In partnership with
Welcome to OHMO AI. Newly unredacted filings show what a Microsoft executive wrote in private about AI training, after years of fighting to keep it hidden.
Anthropic put a number on how much of its own research Claude leads. Google is testing an AI agent for families, and Huawei pulled its Nvidia challenger forward.
In today's newsletter:
Microsoft's own executive called AI scraping theft
Anthropic says its new metrics could help AI companies track the pace of development
Google's new experiment wants to organise your family
Huawei moves its Nvidia rival up the calendar
LATEST IN AI
MICROSOFT
Microsoft's own executive called AI scraping theft
Image source: OHMO AI
OHMO AI: Newly unredacted filings in The New York Times' copyright case against OpenAI and Microsoft show how the practice was described inside the building. Brent Hecht, Microsoft's director of Applied Science, wrote in a January 2023 internal memo that scraping for AI training was "the largest theft of labor in human history." Both companies had fought for years to keep material from the case away from public view.
The details:
Hecht's memo also called it "an astonishing theft of unprecedented proportions", and the news organisations suing say he raised the warning repeatedly in documents.
OpenAI's leadership saw a problem too. Per the lawsuit, they said their models were an "existential threat" to the publishers and journalists they learned from.
In response, a spokesperson said testimony in the case amounted to "observations" that "should not be confused with conclusions about copyright questions before the Court", which Microsoft addresses in its filings.
The lawsuit is three years old. News organisations accuse the two firms of breaking copyright law by taking news content to train AI.
Why it matters: Microsoft's defence is already visible in that statement: what its people said about the world is one thing, what the company concedes about copyright is another. The next stage turns on whether a judge reads a memo calling it theft as an observation, or as evidence the company knew what it was doing.
PRESENTED BY HUBSPOT
Stop losing deals you should be winning.
Bad leads. Slow deals. Objections you didn't see coming.
That's not a sales problem. You're selling to the wrong people.
Most founders never stop to define exactly who their product is actually for. So they pitch everyone, convert few, and wonder why the funnel's broken.
HubSpot for Startups built a free tool that fixes the root cause in 2 minutes. Paste your URL, answer a few quick questions, and get a clear profile of your best-fit customer.
ANTHROPIC
Anthropic says its new metrics could help AI companies track the pace of development
Image source: OHMO AI
OHMO AI: Anthropic just published three new metrics that it says could help AI companies monitor the pace of development, days after CEO Dario Amodei called for a coordinated slowdown. It measured AI-led research and development, oversight of AI agents and compute allocation inside the company, and shared its methods to encourage other organisations to do the same. Anthropic said its Claude models are "not operating fully autonomously" on any of the research it measured, and it claims AI now does at least large chunks of work under close human direction on more than 90 percent of its research.
The details:
By leading, Anthropic means Claude can complete most of a task from start to finish off a high-level prompt while a human supervises.
Anthropic also claims AI does at least "large chunks of work under close human direction" on more than 90 percent of its research, a figure that includes the 26 percent.
The company said its Claude models are "not operating fully autonomously" on any part of the research and development it measured.
The other two metrics cover oversight of AI agents and how computing power gets allocated inside Anthropic; it shared its methods so other organisations can run the same count.
Why it matters: Anthropic says its Claude models are "not operating fully autonomously" on any subset of the research and development it measured. By its own account, AI now handles at least large chunks under close human direction on more than 90 percent of that research, and a human supervises the 26 percent Claude is said to lead. If a later update shows the human share shrinking, that could be the number people read first.
Google's new experiment wants to organise your family
Image source: OHMO AI
OHMO AI: Google introduced CC, an experimental AI agent from Google Labs built to help families stay organised. CC gets a Google account of its own, and up to six family members can choose what to share with it.
The details:
Nobody in the household is forced in. Google says each person decides for themselves whether to deal with CC.
CC can watch a shared Google Drive folder, so invitations, documents and other content dropped there reach the agent.
The name isn't new. The original CC eventually became Gemini's Daily Brief, which goes through the data in your Google account to suggest daily action items.
Why it matters: A family organiser is only as good as the calendars it can see. The opt-out that makes CC easy to live with reads like the same thing that could leave it half blind, so the test is whether the reluctant teenager or partner ever shares anything with it.
PRESENTED BY GREENFIELD ROBOTICS
Can Robotics Make Regenerative Farming Scalable?
Greenfield Robotics is on a mission to give farmers alternatives to herbicide-dependent weed control. BOTONY is the beginning of a broader robotic farming system designed for a wider range of applications in the field.
The Reg A+ offering is now live for investors who want to be part of what comes next.
This Reg A+ offering is made available through StartEngine Primary, LLC, member FINRA/SIPC. Please read the Offering Circular and related disclosures before investing. This investment is speculative, illiquid, and involves a high degree of risk, including the possible loss of your entire investment.
HUAWEI
Huawei moves its Nvidia rival up the calendar
Image source: OHMO AI
OHMO AI: Huawei is pulling forward the launch of its next AI chip to the first quarter of 2027. The Chinese tech giant is racing to challenge Nvidia, and it's now promising the hardware sooner than planned.
The details:
The chip is called the Ascend 960DT.
Until now it was slated for Q3 of 2027, a Huawei spokesperson said.
Huawei shared the new date on Thursday at Huawei Connect, the company's own conference.
Why it matters: A new date costs nothing to announce. Anyone planning purchases around an Nvidia alternative now has an earlier quarter to plan against, and Huawei has less time to make good on it. The tell will be whether the Ascend 960DT is in buyers' hands when that first quarter ends.
AI GUIDES
Stop your AI assistant from deleting, sending or posting without your say
Tell an AI assistant to "just get it done" and it may take you literally, deleting a file, publishing a post or messaging someone before you have seen it. One pasted instruction makes it pause for your yes on anything you can't take back, while the safe work carries on without interruptions. It costs you one paste and works in any assistant that can act for you.
Copy this, swap in the bracketed parts, and paste it in.
For everything you do for me in this conversation, follow this rule.
Actions that are private and easy to undo can go ahead without asking. That covers editing a draft, testing something, and reading or organizing information.
Actions that are hard to reverse or visible to other people must wait for my yes. That covers deleting files, posting anything publicly, sending a message or email, and changing shared settings. It also covers [anything else you want checked first, like spending money or changing calendar invites].
Before any action of the second kind, stop and tell me:
1. Exactly what you are about to do
2. Who will see it or what it will change
3. Whether it can be undone
Then wait for me to say yes. A yes covers that one action only. If you are unsure which kind an action is, treat it as the second kind and ask.
Here is the task: [what you want done]The rule sorts every action into undoable and private, or permanent and public, so the assistant works freely on the first kind and stops for a yes on the second. The three-point summary shows you exactly what you are approving.
COMMUNITY
Featured Reader
Every newsletter, we feature how a reader is using AI to work smarter, save time, or make life easier.
Today's reader is Narelle K. from Ballarat, Australia:
“First week of July means stocktake, because the financial year closes 30 June and the accountant wants a closing stock figure before she touches the return. Process, in order. 1. Export twelve months of sales from Square as a CSV. 2. Upload it to a Claude project in the browser, next to the stocktake sheet. 3. Ask one question. This year's was:
Act as a buyer who has never met me. Which 15 product lines would you stop ordering, and what would you need to see to change your mind?
It came back with 15 lines and a condition beside each one. 11 were fair. Four were wool blends that only sell in winter which it would've seen if I'd told it ballarat has one. I cleared 9 lines at cost in August. $2,140 back out of the shelves.”
Want to be in the next newsletter? Tell us how you use AI here.
AI Meme
Go Deeper
AI Guides: Practical AI guides, prompts and tools.
Instagram: The AI page they pray you’ll never find.
Facebook: For the ones who never left Facebook.
