{"slug": "roundup-new-interview-llm-flip-flopping-cyberattacks-jobs-etc", "title": "Roundup: New Interview, LLM Flip-Flopping, Cyberattacks, Jobs, etc.", "summary": "Goldman Sachs raised its humanoid robot sales forecast by 5x in a new report, citing faster-than-expected adoption in logistics and manufacturing and rapid progress in robot learning data. The roundup also highlighted an arXiv paper finding that quality-adjusted inference prices are falling roughly seven times faster than conventional measures show while cost per completed task has stopped falling, because reasoning models consume more tokens as capability rises. Other items covered models easily flipping response states, AI-driven cyberattacks on critical infrastructure, and signs of AI automation in declining Texas job postings.", "body_md": "### What's Ahead\n\n1. **\"Big Tech\" Interview**\n2. **The Price of Intelligence**\n3. **Goldman Does a 5x on Its Humanoid Robot Forecast**\n4. **Models Easily Flip Response States**\n5. **AI Is Driving a Wave of Cyberattacks on Critical Infrastructure**\n6. **A New World Inflationary Order, Driven by CapEx and Duplication**\n7. **Signs of AI Automation in Declining Texas Job Postings**\n\n### \"Big Tech\" Interview\n\nA new interview with me that I should have mentioned in the last roundup.\n\n### The Price of Intelligence\n\nQuality-adjusted inference prices are falling roughly **seven times faster** than conventional measures show, yet **the cost per completed task** has stopped falling. What gives?  In short, reasoning models are consuming more tokens as capability rises. Why? Because cheaper intelligence is being spent immediately on more computation, masking the expensive [flatlining of models](https://paulkedrosky.com/why-400-hitters-disappeared-and-what-it-means-for-ai/) by **throwing more tokens** at the problem.\n\n**Source:** [The Price of Intelligence: A Quality-Adjusted Price Index for AI Services](https://arxiv.org/pdf/2608.29843)\n\n### Goldman Does a 5x on its Humanoid Robot Forecast\n\nIn one of the largest upgrades I have seen since the dot-com days, Goldman Sachs has, in a new report upped its forecast for **humanoid robot sales by 5x**. The firm argues that **adoption is faster than expected**, especially in logistics and manufacturing, and **robot learning data** is moving quickly as well. Expect this to be fundamental to the bull case over the next 24 months as large language models come to be seen as plateauing and limited.", "url": "https://wpnews.pro/news/roundup-new-interview-llm-flip-flopping-cyberattacks-jobs-etc", "canonical_source": "https://paulkedrosky.com/roundup-new-interview-llm-flip-flopping-cyberattacks-jobs-etc/", "published_at": "2026-09-01 21:06:10+00:00", "updated_at": "2026-09-17 00:54:31.905451+00:00", "lang": "en", "topics": ["robotics", "artificial-intelligence", "large-language-models", "ai-safety"], "entities": ["Goldman Sachs", "arXiv", "Texas"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/roundup-new-interview-llm-flip-flopping-cyberattacks-jobs-etc", "markdown": "https://wpnews.pro/news/roundup-new-interview-llm-flip-flopping-cyberattacks-jobs-etc.md", "text": "https://wpnews.pro/news/roundup-new-interview-llm-flip-flopping-cyberattacks-jobs-etc.txt", "jsonld": "https://wpnews.pro/news/roundup-new-interview-llm-flip-flopping-cyberattacks-jobs-etc.jsonld"}}