I'm curating SN's AI posts site-wide (not just ~ai) weekly. Of course, I'm late on the second edition because I was traversing the skies and I underestimated the work I needed to do. Hopefully I can hit the post button on this on Monday from here on out.

Correction from last week
I missed @noknees' post last week: I told ChatGPT, it was his last day

What can possibly go wrong?What can possibly go wrong?

Where last week was mostly about AI tools, this week was about... AI fools?

Claude runs a vending machine: decides to stock tungsten cubes

@ScoresBy shared an awesome article about an experiment where the currently regarded "best LLM to use for integrations" can really screw things up if you make it responsible for real world applications, like running a vending machine.

Yes you may buy books, digitize them and destroy them

Staying with Anthropic, @carter first shared that Fed judge sides with Anthropic over training AI on books w/o authors' permission and then 3 days later that PS: those books are destroyed. Maybe one day society will get something back from this, or perhaps we can reproduce them by having the AI cite them word for word, thanks to another share from @carter: Extracting memorized pieces of (copyrighted) books from open-weight llm

More on safety

• An AI Takeover Scenario by @OT
• ‘Unethical’ AI research on Reddit under fire by @klk
• SynthID - A tool to watermark and identify content generated through AI by @carter
• Hundreds of MCP Servers Expose AI Models to Abuse, RCE by @ch0k1

Other news:

• At 20 years old, Reddit is defending its data and fighting AI with AI by @Coinsreporter
• Gemini 2.5 allegedly threatened to kill itself by @sox
• SoftBank aims to become leading artificial super intelligence platform provider by @ch0k1
• OpenAI turns to Google's AI chips to power its products, source says by @ch0k1

Stacker 2 StackerStacker 2 Stacker

NotebookLM Workshop

@cryotosensei shared a detailed report about their workshop on NotebookLLM. Interesting stuff!

What's the Best Consumer Application of AI?

@Coinsreporter dissed @Aardvark and asked SN a question in ~ai, which probably didn't help discoverability of the question. However that did not prevent @Car from commenting with a video of the most legendary David Bowie interview ever in response.

OpinionOpinion

• The Monster Inside ChatGPT by @SimpleStacker
• AI Slop: Last Week Tonight with John Oliver by @south_korea_ln
• Nils Pihl - The Great Reversal: Digital Beings in the Physical World by @ScoresBy
• I Built an AI Career Coach. I’ve Never Had a Better Coach by @jakoyoh629 
• The Wolf-Krugman Exchange: AI hype vs reality by @0xbitcoiner 
• Will AI end the resume? by @suraz

IntegrationsIntegrations

• Make Sense of 10K+ Line GitHub Repos Without Reading the Code by @optimism
• Sovereign "deep research" with jan-nano by @optimism
• TiresiasIQ: Human Behavior Prediction Engine by @noknees 
• Leica Geosystems unveils AI-powered safety solution for construction vehicles by @BlockchainB
• GitHub Wants to Change How Developers Work on Tasks by @smoking_snake

ResearchResearch

• MIT says using ChatGPT can rot your brain. The truth is more complicated by @south_korea_ln
• AGI is Mathematically Impossible 2: When Entropy Returns by @carter
• 1-Bit LLM: The Most Efficient LLM Possible? by @carter
• Continuous Thought Machines by @carter
• Charted: Productivity Gains from Using AI by @0xbitcoiner 
• Gemini Robotics On-Device brings AI to local robotic devices by @carter
• The LLM's RL Revelation We Didn't See Coming by @carter
• The Bitter Lesson is coming for Tokenization by @carter
• MUVERA: Making multi-vector retrieval as fast as single-vector search by @carter
• AlphaGenome: AI for better understanding the genome by @carter

ToolsTools

• Imagen 4 is now available in the Gemini API and Google AI Studio by @0xbitcoiner 
• Building agents with OpenAI and Cloudflare’s Agents SDK by @0xbitcoiner 
• A Marketplace For AI Models by @0xbitcoiner
• Google is bringing Gemini CLI to developers’ terminals by @ch0k1 - and #1015464 by @carter

PolicyPolicy

• Debate over future of US AI regulation hinges on broadband funding by @Coinsreporter
• docs: define policy forbidding use of AI code generators by @carter