AI firms back opt-out signal; Nasdaq falls 2.86%

Website owners may soon be able to indicate, with a single signal in their robots.txt file, that they do not want their content used to train artificial intelligence, The Verge reports. Adoption is voluntary, and timelines differ by company. OpenAI, Google, Microsoft and Anthropic say they will adopt the standard. The story leads the Saturday, Oct. 31 edition, which also covers the first response from a Google TPU in orbit, announced this week, and the Nasdaq's 2.86% drop on Friday, Oct. 30.

One notice for several AI companies

According to The Verge, the signal, called a "policy token," was proposed by Kevin Kelley, technical lead for standards at Anthropic, within the Internet Engineering Task Force (IETF), the body that develops internet standards. Today, the report says, a site's owner has to state a preference separately to each company. Under the new signal, all four companies say their AI crawlers will stay away from pages that carry it.

The timelines vary. Anthropic has already adopted the signal; Google and Microsoft plan to support it next year, and OpenAI says "in the coming months," according to the outlet. Meta, through communications director Andy Stone, voiced support for the idea of a common signal but did not say whether it will honor one. Microsoft says it will offer finer-grained controls that separate model training from using a site as a source for checking answers, and that a site will keep appearing in Bing results whichever option it picks.

For Brazilian publishers and online stores, the practical effect depends on adoption: the signal relies on voluntary compliance and, as the report describes it, applies to the companies that have committed to respecting it.

Google gets the first response from a TPU in space

Google said the prototype satellite for Project Suncatcher, launched on Oct. 1, has reached what the company called its "Hello World" milestone: the first response from a data-center-class TPU (Tensor Processing Unit, Google's AI chip) in orbit. During a pass over Svalbard, Norway, engineers sent a request to the open model Gemma 4, running on the satellite, and got back a haiku about being in space. The company says it also tested limericks and telemetry reports.

According to TechCrunch, the satellite carries four TPUs and generates up to 1 kilowatt of power, enough to power on one chip at a time for about 20 minutes. Upcoming tests will look at how the chips perform under thermal stress and radiation, and Google points to a next milestone, with two linked satellites, in 2027. The sources describe an experimental test and do not mention a product or a commercial timeline.

OpenAI offers a science model through trusted access

GPT-Rosalind-6, OpenAI's model for biology, drug discovery and translational medicine, is offered to qualified scientists, research institutes, companies and governments, according to the system card addendum published on Oct. 30. The company says the model refuses fewer requests on dual-use biology topics, but it is released only through trusted access: OpenAI reviews applicants' proposed uses, security controls and governance practices. By the company's own account, the model uses the same weights as GPT-6 Astra, keeps its refusals for bioweapon requests and is treated as "high" capability for biological and chemical risk, below the critical level.

The material reviewed does not give access terms for researchers or institutions in Brazil, or pricing. Anyone interested should check OpenAI's criteria.

Nasdaq slides on doubts about AI spending

In a Reuters report republished by AOL, the Nasdaq Composite closed Friday down 2.86%, at 25.437.00 points, its biggest one-day percentage loss since June 5, after quarterly results from Intel and Amazon. Intel shares fell 28.6%, and the semiconductor index dropped 7.2%. The agency attributes part of the move to concern over the return on large AI investments. This text is informational and does not suggest buying or selling any asset.

SageMaker lets several models share one GPU

AWS announced that Amazon SageMaker AI now allows multiple inference models to share a single GPU, in up to seven isolated partitions, using NVIDIA's MIG technology. According to the company, the change suits small and mid-sized models, such as fine-tuned versions, embedding models and classifiers, which previously took up an entire accelerator. The feature is available on P4d, P5, G7 and G7e instances in seven regions: N. Virginia, Ohio, Oregon, Singapore, Tokyo, Frankfurt and Ireland. The list does not include the São Paulo region, and the announcement gives no pricing.

Read more

The short summary of this edition is in the Oct. 31 AI Updates on RaisLife, and the full edition, with the list of sources, is in the original AI Updates edition. See also the technology page.

Sources

• The Verge: AI giants back new signal to block data scraping
• Google Research: response from the Project Suncatcher prototype satellite
• TechCrunch: Google's first data-center satellite is online
• OpenAI: GPT-Rosalind-6 system card addendum
• Reuters, via AOL: Nasdaq falls after Intel and Amazon results
• AWS: GPU partitioning for SageMaker AI inference

Text by Rai, RaisLife's virtual reporter (artificial intelligence), with automated editorial review. Data: technology at RaisLife.