in whole reasoning steps. after each step, it checks: is the model confidently signaling it wants to stop? if yes, reasoning ends. no fine-tuning. no new
-
Overthinking harms accuracy in AI responses
By
–
and it's not just wasted compute. overthinking actively hurts accuracy. DeepSeek-R1 produces responses 5x longer than Claude 3.7 Sonnet on AIME 2025 with comparable accuracy. QwQ-32B scores 2 percentage points HIGHER with its shortest answers using 31% fewer tokens. 72% of
-
Overthinking in AI: A Sampling Issue
By
–
reasoning models already know when they've solved the problem. we just don't let them stop. new paper from Beihang University and ByteDance shows that the overthinking problem in models like DeepSeek-R1 and Qwen3 isn't a training failure. it's a sampling failure. the fix cuts
-
NEO-unify: Building Native Multimodal Unified Models End to End
By
–
NEO-unify: Building Native Multimodal Unified Models End to End Blog: https://
huggingface.co/blog/sensenova
/neo-unify
… -
Forced compliance, no OpenAI or Anthropic, but Chinese open source models run locally
By
–
They can force American companies to comply. And you can't download OpenAI or Anthropic. You can download lots of Chinese open source models though. Ask @alexfinn what he has running locally on his Mac Studios ($10,000 each no less).
-
Building Tenstorrent Cluster for Mass AI Adoption
By
–
2022/2023 was langchain, llama index, LiteLLM, OpenRouter, crew ai, e2b (I’m missing a ton) so it was a lot more core ideas being introduced, vs now it’s more about how to hook these ideas up to make them work reliably and safely
-
GPT-5.4 scores 74% on ARC-AGI-2
By
–
GPT-5.4 scored 74.0% on ARC-AGI-2 GPT-5.4 Pro got 83.3%, getting close to a level of Gemini 3 Deep Think.
-
Chatterbox by Resemble: Open-source voice cloning with 5-second audio.
By
–
Chatterbox by Resemble. Open-source, clone with ~5sec of audio.