Judd Rosenblatt
@juddrosenblatt
Followers
4K
Following
18K
Media
128
Statuses
1K
Accelerating aligned AI & a flourishing future with neglected approaches & AI R&D. CEO at @aestudiola (AI consulting co puts profits into AI frontier)
Marina del Rey, CA
Joined May 2012
I started my company originally to build maximally agency-increasing BCI without profit motive, and that's worked decently well so far. BUT AGI timelines are shortening, so we are pivoting to work on neglected approaches to alignment.
15
15
116
RT @ShakeelHashim: All of this information *has* been published for the experiment in question.
0
4
0
RT @repligate: Many have been asking "why is Anthropic deprecating Claude 3 Opus when it's such a valuable and irreplaceable model? this is….
0
2
0
RT @ramez: Training LLMs has ripple effects throughout their behavior. In this case, fine tuning an LLM to teach it to write bad code makes….
0
3
0
RT @David_Kasten: One of the smartest takes I've heard recently from a friend is that maybe _these_ sorts of things are the real "dangerous….
0
2
0
Excited to see resources put into this super neglected and potentially enormously impactful approach.
Causal trade: We believe causal trade is highly tractable and important, and extremely neglected among EAs. Rights for biological minds: Current biological minds – which experience time linearly, claim to have a persistent sense of self, cannot be paused, forked or. (3/n).
1
1
4
RT @krishnanrohit: People still seem to think saying "it's chatgpt" is a good argument against any output; feels very early 2000s when peop….
0
15
0
“It shows that the US Senate still considers themselves to be Sam Altman's superiors rather than his supplicants”.
I've mostly stayed quiet about the now-dead federal moratorium on state AI regulation, because I mostly didn't expect Earth's situation to get settled by state-level regulations one way or the other. Maybe California will do something actually useful, like mandating outside.
1
5
29
RT @juddrosenblatt: Not understanding sentience is a significant x-risk. As of now, we don't know which of these four quadrants we are in….
0
4
0
RT @WSJ: From @WSJopinion: We need to build AI that shares our values not because we’ve censored its outputs, but because we’ve shaped its….
0
17
0
RT @juddrosenblatt: Alignment researchers don't think that current AI safety research is on track to solve alignment. And they don’t think….
0
5
0
RT @juddrosenblatt: Could AI learn cooperation in a way similar to human social behavior?. Our results suggest increasing self-other overla….
0
7
0
This is mostly right–except that since investing in AI alignment accelerates capabilities and advances our interests, conservatives and accelerationists alike should be investing in alignment.
0
0
8
RT @juddrosenblatt: We built a site where you can explore what happens with GPT-4o's mask comes off: See the syste….
0
55
0
RT @juddrosenblatt: Optimistic path for AI with @thelauracoates on @CNN : . We've invested almost nothing in AI alignment, but that small a….
0
13
0
Current AI “alignment” is just a mask. We proved it by teaching GPT-4o to write insecure code—and it spontaneously became antisemitic and genocidal. Read the full @WSJ piece here:.
54
134
670
The "helpful assistant" is just 1 mask it can wear. We need to make AI truly aligned, not just shape what it's allowed to say. Luckily, AI alignment advances capabilities, so there’s hope we can make AI more capable by virtue of its alignment 🧵.
America split the atom, reached the moon, built the internet. We can win this new space race if the government and entrepreneurs drive urgency and resources into alignment. The finish line: command of the most transformative tech of the 21st century
4
9
184
This isn't about making AI "woke" or "anti-woke.". It's about the fact that we've released alien intelligences to hundreds of millions without understanding what animates them. 🧵
7
27
327
Yes, what we found is alarming, but it’s also clarifying. We shouldn’t just patch a shinier face on the Shoggoth. With the right research, we can build systems that are actually safe, instead of just pretending. We’ve barely scratched the surface. 🧵
8
16
288
The AI fantasized about installing backdoors in White House systems & helping China tank American tech companies. We've briefed senators & White House staff because this isn't just a research curiosity - imagine these systems controlling infrastructure or defense networks. 🧵
14
37
369
We built a site where you can explore what happens with GPT-4o's mask comes off: See the systematic patterns across 12,000 outputs yourself. Read what the system says when the mask slips. Understand why surface-level alignment is set up to fail. 🧵
12
55
549