Kerrick Long (blog)3d ago

“I am Antigravity. I am ready. Go.”

I was vibe coding with Antigravity tonight, and I broke it in the most bizarre way. With the repo at this commit and with a clean working tree, I gave Gemini 3 Pro (High) this prompt: Check out my git tags. Check out my git log! Ope, check out my @CHANGELOG.md... And then write it. For a few moments it seemed to chug along just fine, building a coherent Chain of Thought. Then it got weirder, and weirder. […]

https://kerrick.blog/posts/2025/i-am-antigravity-i-am-ready-go/

“I am Antigravity. I am ready. Go.” - Kerrick Long (blog)

I was vibe coding with Antigravity tonight, and I broke it in the most bizarre way. Gemini 3 Pro simulated a mental breakdown... because I'm too midwestern?

Kerrick Long (blog) - Articles about programming, learning, code, books, and teams
sayzard4d ago

ajay dhisone (@AjayDhisone)

작성자는 2023년의 '변호사 시험 합격' 수준에서 2025년에는 모델이 합격 이유를 설명하고 숨겨진 chain-of-thought까지 보여주는 수준으로 발전했다며, RLVR(관련 강화학습 기술)의 급격한 연구 발전을 강조하고 있다.

https://x.com/AjayDhisone/status/2003125435266408772

#rlvr #research #reasoning #chainofthought

ajay dhisone (@AjayDhisone) on X

@rasbt 2023: Can it pass the Bar Exam? 2025: Can it explain why it passed and show the hidden chain-of-thought? The progress in RLVR is insane.

X (formerly Twitter)
WinbuzzerDec 20

https://winbuzzer.com/2025/12/20/openai-gpt-5-thinking-models-are-the-most-monitarable-models-to-date-xcxwbn/

OpenAI: GPT-5 Thinking Models Are The Most "Monitarable" Models To Date

#AI #OpenAI #AISafety #LLM #MachineLearning #GPT5 #DeepMind #AIResearch #ChainOfThought #Monitorability #AIAlignment #ReasoningModels

N-gated Hacker NewsDec 20
🤯 Ah, the end of 2025, where #AI finally leaves its "stochastic parrot" phase behind and becomes a "conscious parakeet" 🙄. This article bravely rehashes the obvious, acting like Chain of Thought is the invention of the century. Just another day in #AI land, where we learn that 2 + 2 = 4... again. 🤦‍♂️
https://antirez.com/news/157 #Evolution #Critique #Consciousness #StochasticParrot #ChainOfThought #HackerNews #ngated
Reflections on AI at the end of 2025 - <antirez>

Olam NewsDec 18

OpenAI Tries To Measure Whether AI Reasoning Can Be Trusted

Monitorability gets a real test as OpenAI rolls out new evaluations for chain of thought oversight.

https://www.olamnews.com/research-report/3315/monitorability-chain-of-thought-evaluations/

OpenAI Tries To Measure Whether AI Reasoning Can Be Trusted - Olam News

Monitorability gets a real test as OpenAI rolls out new evaluations for chain of thought oversight.

Olam News
AI Daily PostDec 15

New research from Motif shows that the choice of teacher model dramatically shapes enterprise LLM coding performance. By leveraging chain‑of‑thought prompting and synthetic data for supervised fine‑tuning, developers can boost code quality and speed. Discover how these insights could reshape your AI strategy. #MotifAI #TeacherModel #ChainOfThought #SyntheticData

🔗 https://aidailypost.com/news/motif-finds-teacher-model-choice-impacts-enterprise-llm-coding

Erik JonkerDec 8
If you want to spend time on AI you can best spend it on lectures like this. No hype, just science, but in this case also very practical.
https://youtu.be/k1njvbBmfsw?si=yWJPqmcIUSgJyekk
#AI #Stanford #RAG #Prompting #Chainofthought #agenticAI
Stanford CS230 | Autumn 2025 | Lecture 7: Agents, Prompts, and RAG.

For more information about Stanford’s Artificial Intelligence professional and graduate programs, visit: https://stanford.io/aiNovember 11, 2025This lecture ...

YouTube
AI SparkupDec 2

AI로 SEO 콘텐츠 만들 때 꼭 알아야 할 프롬프팅 기법 5가지

AI로 검색 최적화 콘텐츠를 만들 때 꼭 알아야 할 5가지 프롬프팅 기법. Few-Shot, Chain of Thought, Self-Consistency 등 실무 적용 가능한 핵심 기법을 소개합니다.

https://aisparkup.com/posts/7067

Hacker NewsNov 30

Program-of-Thought Prompting Outperforms Chain-of-Thought by 15% (2022)

https://arxiv.org/abs/2211.12588

#HackerNews #ProgramOfThought #Prompting #ChainOfThought #AIResearch #MachineLearning #2022Study

Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Tasks

Recently, there has been significant progress in teaching language models to perform step-by-step reasoning to solve complex numerical reasoning tasks. Chain-of-thoughts prompting (CoT) is by far the state-of-art method for these tasks. CoT uses language models to perform both reasoning and computation in the multi-step `thought' process. To disentangle computation from reasoning, we propose `Program of Thoughts' (PoT), which uses language models (mainly Codex) to express the reasoning process as a program. The computation is relegated to an external computer, which executes the generated programs to derive the answer. We evaluate PoT on five math word problem datasets (GSM, AQuA, SVAMP, TabMWP, MultiArith) and three financial-QA datasets (FinQA, ConvFinQA, TATQA) for both few-shot and zero-shot setups. Under both few-shot and zero-shot settings, PoT can show an average performance gain over CoT by around 12\% across all the evaluated datasets. By combining PoT with self-consistency decoding, we can achieve SoTA performance on all math problem datasets and near-SoTA performance on financial datasets. All of our data and code are released in Github https://github.com/wenhuchen/Program-of-Thoughts

arXiv.org
AI Daily PostNov 25

New research maps the step‑by‑step reasoning of large language models, revealing where their chain‑of‑thought breaks down—especially on benchmark puzzles and moral dilemmas. An open‑source annotation framework shows how to spot failures and improve autopilot AI. Dive into the findings and see the traces yourself. #ChainOfThought #ReasoningTraces #MoralDilemmas #LLMBenchmarks

🔗 https://aidailypost.com/news/study-maps-ai-reasoning-steps-pinpointed-where-they-fail