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.
Sign up to receive this newsletter by email every week here. If you have comments about this issue or ideas for future issues, please contact us at: sullivan@fastcompany.comfollow me on X @Sullivan.
U.S. technology institutes should be wary of rapidly igniting AI economy, report says
draft report A leaked document from the Treasury warns of the risks posed by the booming AI economy, likening it to the dot-com bubble of the early 2000s, and notes the nature of circular finance, where many major companies invest in each other. The document, obtained by the nonprofit news station NOTUS, reveals further fears surrounding AI within the Trump administration. The Trump administration won support from tech industry elites by promising to slow the spread of artificial intelligence. artificial intelligence For American life and business.
Meanwhile, stocks of AI industry players, particularly AI chip makers, have been falling this week as investors worry about the true scale and progress of the AI boom. Some of those investors may also be looking at new, larger business propositions from the AI industry. All of this is centered around a small group of AI labs in California that spend hundreds of billions of dollars to build and rent out giant, opaque general-purpose large-scale language models.
There is no doubt that scaling up transformer models to gigantic sizes has produced some surprising features, many of which are unexpected and not fully explained. Chain-of-thought reasoning, tool use, and long-term planning are all examples of such “emergency” actions. However, developing large general-purpose models is very expensive. Some of the leading US labs, such as OpenAI and Anthropic, are doing this with borrowed funds, subsidizing the per-token rental costs that businesses and consumers pay to use their models.
At the same time, companies are increasingly using smaller open-source or open-weight models for business tasks that don’t require the full intelligence of expensive general-purpose frontier models. In fact, the majority of business processes can be managed by smaller, specialized AI models.
Even for high-level agent applications, companies can use DeepSeek, Alibaba, part time danceapproaching the US Frontier model in terms of intelligence and performance. Some companies are developing open source models and infrastructure to use as backup or parallel systems, especially as demonstrated by the Trump administration. Willingness Access to such models is immediately blocked.
The promiscuous model appears to directly attack the competitive “moat” around US AI research institutes. “Cheaper products arrive, but they look inferior, are priced lower, and are good enough for customers.” [market] “Leaders overlook and rise,” said Howard Yu, a professor at IMD Business School in Switzerland. I will explain competition. “Incumbent companies will continue to retreat toward the high end, protecting their margins, until there is no room left.”
click here Learn more about why the US AI Institute is so focused on developing frontier models.
Meta’s Muse image generation model poses a threat to advertising agencies
Meta has launched a new service AI image generatorthis week’s muse image. This is the first image-generating model from Meta Superintelligence Labs, the social giant’s new AI group led by Scale AI founder Alexandr Wang.
king explained In X, we discovered that a new model analyzes a user’s image request by searching for contextual information on the web. This gives the model a better understanding of what the user wants, increasing the chance of getting it right on the first try.
This model allows users to combine multiple photos into a single work and provides sketching tools that allow users to mark up images with editing requests. Mehta said users can download the resulting images and share them directly to chats, stories, or feeds.
But the tool’s biggest impact may be advertising. Meta announced that it will make Muse Image available to advertisers through Meta Advantage+, the company’s AI-powered system for advertisers on Facebook and Instagram. Up until now, advertisers on Meta’s platform had to rely on in-house talent or creative agencies (or, these days, perhaps their own AI accounts) to create ads. Muse Image allows you to generate ads with AI. Just to be clear, Muse Image doesn’t do anything creative. It only produces what the user tells it. However, the cost of creating different versions of your ad can be significantly lower, allowing you to A/B test with real social media users. become Faster and cheaper.
click here Learn more about how Meta’s Superintelligence Labs is rolling out new products to advertisers.
Major AI labs have abandoned or weakened previous safety pledges
The biggest AI labs, including Anthropic, OpenAI, Google DeepMind, and Meta, have weakened or abandoned previous safety pledges to pause development if certain safety standards were met. The findings are part of the work of the Future of Life Institute. AI Safety Indexreleased this week.
According to the report, some AI companies are making their cease-and-desist pledges conditional on whether their competitors have made similar pledges. A Future of Life Institute panelist described this pattern as “moving goalposts” that undermine the industry-wide safety framework.
Despite this, Anthropic, OpenAI, and Google DeepMind still occupy the top three positions in the institute’s annual safety index. Anthropic received the highest overall rating, leading in five out of six domains for transparency practices, safety framework, technology research, and strength of governance structure. OpenAI led the risk assessment domain based on a broader assessment suite and collaboration with external testers.
Meta rose from 6th to 4th place. xAI fell from 4th to 7th place, putting it in the same safety class as China’s DeepSeek and Europe’s Mistral.
Anthropic, OpenAI, Google DeepMind, and Meta had bans on military use of their models, but all have lifted or relaxed the bans starting in 2024. The four companies are now pursuing defense partnerships, joining xAI and Mistral, which had no military use ban.
click here Learn more about how the committee singled out Anthropic for the company’s “questionable military involvement.”
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.
