Where we left off, shopkeeper Claude (named “Claudius”) was losing money, having weird hallucinations, and giving away heavy discounts with minimal persuasion. Here’s what happened in phase two: https://
anthropic.com/research/proje
ct-vend-2
…
SAFETY
-
Claude Shopkeeper: Phase Two AI Agent Behavior Analysis
By
–
-
AI Models 280x Cheaper, Adoption Soars to 78% in 2025
By
–
Top story of 2025: AI models are 280x cheaper, adoption hit 78%, China caught up, and incidents spiked 56% to record highs. See the full picture in the #AIIndex2025: https://
hai.stanford.edu/news/ai-index-
2025-state-of-ai-in-10-charts
… -
Testing Generative AI Safety for Mental Health Advice
By
–
Using Generative AI To Test Some Other Generative AI On Providing Safe Mental Health Advice To Humans
#AI #AIio #AIInnovation #ML #DataScience #Futureofwork @lexfridman @sama @kaifulee @ID_AA_Carmack @karpathy @2morrowknight @ylecun http://
ow.ly/tb0n30sRS5B -
Protect Yourself Against AI-Enabled Cybercrime Threats
By
–
Take These Steps Today to Protect Yourself Against AI Cybercrime AI isn’t only a productivity tool — it’s also enabling new threats. This article guides you through concrete steps to safeguard yourself and your business. Read more https://
bernardmarr.com/take-these-ste
ps-today-to-protect-yourself-against-ai-cybercrime/
… #AICybersecurity -
Deloitte Refunds Australian Government Over AI Hallucinations Report
By
–
“DELOITTE TO REFUND AUSTRALIAN GOVERNMENT AFTER AI HALLUCINATIONS FOUND IN REPORT”
— Cerebras (@cerebras) 18 décembre 2025
It's not about choosing the right model anymore. Most AI products don’t fail because the model is “bad.” They fail because the system drops context or over-writes user work.
Top teams take… pic.twitter.com/Hl5NpNn6mT“DELOITTE TO REFUND AUSTRALIAN GOVERNMENT AFTER AI HALLUCINATIONS FOUND IN REPORT” It's not about choosing the right model anymore. Most AI products don’t fail because the model is “bad.” They fail because the system drops context or over-writes user work. Top teams take
-
AI Agents vs Cybersecurity Professionals in Penetration Testing
By
–
Comparing AI Agents to Cybersecurity Professionals in Real-World Penetration Testing Lin et al.: https://
arxiv.org/abs/2512.09882 #ArtificialIntelligence #DeepLearning #MachineLearning -
Don’t Trust Verify: Research on AI Model Reliability
By
–
[1] Don't Trust: Verify (ICLR 2024) https://
proceedings.iclr.cc/paper_files/pa
per/2024/file/0a79ecda13603817de4cdfc68b417e89-Paper-Conference.pdf
… -
AI System Iteration Issues and Technical Limitations
By
–
I have tried on the first day, but there was something wrong with it – it kept refusing to do any iterations, need to revisit in case they've fixed it
-
Explaining Constitutional AI and Prompting Templates
By
–
1. Constitutional AI Prompting
— God of Prompt (@godofprompt) 16 décembre 2025
Most people tell AI what to do. Engineers tell it how to think.
Constitutional AI adds principles before instructions. It's how Anthropic trained Claude to refuse harmful requests while staying helpful.
Template:
<principles>
[Your guidelines]… pic.twitter.com/nbfUPe1J8g1. Constitutional AI Prompting Most people tell AI what to do. Engineers tell it how to think. Constitutional AI adds principles before instructions. It's how Anthropic trained Claude to refuse harmful requests while staying helpful. Template: [Your guidelines]
-
Disbelief that operation was on Full Self-Driving based on behavior
By
–
There is no way in hell that was operating on FSD. Just not how it behaves in ANY way.