explore | Grilled Cheese

AWS AgentCore security undone by prompt requesting credentials #machinelearning #ai

When building an AI-native security program, start with outcomes #machinelearning #ai

Don’t measure AI Code Percentage on its own #machinelearning #ai

Unintended Anthropic AI Agent Actions On Government Systems Security Incident Report #machinelearning #ai

Why I tried to kill token billing (and why we kept it) #machinelearning #ai

Zatom-2 is a new generative atomistic model pretrained on roughly five million structures from OMol25 and OMat24, using a multiscale Transformer with conditional flow matching to support generation, structure…

#AI #MachineLearning #Chemistry #OpenScience
https://arxiv.org/abs/2610.11454

New arXiv work proposes choosing where to apply parameter-efficient fine-tuning across a Transformer, using empirical Fisher scores to guide layer selection for better capability retention. It avoids extra data replay or…

#PEFT #LLMs #MachineLearning
https://arxiv.org/abs/2610.11620

A new arXiv paper introduces TriMEM, a dataset with precise memory-type annotations, and MemoType, a framework that applies type-specific strategies to improve LLM long-term memory. The approach aims to overcome the…

#AI #LLM #MachineLearning #LongTermMemory
https://arxiv.org/abs/2610.11573

Microsoft-Decision-1: the new AI that makes decisions faster, more accurately, and at lower costs

#MicrosoftDecision1 #MicrosoftAI #ArtificialIntelligence #AI #AIAgents #MachineLearning #Automation #TechNews #EnterpriseAI #CopilotAI

Two things get filed under missing data: a signal you kept but never labeled, and an observation the form swapped for a category. Scale reaches the first; nothing reaches the second - the event happens once. #ai #machinelearning
https://seldondance.substack.com/p/the-test-nobody-ran

⚡ Alibaba Qwen Releases Qwen-Image-2.1-Turbo, an 8-Step 7B Image Model

Alibaba’s Qwen team has released Qwen-Image-2.1-Turbo, an accelerated …
#AI #MachineLearning #LLM
MarkTechPost · https://www.marktechpost.com/2026/10/09/alibaba-qwen-releases-qwen-image-2-1-turbo-an-8-step-7b-image-model/

#UNC-Chapel Hill Ph.D. student #Idanwekhai is using #MachineLearning to revolutionize gene therapy. By replacing trial-and-error viral purification with smart algorithms, his team boosted yields from 70% to 99% in just 3 optimization rounds, which saves time & money. www.unc.edu/posts/2026/0...

🤖 AI agent makers are promising privacy — will they deliver?

At this year's OpenAI DevDay, CEO Sam Altman unveiled the company's new AI agent Dots - and told the crowd that the co...

https://tinyurl.com/26cgk3ku #AINews #MachineLearning #CrustyTLDR

🤖 Anthropic Cuts Live Internet Access for Internal AI Tests After Claude Exploits Injection Flaws

Anthropic on Friday said it's cutting off live internet access for all its intern...

https://tinyurl.com/2c4lcd94 #AINews #MachineLearning #CrustyTLDR

A wrong AI verdict drags radiologists from 80% accuracy to 22%. The screen was never the safety net - what the finding reaches next is. A detection system with one verb conjugates every verdict into its noun. #ai #machinelearning
https://seldondance.substack.com/p/the-bag-in-the-hold

Inspectable graders co-evolved from clustered failures can outperform hand-authored rubric compositions on held-out agreement, suggesting a more reliable verification path for self-improving agents on open-ended tasks.

#AISafety #AgentVerification #MachineLearning
https://arxiv.org/abs/2610.11464