Associate Dean for Research & Chair of Information Management at Bayes Business School (formerly Cass), City St George’s, University of London
London, England
Joined August 2011
- Tweets4.2K
- Following1.3K
- Followers619
- Likes7.9K
Feng Li retweeted
Human-in-the-loop (HITL) is widely recommended for oversight & safety, yet current AI agent design requires humans to keep up with unreasonable volume & complexity, promotes fatigue & disengagement, and degrades the very humans skills it relies on. 1/ arxiv.org/abs/2608.23642
How many people already let AI agents filter their inbox so they rarely look at it themselves? What does this mean for newsletters & direct relationships with readers, indeed relations between people? Who decides what deserves attention, including ideas we didn’t know we needed?
a big take I've been noodling on and I’m ready to call:
the newsletter is dead
@every's core business is a newsletter with 120k+ subscribers so I don't say this lightly.
persistent agents mean i dont look at my inbox
with dots (GrokBots, Muse, etc) I almost never look at my inbox. if I do it's for a specific thing I need to reply to.
i think in a year I'll be doing all of my email this way - fully filtered by my dot / something like it.
this is not new: it's a continuation of the trend of having a “newsletter” folder in Gmail, but 10x more filtered.
what it means
It doesn’t mean people won't read newsletters, but it does mean that getting a person’s email is no longer the guarantee that you have a relationship with them that it used to.
luckily, with persistent agents (and AI coding more generally) there is an entirely new playing field to build relationships with readers. instead of fighting to get an email address, publishers and creators will need to make sure we're installed in our reader's agent harness of choice, so that they get directly notified when we do something new.
we're already starting to experiment with this @every with our MCP and the frontier alerts we send users of our slack agent (every.to/agent)
it's also more affordable than ever for publishers and creators to build their own platform ecosystem: you can have your own website, app etc. and you can use AI to publish 100x more high-quality stuff to your community—so we may see a world where many creators themselves are running their own little empires that have their own apps, feeds etc.
to be clear: It'll take a while before this actually happens, but the idea of getting someone's email creating a direct relationship with them is over.
here's a set of predictions I made 3 years ago about AI and media that panned out pretty well: every.to/chain-of-thought/we…
“It’s about strategy and management as much as subsidies. The likes of BYD certainly benefited from past state support, but their operations were refined in a ferociously competitive domestic Chinese EV market. Meanwhile, European car manufacturers were…” giftarticle.ft.com/giftartic…
“Schools and universities should draw a similar distinction: abundant AI resources and support during study and strict independence in assessment.”
How I used AI to become a chess grandmaster at 16 via @FT
giftarticle.ft.com/giftartic…
Hmmm i wish i could be this positive - “The [global] economy has been resilient, but that resilience is reaching its limits,” Tenreyro warned. ft.com/content/4a7a7282-74a5…
More R&D? Better social care? A stronger NHS? Thriving regions? All worthwhile. But money is finite. What we choose not to fund matters as much as what we fund. Leadership and strategy require choices.
My BAM piece:
britishacademyofmanagement.s…
@bam_ac_uk @BayesBSchool @CityStGeorges
Feng Li retweeted
Intelligence must draw on memory. The mind must not simply compile data — as an algorithm now does more quickly than we can — but recall lived experiences, which contain depths of meaning that only the human soul can recognize, and connect more deeply with the goodness of these experiences. Everything that is good begins “with the Spirit” and thus possesses a breath, a dynamism and a sense of communion that, once experienced, are engraved upon the heart. vatican.va/content/leo-xiv/e…
Feng Li retweeted
"I believe there’s about a 50% chance we all die because of the development of smarter-than-human AI systems," writes Geoffrey Irving.
time-magazine.visitlink.me/c…
So now we use AI to rewrite what we write in our own words to sounds not like AI, just to evade detection by AI detectors? What the hell has the world come to?
And is this tweet written by AI?
wrote it yourself. flagged as ai anyway.
walter checks it first, and helps it read unmistakably like you. keep it yours.
trywalterai.com/ai-humanizer…
Feng Li retweeted
This could be a problem: Goldman sensitivity table shows that in a worst case scenario where hyperscaler ROIC on capex is 0 (token costs collapse, token demand goes to open models, etc) they still need to spend $920BN every year just to cover depreciation and running costs
Goldman Calculates How Much Revenue Is Needed To Justify $1.7 Trillion In Hyperscaler Capex (And What It Leaves Out) zerohedge.com/markets/goldma…
where do toy stand on such issues? - German unions are fighting to defend the 35-hour week economist.com/business/2026/…
Today is @BayesBSchool AnnualResearch Day. We are launching four new white papers around our Strategic Research Themes, setting out key research agendas and the next steps for collaboration, funding, publications and impact. Get involved! @CityStGeorges
bayes.citystgeorges.ac.uk/fa…
Shared a PowerPoint through Outlook on my iPhone. Microsoft silently turned its own .pptx file into .odp, which the iPhone then refused to open. In 2026, sharing a PowerPoint file through Outlook on iPhone apparently remains an unsolved technical challenge. How Evil!
New paper out in Information Technology & People, based on Jafar Sabbah’s PhD research with Aneesh Banerjee and me. GenAI does not automatically improve ideation. There are trade-offs between novelty, utility and feasibility. @NewsfromCity @BayesBSchool
emerald.com/itp/article-abst…
Feng Li retweeted
Oxford researchers just published a paper arguing LLMs cannot invent anything. Mathematically impossible.
The reason is simple and brutal. A model trained to predict the next word can never believe something the existing data says is wrong. And every real breakthrough in history started with exactly that belief.
In 1903 every prediction machine on Earth would have told the Wright Brothers that human flight was one to ten million years away. Nine weeks later they flew. Not because they had better data. Because they had a theory the data hadn't caught up to yet.
That is the gap between AI and human thinking. LLMs mirror the past. Humans reason into a future that doesn't exist yet. Oxford just proved mathematically that those are two different things.
I use these models every day. This matches what I see. The new stuff always comes from the human at the keyboard who decides the data is wrong.
LLMs don't think. You do.
Readers added context they thought people might want to know
The paper was not "just published", it was published on December 2nd, 2024 - 21 months and 16 days before the post.
pubsonline.informs.org/doi/10.1287/st…
“these companies alone could be worth well north of $5tn…Now compare that with the entire history of IPOs from 1980 to 2025. The 3,365 tech companies that went public in that period were worth a combined $4.1tn when they started trading”
ft.com/content/8c4143c7-f846…
“The labs are saying to governments: ‘We’re quite likely to kill every human being on Earth; please stop us.’ Governments respond with ‘Can we give you a tax break? Build you a datacenter?’ Perhaps at some point will remember voters prefer not to be dead”
ft.com/content/b7fe0fe0-0463…
Europe’s potential zone of AI sovereignty is narrow. Its frontier labs cannot at present compete financially with their American and Chinese competitors, and leading-edge chip production is far behind. ft.com/content/f054f927-b512…
