@TimaeusResearch

Timaeus is an AI Safety Research Organisation working on Singular Learning Theory and Developmental Interpretability.

Joined June 2023
  • Tweets16
  • Following25
  • Followers330
2Media
Timaeus is joining ⊢ Sequent Research!
Geoffrey Irving
@geoffreyirvingJun 10
We are starting a new, nonprofit alignment organization, ⊢ Sequent Research, bringing together researchers previously on UK AISI’s Alignment Team, Timaeus, and elsewhere to research how to align superintelligence. We are hiring! 🧵
1
3
60
4,501
Please reach out if you’re interested in working with us! Sequent will have a large in-person presence in Berkeley, as well as researchers remote from London, Melbourne, and elsewhere. 🇺🇸🇬🇧🇦🇺 1. Full post: sequent.org/launch 2. Express interest: sequent.org/apply

Sequent: Scale and Automation for Higher Confidence in Alignment — Sequent

Sequent founding announcement.

sequent.org
5
3
83
4,737
I’m incredibly proud of what we’ve built with Timaeus and all the research we’ve accomplished. But given how much research will change over the next few years, it’s time to make another bet. 9/9
1
1
23
534
This is why a strong human component is still necessary in alignment research at Sequent. Maybe that’s you. You should consider dropping what you’re doing and helping. A lot of other theoretical and empirical research can be left to the ASI, but alignment can’t (responsibly).
1
1
1
28
2,593
Timaeus retweeted
george
@georgeyw_Jan 27
“Neural networks are grown, not programmed” We’re changing that. Mechinterp investigates how models generalize beyond their training data by studying the resulting internal structure. We introduce patterning as the dual: given desired structure, determine what data produces it.
19
155
10
1,070
92,295
Well, we haven't solved all the problems yet, so the offer stands... If you want to solve deep scientific problems in AI safety, come work with me through the Pivotal Fellowship 1/4
Jesse Hoogland
@jesse_hoogland1 Oct 2025
Calling all mathematicians, physicists, computer scientists, statisticians, learning theorists, engineers, neuroscientists, developmental biologists, etc.... If you want to solve deep scientific problems in AI safety, apply to our MATS stream 1/3
1
4
18
1,910
Training Data Attribution (TDA) should account for learning dynamics! The same data can influence model behavior in dramatically different ways at different time points of training. We call for a shift towards stagewise data attribution and the study of influence dynamics. 1/11
1
11
2
46
4,534
How does training data shape model behavior? Well, it’s complicated… 1/10
15
138
5
961
97,071
We're hiring for research engineers! Apply by *this Sunday* (Sep 14th)!
1
4
1
19
7,085
Using singular learning theory, we study how data shapes the structures that neural networks learn during training, and how those structures enable generalization. This gives us a unique lens on interpretability and alignment. xitter.cf/danielmurfet/status/19…
Daniel Murfet
@danielmurfet5 Aug 2025
Neural networks are grown, not programmed. What does that growth process look like? Like this! This is a small language model (3M) across training, visualised with a new interpretability technique: susceptibilities. We call this handsome critter the rainbow serpent.
1
2
427
If you're excited to support our researchers and contribute to this agenda, apply today! airtable.com/appMGBqrZQpiKXR… More details: timaeus.co/blog/updates/2025…
3
246
1/ AI is accelerating. But can we ensure that AIs truly share our values and follow our goals? We argue that aligning advanced AI systems requires cracking a core scientific challenge: how data shapes AI's internal structure, and how that structure determines behavior.
30
90
13
539
105,403
Timaeus retweeted
george
@georgeyw_11 Feb 2025
Our paper has been accepted to ICLR as a spotlight paper! We introduced the refined Local Learning Coefficient, which measures “how much structure” there is in particular parts of the model associated to particular datasets or behaviors.
george
@georgeyw_16 Oct 2024
1/ How do attention heads form? With our new approach, we show that attention heads have distinct developmental signatures. These signatures reveal how heads develop distinct functional roles specialized to different subsets of data. In the process, we discover a new circuit.
4
15
92
8,433
We’re hiring! We’re looking for research scientists, engineers, and a research lead to work on AI alignment through singular learning theory
1
9
5
35
9,503
More details: $70-140k, initially remote with possible relocation to Berkeley/London. Rolling applications until Feb 13th. Announcement: lesswrong.com/posts/g8e4pz7a… Apply directly: airtable.com/appMGBqrZQpiKXR…

Timaeus is hiring researchers & engineers — LessWrong

TLDR: We're hiring for research & engineering roles across different levels of seniority. Hires will work on applications of singular learning theory…

lesswrong.com
1
3
645
1/8 How do transformers learn? In our new work, we find that transformers develop in-context learning in discrete stages that can be automatically discovered. 🧵 arxiv.org/abs/2402.02364 Joint work w/ @georgeyw_, Matthew Farrugia-Roberts, @lemmykc, Susan Wei, @danielmurfet
2
85
11
424
37,090
Load more