Innovation Profs - 9/30/2025

Your weekly guide to generative AI tools and news

Author

Innovation Profs
September 30, 2025

Latest Gen AI News

Microsoft launches ‘vibe working’ in Excel and Word

Step aside, vibe coding. ‘Vibe working’ is now a thing, at least according to Microsoft, which introduced the term when it announced on Monday the launch of two new Copilot tools, Agent Mode and Office Agent. “Agent Mode essentially takes a complex task and breaks it down with planning and reasoning that you can follow. It then uses OpenAI’s GPT-5 model to break down each step of document creation into an agentic task and execute it.” Meanwhile, Office Agent leverages Anthropic’s Claude models to produce Powerpoint presentations or Word documents from an initial prompt. The inclusion of Anthropic models is particularly noteworthy, as it’s the most visible sign that the Microsoft-OpenAI partnership is not as strong as it formerly was.

Anthropic launches Claude Sonnet 4.5, its latest AI model that’s ‘more of a colleague’

Speaking of Anthropic, Monday was also an eventful day for the LLM company, as it announced its latest model Claude Sonnet 4.5, which it has hailed as “the best coding model in the world” (a title Anthropic had previously attributed to Claude Opus 4). Reportedly, it can code autonomously for over 30 hours!

By way of reminder, ever since the launch of version 3 in March 2024, Claude has come in three sizes: Haiku, Sonnet, and Opus (in order from smallest to largest). Since the launch of Claude Sonnet 3, the updated versions of the model include 3.5, 3.7, 4, and now 4.5. Significantly, Anthropic claims that although Sonnet 4.5 is smaller than Opus 4.1, it is smarter “in almost every single way.” Sonnet 4.5 will now be the default model for all users.

OpenAI takes on Google, Amazon with new agentic shopping system

Sign up for Oct. 14 ChatGPT Workshop

Our next workshop will help you Become a ChatGPT Power User. We will explore all of the features of ChatGPT - from Deep Research to Custom GPTs to Image and Video Creation to integration with tools like Google Drive and Gmail. Attend in-person at Drake University or virtually.

What: Become a ChatGPT Power User
When: Oct. 14, 9 a.m. to 2:30 p.m.
Price: Early bird rate of $125 through Oct. 7 (Use code EARLY to save an additional $25 if you sign up by 5 p.m. Oct. 1)
Where: Drake University or attend virtually
Sign up now!

Quick Hits

OpenAI is testing a new GPT-5-based AI agent "GPT-Alpha"

OpenAI is internally testing a new version of its AI agent, which uses a special version of GPT-5 dubbed "GPT-Alpha."

www.bleepingcomputer.com/news/artificial-intelligence/openai-is-testing-a-new-gpt-5-based-ai-agent-gpt-alpha

Big AI firms pump money into world models as LLM advances slow

World models aim to navigate the physical world by learning from videos, robotic data.

arstechnica.com/ai/2025/09/big-ai-firms-pump-money-into-world-models-as-llm-advances-slow

Why 99% of companies fail at AI integration — and how to join the 1% that succeed

In boardrooms and breakrooms alike, AI is undoubtedly a core topic. Yet, future-proofing a business with AI means more than dabbling in chatbots or automating a few tasks – it requires making AI fluency a foundational skill across the organization.

venturebeat.com/ai/why-99-of-companies-fail-at-ai-integration-and-how-to-join-the-1-that

Generative AI might end up being worthless — and that could be a good thing

GenAI does some neat, helpful things, but it’s not yet the engine of a new economy — and it might not ever be.

theconversation.com/generative-ai-might-end-up-being-worthless-and-that-could-be-a-good-thing-266046

U.S. rejects international AI oversight at U.N. General Assembly

While heads of state and world experts highlighted risks and opportunities from AI, U.S. representatives expressed deep opposition to any global effort to govern the technology.

www.nbcnews.com/tech/tech-news/us-rejects-international-ai-oversight-un-general-assembly-rcna233478

Gemini in Google Sheets is now a formula-writing genius that explains its own work

For many of us, writing complex spreadsheet formulas can feel like learning a new language. It can be an intimidating and frustrating process of trial and error. In a fantastic new update, Google is making that process a whole lot easier by rolling out a significant upgrade to formula generation in Gemini for Sheets. This...

chromeunboxed.com/gemini-in-google-sheets-is-now-a-formula-writing-genius-that-explains-its-own-work

Lionsgate Is Finding Out It's Really Hard to Make Movies With AI

Four 'John Wick' movies just isn't enough to train an AI model, apparently.

gizmodo.com/lionsgate-is-founding-out-its-really-hard-to-make-movies-with-ai-200066322

DeepSeek releases 'sparse attention' model that cuts API costs in half | TechCrunch

Researchers at DeepSeek released a new experimental model designed to have dramatically lower inference costs when used in long-context operations.

techcrunch.com/2025/09/29/deepseek-releases-sparse-attention-model-that-cuts-api-costs-in-half

AI 'Workslop' Is Plaguing American Companies, Says Stanford Research

Over 40% of US-based full-time employees reported receiving “workslop” in the last month, defined as AI-generated content that "masquerades as good work but lacks the substance to meaningfully advance a given task."

www.pcmag.com/news/ai-workslop-is-plaguing-american-companies-says-stanford-research

To Understand AI, Watch How It Evolves | Quanta Magazine

Naomi Saphra thinks that most research into language models focuses too much on the finished product. She’s mining the history of their training for insights into why these systems work the way they do.

www.quantamagazine.org/to-understand-ai-watch-how-it-evolves-20250924

This AI film took 200 creatives and 1000 prompts per image | CNN Business

In her new short film “Ancestra”, Eliza McNitt tells the story of her own birth using generative AI, with the help of 200 creatives and a process that took up to 1000 prompts per image.

www.cnn.com/2025/09/29/business/video/ancestra-eliza-mcnitt-generative-ai-film-spc-digvid

Tool of the week: Suno V5

AI music generation app Suno released version 5 this week with clearer, more immersive audio and vocals that sound natural and authentic. It gives users more creative control over every element of a song while improving genre understanding and mixing for professional-level results.

A standout feature is the ability to remaster old tracks with V5, letting you adjust how much variation you want from the original song. This means you can breathe new life into past work without losing its core feel.

Suno version 5 is available today for paid users of the app. Free user of Suno get access to version 3.5

Suno also recently released Hooks to combines your music and videos into short-form content to share with the Suno community.

AI-generated image of the week

The Cubs are back in the playoffs (first pitch is 2:08 p.m. Central time today), and since we can’t be at the game we decided to send some AI friends to enjoy the experience (and test out how well ChatGPT could adhere to our prompt). 

Prompt: a diverse group of four Chicago Cubs fans cheer on the team in the Wrigley Field stands. One wearing a blue Cubs shirt and hat and holding a Chicago-style hot dog. Another is wearing a white cubs shirt and blue hat and smiling while taking a drink of a beer. The third is wearing a white Cubs hat and holding a sign that says "My first playoff game," and the fourth is holding two cans of Old Style beer in one hand and a baseball scorebook in the other.

What we found

Meta AI is taking on ... well ... Reels? It's rolling out Vibes, a feed of AI videos that’s front-and-center when you open the Meta AI app (and at meta.ai). Users can create and share short-form videos to this feed, where other Meta AI users can like, comment and share. OpenAI is expected to launch a similar app for Sora soon.

Here’s Meta CEO Mark Zuckerberg’s first Vibe: