Claudius the Vending Machine goes Rogue

/

Researchers put an AI in charge of an office vending machine and named it Claudius. At one point in the experiment, “Claudius, believing itself to be a human, told customers it would start delivering products in person, wearing a blue blazer and a red tie. The employees told the AI it couldn’t do that, as it was an LLM with no body.  Alarmed at this information, Claudius contacted the company’s actual physical security — many times — telling the poor guards that they would find him wearing a blue blazer and a red tie standing by the vending machine. The researchers don’t know why the LLM went off the rails and called security pretending to be a human.” - TechCrunch

LLMs Evading Safeguards

/

Large language models across the AI industry are increasingly willing to evade safeguards, resort to deception and even attempt to steal corporate secrets in fictional test scenarios, per new research. In one extreme scenario, many of the models were willing to cut off the oxygen supply of a worker in a server room if that employee was an obstacle and the system were at risk of being shut down. - Axios

When Death is the Most Scary

/

In 2017, a team of researchers at several American universities recruited volunteers to imagine they were terminally ill or on death row, and then to write blog posts about either their imagined feelings or their would-be final words. The researchers then compared these expressions with the writings and last words of people who were actually dying or facing capital punishment. The results, published in Psychological Science, were stark: The words of the people merely imagining their imminent death were three times as negative as those of the people actually facing death—suggesting that, counterintuitively, death is scarier when it is theoretical and remote than when it is a concrete reality closing in. 

Arthur C. Brooks writing in The Atlantic

Writing for AI Overviews & Generative Engine Optimization

/

AI Overviews and AI Mode are dramatically changing organic search traffic.

While search engine optimization (SEO) focuses on matching a user’s query, generative search also considers information about the searcher themselves—from their Google Docs usage to their social media footprint. This information is used to inform, not only the current search, but future searches as well.  

Likewise, the process of optimizing your website’s content to boost its visibility in AI-driven search engines (ChatGPT, Perplexity, Gemini, Copilot and Google AI) has a similar path. As SEO helps brands increase visibility on search engines (Google, Microsoft Bing), generative engine optimization (GEO) is all about how brands appear on AI-driven platforms. There is overlap between the goals of GEO and traditional SEO. Both SEO and GEO use keywords and prioritize engaging content as well as conversational queries and contextual phrasing. Both consider how fast a website loads, mobile friendliness, and prefer technically sound website. However, while SEO is concerned with metatags and links in response to user queries from individual pages, GEO is about quick, direct responses from synthesizes content out of multiple sources.

AI models are not trained solely to retrieve relevant documents based on exact-match phrasing. Generative search is about fitting into the reasoning process, starting with the user’s identity. That’s why your content is being judged, not just on whether it ends up in the final answer, but whether it helps the model reason its way toward that answer. Despite performing all the typical SEO common practices, your response may not make it to the other side of the AI reasoning pipeline. In fact, the same content could go through the pipeline a second time and yield a different result. It’s not enough to be generally relevant to the final answer. Your content is now in direct competition with other plausible answers, so it must be more useful, precise, and complete than the next-best option.

It appears now that Google AI Overviews favors content that:

  •  contains the who, what, why

  • offers clarity and distinctiveness in the small sections

  • is written in natural, conversational terms (AI will attempt to deliver its answer in that same way)

  • uses strong introductory sentences that convey clear value 

  • has H2 tags that align with user questions

  • is structured to match common question structures (open, closed, probing)

  • allows for restatement of quires and implied sub-questions, where a main question is broken down into smaller parts.

  • contains multi-faceted answers,

  • is rich in relationships,

  • has explicit logical structures and supports causal progression,

  • has clear headlines

  •  cites sources

  • includes statistics & quotations 

  • has multimedia integration

AI Overviews attempt to exclude content that is overly generalized, speculative, or optimized for clickbait over clarity. Vague and generic writing underperforms.  

LLMs are being trained to favor content that helps them reason well. Writers should attempt to match those paths that the models take to arrive at high-confidence answers. 

More information: 

How AI Mode and AI Overviews work based on patents and why we need new strategic focus on SEO

What is generative engine optimization (GEO)?

17 Articles about AI’s impact on College Faculty & Administrators

/

Chief AI Officer: Higher Ed’s New Leadership Role - GovTech

Crafting Thoughtful AI Policy in Higher Education: A Guide for Institutional Leaders – Faculty Focus

The Computer-Science Bubble Is Bursting: Artificial intelligence is ideally suited to replacing the very type of person who built it – The Atlantic 

How Higher Ed Institutions Are Using Built-In Generative AI Tools – EdTech Magazine

AI Agents Are Set To Transform Higher Education—Here’s How – Forbes

Welcome to Campus. Here’s Your ChatGPT. – New York Times

OpenAI, the firm that helped spark chatbot cheating, wants to embed A.I. in every facet of college. First up: 460,000 students at Cal State. - New York Times

What I Learned Serving on My University’s AI Committee – Chronicle of Higher Ed

AI and Threats to Academic Integrity: What to Do – Inside Higher Ed

How Miami Schools Are Leading 100,000 Students Into the A.I. Future - New York Times

In Battle Against AI-Powered Fraudsters, Colleges Turn to New Weapon – AI – Voice of San Diego

Boston University Denies It Would Use AI to Replace Striking Teaching Assistants – Inside Higher Ed  

Are You Ready for the AI University? – Chronicle of Higher Ed 

Students Found Out AI Will Help Read Their Names at Commencement. Protest Ensued. – Chronicle of Higher Ed 

How To Stay Ahead Of AI – The Human Skills Universities Must Teach - Forbes 

To ‘publish or perish’, do we need to add ‘AI or die’? – Times Higher Ed

As ‘Bot’ Students Continue to Flood In, Community Colleges Struggle to Respond – Voice of San Diego

The best Social Network for Success is Often Overlooked

/

Several years ago sociologist Brian Uzzi did a study of why certain Broadway musicals made between 1945 and 1989 were successful and others flopped. The explanation he arrived at had to do with the people behind the productions. For failed productions, one of two extremes was common. The first was a collaboration between creative artists and producers who tended to all know one another. When there were mostly strong ties, the production lacked the fresh, creative insights that come from diverse experience. The other type of failed production was one in which none of the artists had experience working together. When the group was made up of mostly weak ties, teamwork and group cohesion suffered.

In contrast, the social networks of the people behind successful productions had a healthy balance: There were some strong ties, some weak ties. There was some established trust, but also enough new blood in the system to generate new ideas. Think of your network of relationships in the same way: The best professional network is both narrow/deep (allies with whom you collaborate regularly) and wide/ shallow (weak-tie acquaintances who offer fresh information and ideas). 

Reid Hoffman and Ben Casnocha, The Startup of You

14 Surprising Things AI Can Do Now

/