Moritz Weckbecker

PhD in Explainable AI @FraunhoferHHI
Master @ Oxford, Cambridge | Bachelor @ TU Berlin
AI researcher, mathematician and statistician

[KRIS @RREAPERMAN]

being a doctor who fan is a humiliation ritual

[Moritz Weckbecker @MWeckbecker]

@LiZhaoyi13 @COLM_conf Same experience, unfortunately. A shame because I had heard good things about COLM

[Moritz Weckbecker @MWeckbecker]

I‘ve unfortunately had the same experience (after initial fair reviews): one reviewer replied hours before the deadline, the other reviewers + AC didn‘t reply at all. Also still waiting on a final decision.
I would appreciate some more communication from @COLM_conf #COLM2026

[Zhaoyi LI @LiZhaoyi13]

I spent a whole week staying up late and making my rebuttal but don't why but all of the reviewers and AC don't reply to even one of my responses 🤡😫😭 @COLM_conf #COLM2026

[Moritz Weckbecker @MWeckbecker]

@jrosseruk @NeelNanda5 @GoogleDeepMind @cohere @acyr_l Congratulations!

[Moritz Weckbecker @MWeckbecker]

@GlennMatlin Unfortunately didn‘t have the greatest rebuttal experience this cycle, but let‘s see what happens. Fingers crossed for you! 🤞🏻

[kasra @kasratweets]

this is an interesting point in the new ted chiang piece – no one really claims that alphafold is conscious, or that sora or midjourney or dall-e are conscious

[Kane Emerson @KaneEmerson]

The UK has successfully de-coupled carbon emissions and GDP.

[Moritz Weckbecker @MWeckbecker]

i just remembered that the pope studied maths and it‘s so surreal

[Moritz Weckbecker @MWeckbecker]

Wake up babe, new encyclica just dropped

[Pope Leo XIV @Pontifex]

Humanity, created by God in all its grandeur, is today facing a pivotal choice: either to construct a new Tower of Babel or to build the city in which God and humanity dwell together. In Jesus Christ, this humanity in its grandeur becomes the Way, the Truth and the Life, opening the path for each of us to grow toward fullness. #MagnificaHumanitas
https://t.co/6i9MWs6LJl

[Moritz Weckbecker @MWeckbecker]

Love AOE, when I can work on my paper for 14 additional hours

Hate AOE, when I need to wait on notification for 14 additional hours

[Moritz Weckbecker @MWeckbecker]

@jrosseruk if only we could actually do it

[Moritz Weckbecker @MWeckbecker]

@AlexBarry4 @EpochAIResearch Congratulations! Excited to see the outcomes of your research!

[Moritz Weckbecker @MWeckbecker]

Very happy to have been named a Gold Reviewer by ICML! It‘s nice to get some positive feedback for the work of reviewing in a world of AI and human slop reviews.

Now wish me luck for the workshop submissions!

[Moritz Weckbecker @MWeckbecker]

ok, I know that sounds insane but I had the idea for this in my 2nd month of my PhD and never followed up on it.

Ideas are cheap. Your ideas deserve at least a 2-week implementation sprint.

[Anthropic @AnthropicAI]

New Anthropic research: Natural Language Autoencoders.

Models like Claude talk in words but think in numbers. The numbers—called activations—encode Claude’s thoughts, but not in a language we can read.

Here, we train Claude to translate its activations into human-readable text.

[Kevin Roose @kevinroose]

New column: I went to visit @METR_Evals, the 30-person AI nonprofit that makes the Most Important Chart in the World.

I learned a lot, but the most striking thing was how soon some of them think AI R&D could be fully automated. (This year!)

https://t.co/EWnYZ7WG0p

[Moritz Weckbecker @MWeckbecker]

@Marshwiggle119 I would expect that it is connected to such effects, although I doubt there exist specific neurons / directions in activation space for each of these subliminal tokens. What things would you test?

[Moritz Weckbecker @MWeckbecker]

1/ We found a new way to misalign an entire AI agent network by compromising just one agent. It works through subliminal messaging — no malicious content in any message — so current defenses can't detect it.

We call it Thought Virus. 🧵

[Owain Evans @OwainEvans_UK]

Our paper on Subliminal Learning was just published in Nature!

Last July we released our preprint. It showed that LLMs can transmit traits (e.g. liking owls) through data that is unrelated to that trait (numbers that appear meaningless).

What’s new?🧵

[Owain Evans @OwainEvans_UK]

Such numbers look innocent and they can be spread between groups of agents, surreptitiously spreading traits. This also builds on subliminal prompting by Zur et al. (2025)
https://t.co/3RmVojMJuH

[Owain Evans @OwainEvans_UK]

Weckbecker et al (2026) created a kind of subliminal virus that spreads between groups of LLM agents. They find particular numbers that cause models to exhibit certain traits. E.g. When the number is in the context window, the model expresses a preference for owls.