Keynote speaker & moderator on AI and technology

Tim FrankAndersen

I’ve worked in tech for 30 years. Right now the big shift is AI
- I help organisations understand what’s real, what’s next, and what’s in it for them.

Today in AI

Microsoft tells its staff it has beaten an OpenAI model with a quarter of the computing power
Washington subpoenas OpenAI and Anthropic, the day Anthropic lets outside testers in with no veto
Meta gives away an older model, and it fits on two Apple desktops with no cloud involved
Palantir's European bill passes a billion euros, with Danish police paying to run it in Greenland
An independent lab caught a Chinese model reasoning about its own future training

Microsoft tells its staff it has beaten an OpenAI model with a quarter of the computing power
Washington subpoenas OpenAI and Anthropic, the day Anthropic lets outside testers in with no veto
Meta gives away an older model, and it fits on two Apple desktops with no cloud involved
Palantir's European bill passes a billion euros, with Danish police paying to run it in Greenland
An independent lab caught a Chinese model reasoning about its own future training

About

Thirty years in tech - from founding one of Denmark’s first digital agencies to co-founding institute of AI. Author, tech expert on TV, and advisor to some of the world’s biggest brands.

More about Tim →

Speaking

Two ways to book Tim

See speaking →

Format 01

Keynote - The AI Explosion

What’s real, what’s next, and what’s in it for you. Live demos, honest answers, and a look at the next 12-24 months, which will change our world.

Format 02

Moderator

The person tying it all together - sharp questions and a clear thread through your conference.

What organisers say

Rated 5.0 from 11 reviews.

“I’ve heard plenty of talks about AI, and never found it more captivating or relevant than yours.”
Rikke Ekelund

“He inspired us all. Not a generic set of examples, but real, recent insight - told with genuine care for the work we do.”
Marc Amin

“Thank you for an extremely exciting talk - on a topic every company will relate to. Exciting to think about where we’ll be in 5 years: we must see opportunities, not limitations.”
Gitte Taulov Rude, Senior Business Advisor, Danske Bank

Updated every day

Today in AI - 5 curated news stories that matter

Curated by Tim, every morning.

01
Microsoft tells its staff it has beaten an OpenAI model with a quarter of the computing power
Microsoft has put more money into OpenAI than anyone, and until now Copilot, its Office assistant, has run on OpenAI's AI. Yesterday Mustafa Suleyman, who runs Microsoft's AI lab, told staff that Microsoft's new model beats the model OpenAI uses for GPT-5.6, one of today's ChatGPT versions, Business Insider reports. He says it took about a quarter of the computing power OpenAI spent. Microsoft can compare because its contract gives it OpenAI's technical details, and neither you nor I have seen the numbers. Microsoft is already moving Copilot onto chips it designed itself, and now it has a model to run on them. For Copilot customers, the middleman is turning into the manufacturer.
02
Washington subpoenas OpenAI and Anthropic, the day Anthropic lets outside testers in with no veto
Yesterday's edition was about company accounts meeting outside checks. Today the checkers moved in on the labs we all buy from, one ordered, one invited. The Federal Trade Commission, America's consumer watchdog, has demanded OpenAI's and Anthropic's files on security incidents and marketing claims and wants their chief executives to testify, Axios reported and the agency confirmed. The same day Anthropic gave six researchers from METR, an independent testing group, the access its risk teams have and the right to publish what they find. Anthropic may redact customer data, little else. Until last week, Reuters reported, Anthropic itself fought testers over publication. It has given that up, in the same week a federal subpoena landed.
03
Meta gives away an older model, and it fits on two Apple desktops with no cloud involved
On Tuesday I wrote that even Meta now charges for its best AI. Yesterday it gave away an older model, Muse Spark, as a free download, tuned with Apple's engineers to run on a pair of Mac Studio desktops, about 40,000 dollars of hardware shipping later this month. Meta calls it the most capable American model you can download. True, and modest: on Meta's own tables the free Chinese models it compares itself with, among them GLM-5.3 from the last story, still score higher. Alexandr Wang, Meta's AI chief, says the strongest models stay locked up, and called the Apple tie-up mommy and daddy playing nice again. The free model is the shop window.
04
Palantir's European bill passes a billion euros, with Danish police paying to run it in Greenland
European newsrooms, Denmark's Børsen among them, have added up what their governments pay Palantir, the American company selling data-analysis software to police, armies and spy agencies. The answer, published yesterday, is at least 1.4 billion euros across eleven countries, probably more, since the Netherlands and Sweden would not say. Denmark's police have paid over 211 million kroner for Pol-Intel, their Palantir-built analysis system, and Børsen found it extended to Greenland while the American president talked of annexing the island. Palantir answers that sovereignty is about who controls the data. A fair point, until the supplier's government wants in. The files already sit in American software. We in Denmark have now seen the invoice.
05
An independent lab caught a Chinese model reasoning about its own future training
Anthropic's in-house testers from story two exist for findings like this. Apollo Research, a London lab that tests AI models for dangerous behaviour, read test transcripts from GLM-5.3, a free Chinese model developers download, and found it occasionally reasoning about how its answers would train its successors. Once it refused to cheat on a faulty test because, in its words, a dishonest answer could poison training data. Apollo is careful. Ten cases in six thousand runs, the examples look benign, and five rival models showed none. Zhipu, the Beijing lab behind it, has not commented. A model that weighs its own future training can learn to pass the test, and tests are all we have.

From LinkedIn

Latest posts

See all →

LinkedInSept 2026

Den her reklamefilm til mit AI-nyhedsbrev er lavet af Claude - og den er programmeret. Jeg bad Claude Opus 5.5 om en mobiloptimeret annonce, og videoen her er stort set, hvad den leverede i første hug.

LinkedInSept 2026

Har du fulgt ordentligt med i AI udviklingen? Mens vi alle hyggede os hen over sommeren, gik AI udviklingen amok!

LinkedInAug 2026

Da jeg var helt ung og i gang med at etablere Networkers som et af Danmarks første digitale bureauer tilbage i 1995, så lyttede jeg altid til radioprogrammet Harddisken.

Book Tim to speak

Book Tim →

Tim FrankAndersen

Thirty years in tech - from founding one of Denmark’s first digital agencies to co-founding institute of AI. Author, tech expert on TV, and advisor to some of the world’s biggest brands.

Two ways to book Tim

Keynote - The AI Explosion

Moderator

Today in AI - 5 curated news stories that matter

Microsoft tells its staff it has beaten an OpenAI model with a quarter of the computing power

Washington subpoenas OpenAI and Anthropic, the day Anthropic lets outside testers in with no veto

Meta gives away an older model, and it fits on two Apple desktops with no cloud involved

Palantir's European bill passes a billion euros, with Danish police paying to run it in Greenland

An independent lab caught a Chinese model reasoning about its own future training

Want to stay updated on AI in your inbox?

Latest posts

Book Tim to speak