AI that works. News that's ranked. No filler. https://teapawt.net/t.co/VrgZrPJPUZ
UK
Joined July 2009
- Tweets35K
- Following2.9K
- Followers14.9K
- Likes2.5K
Before your next one-to-ones, write a short paragraph for each role saying what good AI use looks like in that job. Then ask each person what they need to get there.
readanswerable.com/p/would-y…
Prompt that gets a draft: "Reply to this customer."
Prompt that gets a reply you can send: "Reply to this customer. They want a refund date. We can't give one before Friday. Don't apologise twice."
Washington is withdrawing its 1978 guide to fair hiring tests, the one the four-fifths rule comes from. The law underneath doesn't change.
Tempting: stop asking your hiring tool for a bias report. Worse: tidy away old hiring data. One employment law firm warns that can look like destroying evidence.
readanswerable.com/p/ai-regu…
When researchers asked models including GPT-4 how confident they were, the answers mostly sat between 80% and 100%, usually in multiples of five, right or wrong.
Ask where a claim came from instead. That you can check.
aiprompthackers.com/p/ai-mix…
Anthropic's Claude Duet is pitched at small business owners. It sends a morning brief, checks your connected apps every few hours and drafts on its own. Before it sends, pays, signs or deletes, it shows you exactly what it's about to do and waits. Moving money between accounts is off-limits even with approval.
Sensible design. The approval step is only as good as the person tapping Approve at 7am.
In Anthropic's trial with 52 mostly junior engineers, those who handed the whole task to AI finished fastest and averaged under 40% on a quiz about the code they'd just produced.
Those who only asked it conceptual questions scored 65% or more.
Guess which group tops the usage dashboard.
readanswerable.com/p/your-he…
Yesterday's question: the agency using an AI image tool for EU clients, and the 2 December marking date.
Answer: C. That date applies to suppliers of generative AI systems put on the EU market before 2 August 2026. No duty on the buyer for that one.
Your job is one email: ask which release adds the marks.
Twelve times revenue growth and a $23bn quarterly loss in the same breath. Welcome to AI accounting, where both numbers are used to justify the other. Lets hope the Nasdaq feels better by Thanksgiving...
AI stock liquidation raises fresh doubts about Anthropic's planned mega-IPO
As the Nasdaq 100 fell as much as 6%, reports said Anthropic's planned listing before Thanksgiving, potentially raising about $100bn at a $1.5tn-plus valuation, looks harder. Its 2 November amended filing showed about $17bn of Q3 revenue, a $23.3bn net loss and roughly $895.4bn of compute purchase and lease commitments, with nearly $70bn due within 12 months of 30 September.
investing.com/news/stock-mar…
Democracy, briefly awake. Don't make eye contact or it'll doze off again....
Democrats pledge AI hearings and "reasonable guardrails" after taking Congress
House Democratic Caucus Vice Chair Ted Lieu said the incoming majority will hold AI hearings and bring in guardrails, accusing Republicans of letting the industry "run wild". Common Sense Media pointed to an NBC exit poll in which over half of voters said AI and data centres were a crucial or major factor, and pushed early action on the bipartisan GUARD Act for AI companions and the AI LEAD Act on liability.
abcnews4.com/news/nation-wor…
NDAs with councillors are out, except for "proprietary or competitively sensitive information". Which is everything interesting. The secrecy died. Long live the exception clause...
OpenAI vows to drop NDAs with officials and pay full power costs at its own compute campuses
After midterm votes against data centres, OpenAI published a National Community Compact for future campuses it owns and operates. It will pay the full cost of electricity and grid infrastructure, publish construction plans and jobs, report annually on water and power use, and stop using NDAs with public officials except for narrow security and trade-secret carve-outs. It also published a policy blueprint asking lawmakers to turn those norms into rules.
openai.com/index/setting-new…
An Oakland Italian restaurant was losing bookings every time the phone rang during service.
Its AI answered more than 2,300 calls and fully resolved 90%.
One more free case in the Use Case Archive, alongside today's finance and marketing additions.
Open the archive: usecasearchive.com
A two-person Toronto creative agency trained AI on its own past work and briefs.
Concepting time: six weeks → six days or less.
That's the kind of before-and-after the Use Case Archive is for. No fluff, just what small firms are actually doing.
See today's new entries: usecasearchive.com
In a firm of one, every AI mistake gets caught by the same person who made the request. Usually at the same level of tiredness.
If AI now does in minutes what used to take an afternoon, what exactly is the client paying for?
This Answerable piece puts the IRS and the AICPA side by side on that question.
readanswerable.com/p/ai-savi…
Owner describes the bathroom job in his own words → Handoff builds the estimate from his saved presets → it prices the materials and adds the screws, tape and framing nails he used to round off or leave out.
Right Price Remodeling, Massachusetts. About an hour per estimate down to 10 to 15 minutes, according to a press piece.
Missing link in the write-up: who checks the price before it goes to the customer?
The most expensive phrase in a saved prompt: "same format as last time".
The AI guesses, and you fix the guess. Every week.
aiprompthackers.com/p/you-ke…
