Stephanie Chan ✓

@scychan_brains

Staff Research Scientist at DeepMind. Artificial & biological brains 🤖 🧠 Societal impacts of AI + Science of AI. Views are my own.

7 498Followers

2 862Following

988Posts total

144.3KViews on collected posts

Posts recentes

Stephanie Chan ✓@scychan_brains2 Nov 2024

@scychan_brains @JacquesPesnot @summerfieldlab No. Humanising AI in research should be pushed back. Silicon Valley blurred fiction and reality, and now, like piecing together a hangover, we’re just starting to connect the dots of bad decisions. https://t.co/48G13IAuTr 97 views · 0 likes · 0 reposts · 0 replies Open on X →

Stephanie Chan ✓@scychan_brains1 Nov 2024

Also wanted to point to a few more results showing correspondence between human and LM cognition, though the direction of inspiration was psychology -> AI instead. (Of course, this direction will continue to be fruitful too!) *https://t.co/3AkevS32l8 *https://t.co/RZQSU1xt5P N/N 1.3K views · 16 likes · 1 reposts · 0 replies Open on X →

Stephanie Chan ✓@scychan_brains1 Nov 2024

TLDR: if "science of AI" wasn't already attractive to you (above-mentioned benefits, interpretability benefits, possibility of identifying areas for improvement), here's another reason 😊 5/ 1.5K views · 7 likes · 0 reposts · 1 replies Open on X →

Stephanie Chan ✓@scychan_brains1 Nov 2024

Perhaps this transfer paradigm will be increasingly fruitful going forward, as AI models gain more complex capabilities, with some correspondences to human intelligence. 4/ 1K views · 8 likes · 0 reposts · 1 replies Open on X →

Stephanie Chan ✓@scychan_brains1 Nov 2024

In "science of AI", we often have faster iteration, cleaner control over experiments, and divergent sources of inspiration. Psychology / neuroscience can benefit from these as well, by sourcing AI results for new experimental hypotheses and paradigms. 3/ 1.3K views · 11 likes · 0 reposts · 1 replies Open on X →

Stephanie Chan ✓@scychan_brains1 Nov 2024

But I'm also excited about being part of this transfer from "science of AI" -> science of human intelligence. This process is not new, but it might soon have a moment of flourishing. 2/ 1.6K views · 10 likes · 1 reposts · 1 replies Open on X →

Stephanie Chan ✓@scychan_brains1 Nov 2024

I'm very excited about these results (kudos to the authors). E.g. it's wild that the transition between in-weights and in-context learning also occurred at a power-law skew of alpha=1 (same as transformers, and also the same skew as human language)! @FelixHill84 1/ 2.6K views · 9 likes · 0 reposts · 1 replies Open on X →

Stephanie Chan ✓@scychan_brains1 Nov 2024

"Do humans learn like transformers?" This is so cool. We previously showed that transformer in-context learning depends on certain distributional properties of the data. Now, @JacquesPesnot and @summerfieldlab have shown similar effects in humans! Some thoughts 🧵👇 90.6K views · 215 likes · 23 reposts · 3 replies Open on X →

Stephanie Chan ✓@scychan_brains30 Oct 2024

@JacquesPesnot and I are very happy to share this paper. https://t.co/VNJpegYhWn. Building on work by @scychan_brains we show that human in-context and in-weight learning vary with the distributional properties of the training data in a very similar way to transformer networks. 44.2K views · 141 likes · 37 reposts · 3 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

44.2K30 Oct

90.6K1 Nov

2.6K

1.6K

1.3K

1.5K

1.3K

972 Nov

Last 9 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.41%30 Oct

0.27%1 Nov

0.39%

0.77%

0.91%

0.89%

0.52%

1.30%

0.00%2 Nov

Reactions — likes, reposts, replies and quotes — divided by views.

What the audience does

Likes63.4%417 in total

Reposts9.4%62 in total

Replies1.7%11 in total

Bookmarks25.5%168 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.

Posts recentes

Growth & engagement

Views per post

Engagement rate per post

What the audience does

Contas semelhantes