University of Pennsylvania
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 4
- Studies: AI-altered images and chatbots can implant false memories
According to research rounded up by The Washington Post, MIT Media Lab professor Pattie Maes and colleagues showed 24 everyday photos to 200 participants; after a break, some saw AI-altered versions with changed facial expressions, added soldiers or climate imagery, and those participants were far more likely to form false memories — and more confident in them. In other experiments, participants questioned by a generative AI chatbot after watching a crime video produced about three times as many false memories as a control group, and nearly half falsely remembered remake films that were never made. Georgetown researcher Mattea Sim's 331-participant experiment found about 45% of those given an AI summary with a wrong traffic sign detail correctly remembered the sign, versus about 84% given an accurate summary, even when told the summary was AI-generated.
IT之家 AI · 🔥 17 - Wharton Study: Most Firms Can't Prove AI Spending Pays Off
A new study from the University of Pennsylvania's Wharton School and consultancy GBK Collective finds that most companies still gauge AI's return by worker efficiency rather than revenue growth, making the payoff hard to demonstrate in financial statements. The study drew on online surveys of 1,400 U.S.-based employees at firms with at least 1,000 staff and more than $50 million in annual revenue. It found 40% of managers measured AI's impact as "hours saved per employee" and 37% tracked error reduction, while only about 27% tracked revenue from customer-facing AI features, apps and tools; 44% said AI's main role over the next one to two years will be productivity, and just 8% expected new revenue or new products.
The Information · 🔥 0 - AgentWorld benchmark: best multi-agent team solves just 52% of tasks
OpenAgents, Columbia University, the University of Pennsylvania, Seoul National University and Penn State built AgentWorld, a benchmark for multi-agent collaboration that puts LLM-driven agent teams into an MMORPG sandbox with 100 hand-designed tasks plus 100 enhanced variants covering division of labour, communication and resource handoffs. Under a shared prompt template, tool set and interaction protocol, Gemini 3 Flash, Claude Haiku 4.5, GPT-5 Mini and DeepSeek R1-70B scored 52.0%, 45.0%, 36.0% and 20.0% on the main task set, with the enhanced set topping out at 26.0%. The paper also introduces a Causal Collaboration Effectiveness (CCE) metric that traces which actions and messages actually contributed to the shared goal.
36氪 人工智能 · 🔥 0 - OpenAI to watermark ChatGPT and Codex text in the EU
OpenAI laid out how it will comply with Article 50 of the EU AI Act: over the coming weeks it will automatically add an invisible watermark to eligible ChatGPT and Codex text output in the EU across all plans, while API customers worldwide can opt in for select models starting today (off by default). The method, called textGrain, subtly shapes the model's word choices to embed a statistical signal that a detector can read; detector access will initially be limited to approved researchers and expert organizations, and OpenAI plans to open-source the technology. OpenAI's evaluations put detection at about 80% for 200-token passages and about 95% for 400-token passages at a 1% false-positive rate, lower for content like mathematics, and replacing 10% of words with synonyms cut detection from about 92% to 66%. The company said it saw no meaningful benchmark differences for its latest frontier model, Astra, with watermarking on.
OpenAI · 🔥 0
Experience and discussion from the community
Share my University of Pennsylvania experienceAsk about University of Pennsylvania
Nobody has shared their experience with University of Pennsylvania yet.