Jesse Hoogland ✓
@jesse_hoogland
Cofounder @resolution_org
2 303Followers
1 767Following
461Posts total
439.7KViews on collected posts
Latest posts
Looking for new posts…
We’re changing our name: Sequent is now Resolution.
🧵 https://t.co/XttKlId6OW
20.2K views · 179 likes
· 18 reposts · 5 replies
Open on X →
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 https://t.co/jhTkdftJFL
524 views · 24 likes
· 1 reposts · 1 replies
Open on X →
Sequent will house a portfolio of different theory-driven AI safety bets. Together, we’ll share the engineering and security needed for frontier-model access, and through close contact with each other, we expect these different bets to become more productive. 8/9
500 views · 21 likes
· 0 reposts · 1 replies
Open on X →
Frontier models have now reached the point that they can significantly contribute to mathematical research. This gives us the opportunity to buy time — at the cost of vast quantities of tokens. As a larger org, Sequent will be able to pay that price. 6/9
293 views · 20 likes
· 0 reposts · 1 replies
Open on X →
But even breakthrough scientific progress means nothing if it does not shift how (and whether) frontier AI is trained. This is an incredible ask that pits us against the largest engineering projects in human history. Fortunately, we’ll have help. 7/9
289 views · 16 likes
· 0 reposts · 1 replies
Open on X →
This and other progress has made me confident that the research is on track to “work” granted enough time. I see a clear path forward from these techniques to aims as ambitious as solving inner alignment. Unfortunately, we may not have enough time, and our research on its own was
301 views · 22 likes
· 0 reposts · 1 replies
Open on X →
And we’re now starting to see success in applying these techniques to steer the learning process, through what we call “patterning”. Soon we’ll share our first results from applying these techniques to solving real problems in LLMs. So in short: SLT works. 4/9
431 views · 20 likes
· 0 reposts · 1 replies
Open on X →
We developed a new interpretability stack (“spectroscopy”) and scaled it to models with billions of parameters. These techniques discover rich internal structure that is competitive with SAEs but based on a different foundation (weights rather than activations). 3/9
867 views · 26 likes
· 0 reposts · 1 replies
Open on X →
Timaeus is joining forces with @geoffreyirving and researchers from UK AISI to found Sequent Research. 1/9
15K views · 112 likes
· 19 reposts · 1 replies
Open on X →
Three years ago, we started @TimaeusResearch with the aim of making breakthrough scientific progress on AI safety. We made a bet that a then-obscure branch of learning theory would open up new directions in interpretability and alignment. That bet succeeded. 2/9
1.2K views · 26 likes
· 0 reposts · 1 replies
Open on X →
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: https://t.co/u0LVaPsRzE
2. Express interest: https://t.co/S02xraCtPb
4.5K views · 83 likes
· 3 reposts · 5 replies
Open on X →
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! 🧵 https://t.co/UziUGbIPdU
269.1K views · 1K likes
· 146 reposts · 28 replies
Open on X →
Soon, most thoughts on Earth will be carried by tokens. Many beautiful; some consequential.
Understanding this rising sea of intelligence is a major scientific problem, and is the aim of interpretability. Our new interp results on susceptibilities for Pythia-1.4B: 🧵 https://t.co/
9K views · 86 likes
· 13 reposts · 2 replies
Open on X →
“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.
92.2K views · 1.1K likes
· 155 reposts · 19 replies
Open on X →
Timaeus is a new research organization, dedicated to making fundamental breakthroughs in technical AI alignment using deep ideas from mathematics and the sciences. Led by @jesse_hoogland @FellowHominid Stan van Wingerden and myself. https://t.co/DpgsiUNJyY [1/n] https://t.co/j5y0
25K views · 112 likes
· 19 reposts · 1 replies
Open on X →
Growth & engagement
How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.
Views per post
9K24 Apr
269.1K10 Jun
4.5K
1.2K
15K
867
431
301
289
293
500
524
20.2K30 Jun
169
Last 14 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.
Engagement rate per post
1.16%24 Apr
0.47%10 Jun
2.01%
2.26%
0.91%
3.11%
4.87%
7.64%
5.88%
7.17%
4.40%
4.96%
1.01%30 Jun
0.00%
Reactions — likes, reposts, replies and quotes — divided by views.
What the audience does
Likes73.5%2 851 in total
Reposts9.6%374 in total
Replies1.8%69 in total
Quotes1.9%73 in total
Bookmarks13.2%512 in total
Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.
The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.
Similar accounts
Estelle Birdy
@BirdyBooky
4.6K followers
· 10.8K posts
Александар Стефановић
@acastef53
4.6K followers
· 32.2K posts
Maikel
@Maikel_MRP
4.6K followers
· 22.6K posts
MIKY #finoallafine®
@Mikfinoallafine
4.6K followers
· 51.4K posts
Theeviixiin
@Theeviixiin
4.6K followers
· 174 posts
NDIKURIYO Guy Raoul ✓
@NdikuriyoRaoul
4.6K followers
· 2.5K posts
DAKO
@dako_tsi
4.6K followers
· 10.9K posts
👾| Ka1ju_Freak_Art |👾
@Kaiju_Freak_v2
4.6K followers
· 10.5K posts