VibeThinker: 3B param model that beats Opus 4.5 on reasoning with novel SFT+GRPO (arxiv.org) | HN Screenshots

YHacker Newsscreenshots

Screenshot of arxiv.org: VibeThinker: 3B param model that beats Opus 4.5 on reasoning with novel SFT+GRPO

First screen, captured 4 months ago.

Looks similar

Screenshot of arxiv.org: TinyLoRA – Learning to Reason in 13 Parameters

TinyLoRA – Learning to Reason in 13 Parameters (arxiv.org)

227 points by sorenjan 7 months ago | 42 comments

Screenshot of arxiv.org: Reasoning LLMs are wandering solution explorers

Reasoning LLMs are wandering solution explorers (arxiv.org)

83 points by Surreal4434 1 year ago | 78 comments

Screenshot of arxiv.org: KV Cache Compression 900000x Beyond TurboQuant and Per-Vector Shannon Limit

KV Cache Compression 900000x Beyond TurboQuant and Per-Vector Shannon Limit (arxiv.org)

40 points by EGreg 6 months ago | 11 comments

Screenshot of arxiv.org: AI Self-preferencing in Algorithmic Hiring: Empirical Evidence and Insights

AI Self-preferencing in Algorithmic Hiring: Empirical Evidence and Insights (arxiv.org)

309 points by laurex 5 months ago | 163 comments

Screenshot of arxiv.org: Multi-Stream LLMs: new paper on parallelizing/separating prompts, thinking, I/O

Multi-Stream LLMs: new paper on parallelizing/separating prompts, thinking, I/O (arxiv.org)

135 points by atomicthumbs 5 months ago | 15 comments

Screenshot of arxiv.org: Large Language Model Reasoning Failures

Large Language Model Reasoning Failures (arxiv.org)

25 points by T-A 8 months ago | 37 comments

Related stories

Screenshot of tedium.co: What ArtCraft's Vibe-Coded Apps Say About Adobe

What ArtCraft's Vibe-Coded Apps Say About Adobe (tedium.co)

23 points by speckx 18 days ago | 29 comments

Tedium article examining ArtCraft's vibe-coded remakes of creative apps and what they suggest about Adobe and Creative Cloud.

adobe creative-cloud vibe-coding artcraft

Screenshot of arxiv.org: UniEvo-VL: Self-Distillation Training for Multimodal Model Self-Improvement

UniEvo-VL: Self-Distillation Training for Multimodal Model Self-Improvement (arxiv.org)

13 points by gmays 19 days ago | 3 comments

arXiv paper introducing UniEvo-VL, an on-policy self-distillation framework where a multimodal model uses its own self-critiques to improve image generation.

multimodal-models self-distillation diffusion-models image-generation

Screenshot of arxiv.org: GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models [pdf]

GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models [pdf](arxiv.org)

382 points by SerCe 1 year ago | 71 comments

arXiv technical report from Zhipu AI and Tsinghua introducing GLM-4.5, a 355B-parameter open-source MoE language model for agentic, reasoning, and coding tasks.

glm-4-5 zhipu-ai large-language-models mixture-of-experts

Screenshot of privatemode.ai: Turning GLM-5.3-Flash into a Jev-like decision model

Turning GLM-5.3-Flash into a Jev-like decision model (privatemode.ai)

116 points by flxflx 1 month ago | 50 comments

Privatemode AI blog post describing how to prompt GLM-5.3-Flash so its first output token gives a decision in a single forward pass, benchmarked against Jev and Laya.

llm glm vllm prompt-engineering

Screenshot of blog.sshh.io: Every SaaS business will become a harness around a model

Every SaaS business will become a harness around a model (blog.sshh.io)

117 points by iacguy 23 days ago | 85 comments

Substack essay by Shrivu Shankar arguing that every SaaS business will become a harness wrapping AI models with tooling and state.

ai-agents llm-harness saas claude-code

Screenshot of autodidacts.io: Vibecoding isn't as fun as writing code by hand

Vibecoding isn't as fun as writing code by hand (autodidacts.io)

134 points by Curiositry 20 days ago | 151 comments

A personal blog post on The Autodidacts arguing that vibecoding with AI is less satisfying than writing code by hand.

vibecoding ai-assisted-coding programming 100daystooffload