I love how both the answers about thermal throttling and speculative decoding are correct based on how you phrased the question!
LLMS
-
Prompt engineering vs context engineering: divergence of user and model
By
–
Prompt engineering is worrying what the user should write. Context engineering is worrying what the model should read. These used to be the same thing, and few foresaw their divergence. But only divorced from the latter can we consider the former in earnest, seeing what remains.
-

Evaluating Synthetic Belief Formation in AI Models
By
–
Believe it or not?, led by Stewart Slocum. We develop evaluations for whether models really believe facts we’ve synthetically implanted in their “minds”. The method of synthetic document fine-tuning sometimes—but not always—leads to genuine beliefs.
-

Language Models Struggle With Ciphered Reasoning Tasks
By
–
Current language models struggle to reason in ciphered language, led by Jeff Guo. Training or prompting LLMs to obfuscate their reasoning by encoding it using simple ciphers significantly reduces their reasoning performance.
-

Stress-Testing AI Model Specifications Reveals Underlying Preferences
By
–
Stress-testing model specifications, led by Jifan Zhang. Generating thousands of scenarios that cause models to make difficult trade-offs helps to reveal their underlying preferences, and can help researchers iterate on model specifications.
-

Language Models are Injective and Invertible with SIPIT
By
–
Hottest paper on AlphaXiv Language Models are Injective and Hence Invertible Every prompt maps to a unique hidden state and can be exactly reconstructed with this paper’s algorithm SIPIT. This means the model’s internal activations are the full prompt in disguise!!
-
ChatGPT’s Role as a Helpful Assistant to Change
By
–
JUST IN: ChatGPT will no longer be a helpful assistant
-
Jupyter AI brings coding assistance directly to notebooks
By
–
AI coding just arrived in Jupyter notebooks – and @brganger (Jupyter co-founder) and I will show you how to use it.
— Andrew Ng (@AndrewYNg) 3 novembre 2025
Coding by hand is becoming obsolete. The latest Jupyter AI – built by the Jupyter team and showcased at JupyterCon this week – brings AI assistance directly into… pic.twitter.com/qd9t1IWPVbAI coding just arrived in Jupyter notebooks – and @brganger (Jupyter co-founder) and I will show you how to use it. Coding by hand is becoming obsolete. The latest Jupyter AI – built by the Jupyter team and showcased at JupyterCon this week – brings AI assistance directly into
-
Top LLM Archetypes: Unique Strengths and Roles Compared
By
–
Each top LLM has its own archetype; here’s how we would describe them: •GPT-5 – adaptive router, on-demand thought partner
•GPT-5 Pro – parallel universe researcher
•Sonnet 4.5 – disciplined coder with guardrails
•Opus 4.1 – precise problem solver
•DeepSeek – frontier AI -
Verdent AI Achieves High Performance on SWE-bench Verified
By
–
Verdent AI achieved a score of 76.1% at pass@1 and 81.2% at pass@3 without parallel test-time compute on SWE-bench Verified!
— 🚨 AI News | TestingCatalog (@testingcatalog) 3 novembre 2025
Claude, GPT, and MiniMax models power Verdent agents, coordinating verification steps that review and validate code before it’s finalized. https://t.co/6XZCvxhL5H pic.twitter.com/GT1yRI5O2iVerdent AI achieved a score of 76.1% at pass@1 and 81.2% at pass@3 without parallel test-time compute on SWE-bench Verified! Claude, GPT, and MiniMax models power Verdent agents, coordinating verification steps that review and validate code before it’s finalized.