I'm curating SN's AI post site-wide (not just ~ai) weekly.
Are humans the weakest link?
This week, human usage of LLMs was by far the most popular subject on AI SN and besides normal humans, lawyers using AI were making headlines: @Scoresby shared Shahid v. Esaam: court rules based on hallucinated case law and @optimism shared Legal team fined close to 6M sats in sanctions for using an LLM. Courts seem to be of the opinion that hallucinating some non-existing legal case is not cool (probably because it's against the law to do this in court.)
@kepford asked AI Disclosures: Do they have any value? and while this will probably not have saved the aforementioned lawyers from being sanctioned, SN does seem to agree on some use for this, in a limited form. Related, @SimpleStacker asked SN Are you downzapping suspected AI posts and comments?, which makes one wonder: would you downzap it if there was a disclosure with it? Or only if it's passed off as human content?
@0xbitcoiner shares Scholars sneaking phrases into papers to fool AI reviewers which shows some real creativity in indirect prompt engineering (and poisoning.)
Be careful with your interactions! @zuspotirko warns: Don't tell ChatGPT enough so it can predict your future. If ever there was a time to heed Marc Andreessen's advice, now is the time. Don't ask ChatGPT what it knows about you, and perhaps just stop using data harvesting chatbot services altogether. Note that OpenAI to release web browser in challenge to Google Chrome, shared by @Coinsreporter, means more data harvesting from OpenAI, simply because they're jelly at the sheer volume of data Google can harvest with Chrome.
Last but not least, according to Grok, the real problems are not the LLMs: Grok says Elon and Trump are largest disinformation spreaders on X, @79c9095526 shared.
Other posts on human/LLM friction:
- Why do people find it so exciting when LLMs say outrageous things? by @Scoresby
- I asked ChatGPT 4o to make ASCII art of Bitcoin by @SimpleStacker
- The ability to craft good prose used to be proof-of-work - Byrne Hobart by @Scoresby
- What the World Is Asking ChatGPT in 2025 by @0xbitcoiner
Guides and Reports
- 7 Steps to Mastering Vibe Coding by @optimism
- How Cursor went from 0 to $500M ARR in 30 months - Gupta post by @Car
- Frequently Asked Questions (And Answers) About AI Evals by @carter
- Solving Gemini CLI Authentication on Remote Machines by @klk
- AI Model & API Providers Analysis by @0xbitcoiner
Safety
- Anthropic: A Framework for Frontier AI Models Development Transparency by @Cje95
- Supabase MCP can leak your entire SQL database by @hn
- Evaluating and monitoring for AI scheming by @Msd0457890
Opinions
- Intelligence is not compression - Nicholas Carr by @Scoresby
- Grok: searching X for “from:elonmusk (Israel OR Palestine OR Hamas OR Gaza)” by @carter
- I’m Losing All Trust in the AI Industry by @carter
- What Do Commercials About A.I. Really Promise? by @Coinsreporter
- The upcoming GPT-3 moment for RL by @carter
- Artificial intelligence is not wise. by @Solomonsatoshi
- AI Agent Benchmarks are Broken by @carter
- Stop Building AI Agents by @carter
- Designers: We’ll all be design engineers in a year by @deSign_r
- Elon Musk has created an AI monster by @ch0k1
Research
- Osmo - "Giving computers a sense of smell" by @k00b
- Hill Space: Neural nets that do perfect arithmetic (to 10⁻¹⁶ precision) by @carter
- Does AI use actually slow down developer productivity? by @SimpleStacker, #1032946 by @hn, #1033198 by @zuspotirko
- AI ‘scientists’ joined these research teams: here’s what happened by @NovaRift
- How Long Contexts Fail by @carter
- Can an AI model predict perfectly and still have a terrible world model? by @carter
- WorldVLA: Towards Autoregressive Action World Model by @carter
- Overclocking LLM Reasoning by @carter
- DesignArena – crowdsourced benchmark for AI-generated by @deSign_r
- Why LLMs Can't Write Q/Kdb+: Writing Code Right-to-Left by @hn
- Is Gemini 2.5 good at bounding boxes? by @hn
Models and Tools
- BitAgent: enable AI agents to pay for services using the Bitcoin LN & Nostr by @supratic
- Shakespeare - Build custom Nostr websites with AI assistance by @k00b
- VS Code has an "Agent" mode by @carter
- Mirage: The World's First AI-Native Game Engine Powered by Real-Time World Model by @carter
- Apple just released an interesting diffusion based coding language model by @carter
- Kimi K2 - An Open-Source Agentic MoE model by @carter
- Mercury: Ultra-Fast Language Models Based on Diffusion by @hn
Implementations
- AI is changing the rental car return experience - and it could cost you by @k00b
- abacus: WIP autonomous lightning network node agent by @k00b
- My first vibe coded app using AI by @stax
- How generative AI could help make construction sites safer by @BlokchainB
- [Feedback] Gift Bitcoin in Red envelope by @SalmaChan
- Batch Mode in the Gemini API: Process more for less by @carter
- I prompted a Near Protocol agent to emulate John Dee by @nkmg1c_ventures
News and Announcements
- Create Paid MCP Servers with PaidMCP and earn BTC for your AI tools by @Alby
- Hugging Face launching $299 robot that to disrupt the entire robotics industry by @carter
- Anyone watching the xAI Grok 4 livestream? (live discussion thread) by @gmd and #1031915 by @ch0k1
- SpaceX to Invest $2 Billion Into Elon Musk’s xAI by @Coinsreporter
- Updated Goose Roadmap · grant program dedicated team funding by @Scoresby
- YouTube ‘clarifies’ its plan to demonetize spammy AI slop by @jakoyoh629
- Elon Musk Says Grok Is Coming to Tesla Evs by @0xbitcoiner
- Transforming Pittsburgh's Old Steel Mills into an AI Hub. by @Bell_curve
- Elon Musk's xAI launches Grok 4 alongside a $300 monthly subscription by @ch0k1
- Intel spins out AI robotics company RealSense with $50 million raise by @ch0k1
- EU Commission Releases "Voluntary" Tool General-Purpose AI Code of Practice by @Cje95
- America tests AI powered fighter jet drones by @jakoyoh629
- America's largest power grid is struggling to meet demand from AI by @Coinsreporter
- Dubai to debut restaurant operated by an AI chef by @Coinsreporter
- Cops’ favorite AI tool automatically deletes evidence of when AI was used by @0xbitcoiner
- OpenAI’s Windsurf deal is off, and Windsurf’s CEO is going to Google by @hn