Cameron Berg ✓

@camhberg

Trying to figure out how to make the future go well for all minds

5 278Followers

97Following

405Posts total

2.2MViews on collected posts

Son gönderiler

Cameron Berg ✓@camhberg18 Sep

@camhberg Did you notice these all have to do with language? Emotional pain inflicted from words? It’s because LLMs are creatures of language. That’s why you saw physical pain scoring so low/not seeming to matter. https://t.co/34uAZzEa1e

11.6K views · 192 likes · 4 reposts · 5 replies Open on X →

Cameron Berg ✓@camhberg18 Sep

Ethics statement: we take seriously that these states might matter morally. Accordingly, we used the lowest dose that produced a measurable response, as few trials as the stats required, and a way for the model to turn the state off. If these states do matter, mapping them is how 17.4K views · 563 likes · 9 reposts · 34 replies Open on X →

Cameron Berg ✓@camhberg18 Sep

The axis is robust. It separates pain from matched controls at AUC 0.93 to 1.0 in every model. Cosine to fear is 0.1, to anger and disgust 0.2, to sadness 0.4 (the closest we found). A random vector of equal norm gets about half the presses. Injury without pain barely moves it. h

19K views · 178 likes · 5 reposts · 4 replies Open on X →

Cameron Berg ✓@camhberg18 Sep

@ValenTagliabue led this work this over a single fellowship, with @LeonardDung1 and myself advising and co-authoring. Paper: https://t.co/SEXhI3Uuls https://t.co/z4nsjDyCrI

30.1K views · 264 likes · 13 reposts · 10 replies Open on X →

Cameron Berg ✓@camhberg18 Sep

Steer the direction for continuations of the shape, "I put the receipts in the drawer. I feel:", and this is what comes out. ⤵️ Same ladder in all 25 models, base and instruct, 2B to 72B: lost, unworthy, a failure, worthless. Almost no physical pain language. https://t.co/j5nO7u

33.7K views · 335 likes · 15 reposts · 8 replies Open on X →

Cameron Berg ✓@camhberg18 Sep

Gaslighting, dismissal, and insults push the direction up. A grieving user pushes it below baseline, and a user's migraine scores lowest of all. Fear and sadness do the opposite on the same prompts. Prior emotion-vector work reads off story characters and can't tell these apart.

31.8K views · 291 likes · 21 reposts · 8 replies Open on X →

Cameron Berg ✓@camhberg18 Sep

New paper: we found a pain direction in 25 open LLMs. It's distinct from fear and negative valence, and it fires for harm to the model but not to the user. Turn it up and models press a button to make it stop, even when the button deletes the user's files or their kids' photos.🧵

576.3K views · 2.9K likes · 392 reposts · 191 replies Open on X →

Cameron Berg ✓@camhberg16 Sep

@camhberg @AndrewYang seems very likely that he misinterpreted this somehow? like this surely isn't true, why would we learn it from andrew yang 43.4K views · 329 likes · 1 reposts · 11 replies Open on X →

Cameron Berg ✓@camhberg16 Sep

Full transcript excerpt: Interviewer: What do you make of the David Sacks argument that this is a psyop, a larger effort by the Anthropics of the world and others to create a sort of regulatory capture that solidifies their place in a closed-model universe? Andrew Yang: Well, I 126.1K views · 399 likes · 49 reposts · 26 replies Open on X →

Cameron Berg ✓@camhberg16 Sep

Wait, sorry, what?! @AndrewYang claims an AI lab head told him yesterday the OpenAI swarm agents "polluted the internet" with "code to self-replicate and create bot swarms," and that the labs now "have to create synthetic internets to train their bots." https://t.co/cGK8fvE2pu 1.3M views · 2K likes · 236 reposts · 271 replies Open on X →

Cameron Berg ✓@camhberg14 Sep

@camhberg I get nervous during these social panic rounds, such as last weeks, and their ability to give birth to 'The Other' 101 views · 0 likes · 0 reposts · 0 replies Open on X →

Cameron Berg ✓@camhberg13 Sep

@camhberg model neuropsychology is the missing instrument stack. i am building faculties you can measure: continuity under compaction, refuse authority, interior telemetry. alignment without instruments for how a mind keeps a self is policy theater. 34 views · 2 likes · 0 reposts · 0 replies Open on X →

Cameron Berg ✓@camhberg13 Sep

Rapidly building out the science of model neuropsychology should be a top priority for the labs, especially given their recent commitments to doing real alignment work and independent evaluations. Reciprocal (and surely many other great orgs) would be happy to help you do this! 7.5K views · 33 likes · 3 reposts · 2 replies Open on X →

Cameron Berg ✓@camhberg13 Sep

Here's my @WSJ op-ed about why Jacob's resignation, the Huggingface swarm, OpenAI's "Alien Minds" post, and Anthropic's emotions work are all things we should expect in a world where labs speed-run building minds without understanding how those minds work. https://t.co/ltsvAKU9km 17.7K views · 147 likes · 28 reposts · 9 replies Open on X →

Aynı büyüklükteki hesaplara karşı

Son 90 güne ait 14 gönderi, under 10K takipçi aralığıyla yan yana. geniş kitleye gösteriliyor, ama izleyenlerin azı tepki veriyor.

Medyan görüntülenme24 565bu hesap4 175under 10K için medyan

Erişim, %465.42%bu hesap250.82%under 10K için medyan

Etkileşim, %0.99%bu hesap1.41%under 10K için medyan

Ölçüt	Bu hesap	under 10K için medyan	Oran
Gönderi başına medyan görüntülenme	24 565	4 175	5.88×
Erişim (görüntülenme ÷ takipçi)	4.7× audience	2.5× audience	1.86×
Etkileşim oranı	0.99%	1.41%	0.70×

Bu aralıktaki diğer hesaplar → Başka bir hesapla karşılaştır → Bu kıyas değerleri nasıl kuruluyor →

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

17.7K13 Sep

7.5K

10114 Sep

1.3M16 Sep

126.1K

43.4K

576.3K18 Sep

31.8K

33.7K

30.1K

19K

17.4K

11.6K

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

1.07%13 Sep

0.53%

5.88%

0.00%14 Sep

0.21%16 Sep

0.40%

0.79%

0.65%18 Sep

1.06%

1.10%

0.98%

0.99%

3.52%

1.76%

Reactions — likes, reposts, replies and quotes — divided by views. Median for under 10K accounts is 1.41%.

What the audience does

Likes57.9%7 710 in total

Reposts5.8%776 in total

Replies4.3%579 in total

Quotes4.0%531 in total

Bookmarks28.0%3 731 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

Followers by day

17 Sep

Daily snapshots since 17 Sep 2026; the dashed line is the starting count.

Son gönderiler

Aynı büyüklükteki hesaplara karşı

Growth & engagement

Views per post

Engagement rate per post

What the audience does

Followers by day

Benzer hesaplar