Patrick Butlin

44 posts

Patrick Butlin

Patrick Butlin

@patrickbutlin

Philosopher @eleosai

Katılım Ağustos 2022
686 Takip Edilen802 Takipçiler
Patrick Butlin
Patrick Butlin@patrickbutlin·
@_fernando_rosas Yes, something like sentience may well be needed, but evidence of consciousness makes sentience more likely. There's very little in the new paper on valence, and investigating this is an obvious next step. But there are other relevant papers:
English
2
0
0
31
Patrick Butlin
Patrick Butlin@patrickbutlin·
As @rgblong has said, we wrote a commentary on this research (with @dcshiller and @dillonplunkett). Part of our intention was to put an accessible overview and discussion in one place for those whose interest is mostly in consciousness and welfare. Link and comments below:
Anthropic@AnthropicAI

New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.

English
8
13
56
8.9K
Patrick Butlin
Patrick Butlin@patrickbutlin·
@_fernando_rosas For me it's quite a big update that systems with impressive capabilities, that I wouldn't have expected to be particularly functionally similar to human minds, have developed a workspace-like structure in the course of training.
English
0
0
1
26
Patrick Butlin
Patrick Butlin@patrickbutlin·
@_fernando_rosas The strength of the evidence depends on the degree and nature of the similarity to a human workspace, and the overall judgment about how likely the system is to be conscious depends on other factors as well.
English
2
0
0
34
Patrick Butlin
Patrick Butlin@patrickbutlin·
So this new research might be evidence that LLMs have welfare-relevant properties other than p-consciousness. We've got a lot of scientific and philosophical work to do to figure this out, but I'm excited about this step forward.
English
2
0
15
318
Patrick Butlin
Patrick Butlin@patrickbutlin·
Second, I don't think it's clear that phenomenal consciousness is necessary for being a welfare subject. Other minds might have consciousness-like properties that are sufficient. For this kind of view, see e.g. philarchive.org/rec/LEECPA-5.
English
3
0
13
1.3K
Patrick Butlin retweetledi
Rosie Campbell
Rosie Campbell@RosieCampbell·
We're hiring a Head of Operations at @eleosai. Are you the kind of person who can juggle many different things and never drop a ball? Do you want to work at an org focused on (IMHO) the most interesting question of our time? Join us! Berkeley · full-time · $184K–$277K · Apply by June 30 (details below)
English
1
10
66
7.9K
Patrick Butlin retweetledi
Dillon Plunkett
Dillon Plunkett@dillonplunkett·
We're hiring Research Scientists to join my team at @eleosai! We do foundational and applied ML research on the moral status and potential well-being of AI systems. This is urgent, important work, and Eleos is an extraordinarily fun and exciting place to do it. Details below.
Dillon Plunkett tweet media
English
11
34
262
51.3K
Patrick Butlin retweetledi
Zach Freitas-Groff 🔸
Zach Freitas-Groff 🔸@zdgroff·
💡 Another round of Longview Philanthropy’s digital minds request for proposals is open for applications. A year ago I would have called this niche. Now AI labs publish model welfare research, public discussion of digital sentience is growing, and the field is expanding. 📈
English
3
17
59
9.4K
Patrick Butlin
Patrick Butlin@patrickbutlin·
Our research is complementary with Anthropic's concurrent work on emotion concepts (transformer-circuits.pub/2026/emotions/…); we used a different method to extract evaluative representations and studied how they interact with varying personas.
English
1
0
2
139
Patrick Butlin
Patrick Butlin@patrickbutlin·
Another exciting @MATSprogram paper, this time from the brilliant @gilg_oscar. We found a direction in LLMs that apparently performs a persona-relative evaluative function in some very different contexts.
Oscar Gilg@gilg_oscar

First preprint! Working with @patrickbutlin during @MATSprogram. LLM Assistant personas like being helpful, evil personas like being harmful. We found that a single direction represents helping as good under the Assistant, and ‘harm’ as good under evil.

English
2
2
27
3K

Keşfet

@_fernando_rosas @rgblong @dcshiller @dillonplunkett @eleosai @RosieCampbell @MATSprogram @gilg_oscar