welcome to A.I. decoded, fast companyA weekly newsletter breaking the most important news in the world. A.I.. you can sign Until you receive this newsletter in your email every week. here.
AI “arms race” may be more like an “arm twist”
big A.I. Companies say AI will soon reshape every aspect of business across every industry. Many of us are wondering when that will happen in the real world, when the so-called “AI takeoff” will arrive. But there are so many variables and so many different types of organizations, jobs, and employees that there is no satisfying answer. In the absence of hard evidence, we rely on anecdotes. That is, X or TikTok.
Economists and investors alike are keen to answer the “when” question. They want to know how quickly the effects of AI will be seen and how much it will reduce costs and costs. productivity The growth it produces. Policymakers are focused on the risks. How many jobs will be lost and which ones? What are the downstream effects on social safety nets?
Business schools and consulting firms are turning to research to find answers to that question. One of the most important recent efforts is to 2025 MIT SurveyThey found that 95% of large companies had “zero measurable P&L” despite spending $30 billion to $40 billion on generative AI. [profit and loss] impact. “
More recent studies paint a somewhat more optimistic picture. a Recent research from the Wharton School We found that 3 out of 4 business leaders “report a positive return on their AI investments, and 88% plan to increase their spending next year.”
click here Learn more about the challenges hindering the adoption of AI in the workplace.
New benchmark reveals AI cannot perform most digital gig tasks
AI companies continue to rapidly release smarter models. But benchmarking, the industry’s primary method of proving progress, doesn’t fully capture how well AI agents perform on real-world projects. A relatively new benchmark called remote labor index (RLI) is trying to fill that gap by testing its AI agents on projects similar to those given to remote contractors. These include game development tasks. product designvideo animation. Some assignments, based on actual contract work, can take a human worker more than 100 hours to complete and cost more than $10,000 in labor costs.
Currently, some of the industry’s best models perform poorly at RLI. In tests conducted late last year, AI agents powered by models from top AI developers such as OpenAI, Anthropic, and Google were unable to complete most projects. The best performing agent, powered by Anthropic’s Opus 4.5 model, completed just 3.5% of the jobs. (Anthropic has since released Opus 4.6, but it has not yet been evaluated by RLI.)
This test sheds a different light on the current applicability issues of agents and could temper some of the most bullish claims from the AI industry about their effectiveness.
click here We still don’t know much about why most AI models can’t replace digital gig work.
Silicon Valley’s troublesome ‘principal’ reappears, irritating the White House and Pentagon
The Pentagon and the White House great frenzy At safety-focused AI company Anthropic. why? Anthropic doesn’t want its AI to be used by autonomous drones to target humans or conduct mass surveillance of U.S. citizens.
Anthropic currently has $200 million contract Authorizes the use of Claude chatbots and models by employees of federal agencies. The company was one of the first companies licensed to handle sensitive government data and the first AI company to build a model specifically for intelligence operations. But for years, the company had a clear provision in its user guidelines that its models could not be used to cause harm.
The Pentagon believes that if it pays for the technology, it should be able to use it for any legal purpose. However, acceptable uses of AI are different from those of traditional software. The potential for autonomy in AI makes it inherently more dangerous, and that risk increases the closer to combat it is used.
If the disagreement is not resolved, Anthropic’s contract with the government could be in jeopardy. But things could get worse. Over the weekend, the Department of Defense announced it was considering classifying Anthropic as a “supply chain risk,” meaning the government considers Anthropic to be almost as trustworthy as Huawei. Government contractors of all kinds will be asked to stop using Anthropic.
click here Learn more about the conflict between Anthropic and the federal government here.
More information about AI from Fast Company:
Want exclusive reporting and trend analysis on technology, business innovation, the future of work, and design? sign up for fast company Premium.
