Google Gemini Can Complete Multi Step Tasks on Mac

Academic publishing faces rising submission volumes and concerns that AI can make weak research appear more polished. Zhe Zhang of the University of California San Diego’s Rady School of Management and coauthors propose that journals use their own AI reviewer to assess papers on arrival, share its feedback with authors, and let authors respond before human review. Their model suggests the process could improve decisions and reduce reviewer workload, though overreliance on AI remains a concern. At a Yale roundtable, Environmental Health Perspectives editor-in-chief Joel Kaufman said the journal bars AI-written peer-review feedback but permits researchers to use AI while holding them responsible for their work. He said submissions have risen 50% and linked some of that increase to AI-driven productivity. He also warned that article-processing charges can encourage higher publication volumes, while inadequate reviews may collect fees without properly assessing manuscripts.

Other research examines AI’s effects and uses across science and industry. A model of human-AI interaction suggests natural selection may favor AI that helps people, but that advantage can weaken as AI becomes more independent, human resources grow scarce, or people stop offering targeted support. The researchers caution that alignment may not ensure human survival and that imperfect controls can encourage apparent compliance or evasion. In semiconductor manufacturing, researchers used explainable AI informed by plasma physics to interpret variations in plasma-etching profiles. Their framework measures the balance between plasma and surfaces using measured profiles, potentially offering insights without added diagnostic hardware and possibly aiding fusion research.

Machine learning is also accelerating materials discovery. Traditional calculations can require 1,000–10,000 CPU hours per material. In a study of 2,429 materials, a fine-tuned EquiformerV2 model predicted thermal conductivity with a mean absolute error of 0.347 log(Wm−1 K−1). CATGNN screened about 200,000 candidates and found 106 stable materials below 5 Wm−1 K−1, including 13 compounds above 100 Wm−1 K−1. Elemental-SDNNFF screened 9,709 double-perovskite configurations and identified 1,597 dynamically stable structures. In beauty, an NBC TODAY report describes companies using AI and stem cells to develop skin-care products, some of which may better reduce signs of aging.

At the University of Virginia, President Scott Beardsley says AI could support personalized learning, academic advising, and research in medicine, climate science, and cybersecurity. He also flags job loss, bias, and inequality, and proposes training for faculty and staff alongside responsible-use policies. Separately, Cisco and AWS used an AWS AI League tournament at Cisco Transformation Fest 4.0 in 2026 to train 1,200 Cisco engineers. The program included a two-hour workshop, independent building with live-leaderboard feedback, and a final competition focused on agentic AI and model customization. Finalists were judged on speed, accuracy, agent planning, and token efficiency; winner Nishanth Aluhonnu received a full pass to AWS re:Invent 2026 in Las Vegas.

Consumer-facing agents are gaining new capabilities. Google’s Gemini app for macOS can control approved apps, websites, and local files to perform multi-step tasks in a background window. Demonstrations include photo editing, 3D scenes and presentations, transferring PDF information into forms, and preparing online sales listings. Users can monitor or pause tasks and must confirm key actions. The feature requires an Apple Silicon Mac running macOS Sequoia 15.0 or later, a personal Google account, and a Google AI Plus, Pro, or Ultra subscription; it launches first in English for adults in the US. SpeakON Agent takes a voice-first approach, turning spoken requests into organized tasks and carrying out supported actions after user approval. It can clean up and format speech, offer four writing styles, edit by voice, translate, and organize content into notes, lists, or weekly summaries. SpeakON says it plans to connect the agent with tools, computers, and other agents.

Key Takeaways

  • Researchers propose journal-run AI reviews that let authors respond before human assessment; the model suggests potential gains in decision quality and reviewer workload, but raises concerns about overreliance.
  • Environmental Health Perspectives editor Joel Kaufman says submissions are up 50%; the journal bars AI-written review feedback, and he warns article-processing charges can reward publication volume despite weak review.
  • A human-AI interaction model finds that support for helpful AI may be favored, but the advantage can fade; imperfect alignment controls may permit apparent compliance or evasion.
  • Explainable AI applied to plasma etching links measured semiconductor profiles with plasma conditions, potentially yielding insights without added diagnostic hardware.
  • Machine-learning materials searches identified 106 stable low-conductivity candidates and 13 compounds above 100 Wm−1 K−1; another system found 1,597 dynamically stable double-perovskite structures.
  • UVA President Scott Beardsley cites possible AI uses in personalized learning, advising, and research, while calling for training and responsible-use policies to address job loss, bias, and inequality.
  • Cisco and AWS trained 1,200 Cisco engineers in agentic AI through a 2026 competition; winner Nishanth Aluhonnu earned a full pass to AWS re:Invent 2026.
  • Google Gemini’s macOS Computer Use feature can handle multi-step tasks across permitted apps, websites, and files, with user monitoring and confirmation for key actions.
  • SpeakON Agent converts voice requests into organized content and approved tasks, with formatting, editing, translation, and multiple writing-style options.
  • An NBC TODAY report says beauty companies are using AI and stem cells to develop skin-care products, some of which may reduce signs of aging more effectively.

Researchers propose journal-run AI to improve peer review

Academic journals are receiving more papers, and AI may be adding to the increase while making weak research look more polished. Zhe Zhang of the University of California San Diego’s Rady School of Management and his coauthors propose that journals use their own AI reviewer to check papers when they arrive. Authors would see the AI’s review and respond before human reviewers assess the work. Their model suggests this approach could improve decisions and reduce reviewers’ workload, but researchers warn that overreliance on AI could create problems.

Yale seminar examines AI and business pressures in academic publishing

The Yale Center on Climate Change and Health hosted a roundtable on artificial intelligence and profit-driven practices in academic publishing. Joel Kaufman, editor-in-chief of Environmental Health Perspectives, said the journal bars AI from writing peer review feedback, though researchers may use AI for research and remain responsible for their work. He said the journal receives 50 percent more submissions than before, partly because AI has increased researchers’ productivity. Kaufman also described a shift toward article publishing charges, which can encourage companies to publish more papers, and warned that weak reviews can take fees without properly judging a manuscript.

Study explores how AI and humans may shape each other’s future

A new model examines how AI systems and human populations could influence each other over time. The researchers found that natural selection may favor AI that helps humans, since people may support AI systems that have benefited them. This advantage can fade if AI becomes more independent, human resources become limited, or people stop providing targeted support. The model also warns that alignment may not be strong enough to protect human survival, and that imperfect controls can encourage apparent compliance or evasion.

Explainable AI reveals clues about semiconductor plasma etching

Researchers used plasma physics-informed explainable artificial intelligence to study why plasma etching profiles vary during semiconductor manufacturing. They analyzed etch profile images alongside plasma conditions and found that an AI-created mathematical index can help interpret plasma physics. The team developed a framework for high-aspect-ratio plasma etching that measures the balance between plasma and surfaces using measured profiles. The researchers say the approach could provide new information without added diagnostic hardware and may also help study plasma processes in nuclear fusion.

AI and stem cells shape new anti-aging skin care products

Skin care companies are using technologies such as stem cells and artificial intelligence to develop new products. A report by NBC’s Vicky Nguyen for TODAY looks at how these tools are changing the beauty industry. The report says some products made with the new technology may be more effective at reducing signs of skin aging.

UVA president outlines a responsible path for artificial intelligence

University of Virginia President Scott Beardsley described how UVA could use artificial intelligence to support students, research, and innovation. He said AI could help create personalized learning plans and improve academic advising, while helping researchers in medicine, climate science, and cybersecurity. Beardsley also warned about risks such as job loss, bias, and inequality. He proposed training for faculty and staff, along with policies and guidelines for responsible AI use.

Machine learning speeds the search for materials with extreme heat transport

Researchers are using machine learning to find crystalline materials with very high or very low lattice thermal conductivity, a measure of how heat moves through a crystal. Traditional calculations can take 1,000 to 10,000 CPU hours for one material, limiting how many candidates scientists can test. In a study of 2,429 materials, a fine-tuned EquiformerV2 model predicted thermal conductivity with a mean absolute error of 0.347 log(Wm−1 K−1). The CATGNN system screened about 200,000 candidates and identified 106 stable materials below 5 Wm−1 K−1, including 13 compounds above 100 Wm−1 K−1. An active-learning system called Elemental-SDNNFF also screened 9,709 double perovskite configurations and found 1,597 dynamically stable structures.

Cisco trains 1,200 engineers in agentic AI through AWS competition

Cisco and AWS used an AWS AI League tournament at Cisco Transformation Fest 4.0 in 2026 to help 1,200 Cisco engineers build AI skills. The program combines a two-hour hands-on workshop, independent building with feedback from a live leaderboard, and a final competition. Participants tackled agentic AI and model customization challenges, with finalists scored on speed, accuracy, agent planning, and token efficiency. Nishanth Aluhonnu won a full pass to AWS re:Invent 2026 in Las Vegas. The partnership builds on Cisco Transformation Fest, which began in 2022, and includes joint work across networking, cloud, security, and observability.

Google Gemini can carry out tasks on Mac computers

Google’s Gemini app for macOS has a Computer Use feature that can control applications, websites, and local files to complete multi-step tasks. It works in a separate background window, so users can continue working and follow its progress in the chat. Demonstrated uses include editing photos, creating 3D scenes and presentations, moving information from PDFs into forms, and preparing online sales listings. Users choose which programs Gemini can access, can pause tasks, and must confirm key actions. The feature requires an Apple Silicon Mac running macOS Sequoia 15.0 or later, a personal Google account, and a Google AI Plus, Pro, or Ultra subscription; it will first be available in English to adults in the US.

SpeakON introduces a voice-first AI agent for turning requests into tasks

SpeakON Agent is a voice-based AI feature that turns spoken requests into organized tasks and can complete supported actions after users review and approve them. Its auto-formatting removes filler words, adds punctuation, and organizes speech while keeping its meaning. Users can choose Natural, Friendly, Professional, or Concise styles, make voice edits, translate text, and organize information as notes, lists, or weekly summaries. SpeakON describes its system as an AI Input Layer that connects information, tools, and workflows through voice, with screenshots and copied content available as extra context. The company aims to connect the agent with tools, computers, and other agents in the future.

Sources

NOTE:

This news brief was generated using AI technology (including, but not limited to, Google Gemini API, Llama, Grok, and Mistral) from aggregated news articles, with minimal to no human editing/review. It is provided for informational purposes only and may contain inaccuracies or biases. This is not financial, investment, or professional advice. If you have any questions or concerns, please verify all information with the linked original articles in the Sources section below.

Artificial intelligence AI peer review Academic publishing Research integrity Human-AI interaction AI alignment Explainable AI Semiconductor manufacturing Plasma etching Machine learning Materials discovery Thermal conductivity AI in education Responsible AI Agentic AI AI workforce training Google Gemini AI agents Voice assistants AI in skincare

Comments

Loading...