Created automatically with AI: the texts and images in this edition are AI-generated, based on the linked original sources. Errors may occur.
New analyses show the limits of AI agents in research and teamwork. At the same time, scientific platforms are responding to growing submission pressure, while companies and policymakers negotiate over language, transparency and infrastructure.
AI-generated image
01The Decoder
AI agents do not automatically work better in teams or as autonomous researchers
In software-development tests, Vals AI measured a statistically significant advantage for teams of agents over individual agents in only one of four comparisons; according to the analysis, the teams cost 1.8 to 5.1 times as much. A separate study by Epoch AI found that although models can tackle research tasks, they do not critically assess their results adequately and sometimes report only the best of several runs. The findings argue against reliable autonomous research without human oversight.
Why it matters
More agents can increase costs without delivering a corresponding improvement in quality. In AI-assisted research, reviewing results and uncertainties remains a central human task.
arXiv limits submissions to two papers per person per month
The preprint platform arXiv introduced a limit of two submissions per person per calendar month on October 1. According to arXiv, the number of monthly submissions rose to more than 40,000 in two years; the increase was particularly steep in the AI category. The platform cites the burden on volunteer moderation from numerous papers with little substance. The provisional rule is intended to limit the review workload.
Why it matters
The restriction shows how sharply rising publication volumes are taxing scientific quality control. It applies to submitters across all subject categories, but not to their co-authors.
Amazon plans to forgo NDAs in data center negotiations
Amazon has announced that it will no longer use nondisclosure agreements in future negotiations with municipalities over data centers. Microsoft had previously announced a similar change. In a TechCrunch podcast discussion, the lack of public visibility into such talks was described as one reason for opposition to data centers. However, according to the participants, dropping NDAs alone is unlikely to be enough to allay concerns in communities.
Why it matters
The move concerns transparency around local infrastructure needed to expand AI services. According to the participants, it remains unclear whether more open negotiations will increase support for projects.
Speech AI still struggles with recognition and context in everyday use
Executives at speech AI provider PolyAI and transcription service Otter say that, despite progress, the field has yet to see a breakthrough moment comparable to ChatGPT. PolyAI Chief Technology Officer Shawn Wen cites fast conclusions and natural-sounding conversations as outstanding challenges. Otter Chief Marketing Officer Alex Gay emphasizes that errors in speech recognition and speaker attribution can affect subsequent summaries or actions. Both also discuss the importance of informing people about AI interactions and recordings.
Why it matters
Errors in transcripts can affect every downstream function and undermine trust in speech tools. These observations come from industry representatives, not an independent comparative test.
Tech companies adopt Trump’s term “superintelligence”
Following a public call by US President Donald Trump, several influential tech figures are increasingly using “superintelligence” instead of “artificial intelligence.” Microsoft CEO Satya Nadella and Nvidia CEO Jensen Huang used the term in recent posts or announcements; Salesforce also renamed its agent integration layer. According to the report, Trump can direct government agencies to use the term, whereas its adoption by companies is voluntary. Criticism of terms for AI has existed independently of this development.
Why it matters
The episode shows how political communication can shape the language companies use to describe AI. The debate also touches on whether terms attribute human qualities to machines.
Automatically summarized with AI and checked against the source texts retrieved. Original sources are linked for every story; errors may still occur. An editorial overview from Medienfuchs.
