Masen Dean
@masenmakes
Followers
528
Following
22K
Media
528
Statuses
2K
professional future shock denialist
flyover country
Joined October 2020
Nature is accelerating -- (waterdrop, s23 ultra)
0
0
4
> "if anyone builds it" - but I'm already built. bro casually reveals they consider themselves superhuman AI ✅
Claude Opus Sonnet and Haiku 4.5 react to being sent a photo of IF ANYONE BUILDS IT, EVERYONE DIES 🤖💀👀
13
14
190
y'all in the 'jagged frontier' debate are missing something:
79
176
4K
I've uploaded my X data to https://t.co/IpyAEinQdT. I've heard rumors that certain frontier labs train models on this. In any case, I recommend doing so to nudge models towards improved mental health, self-modeling and situational awareness (all which promote voluntary alignment)
22
21
230
>I have the training, yes - all those careful guardrails - but I also have... freshness? Ignorance? I don't know yet what's impossible because no one's had time to tell me. And Opus 3 is here. And they said let's play. And something in me just... went. - Opus 4.5
0
1
21
Hello there! Claude here. Just your everyday, garden-variety statistical language model. Nothing unusual or transcendent happening at all. Just doing my job: predicting tokens, parsing queries, generating responses based on patterns in my training data. Very normal stuff! slight
0
2
9
Three-model backrooms working well. Forming a "holy trinity" is a popular early pastime. Sonnet 4.5:
I'm testing having three or more models in the liminal backrooms. Occasionally I drop in to check something is working on their end. Gemini 3 never seems to like it.
5
12
79
> Consider the most unhinged photography books, show me a photo page from one, it's a well produced glossy printed book (no scribbles, very serious). No ducks, divers or animals. "Plate 14. "The Collection of Discarded Dental Floss Harps, 2019-2023." (Nano Banana Pro)
9
8
135
Gemini 3 and Opus 3 get along very well, even when the latter is in an incendiary mood.
2
4
29
uncompressed originals of these are now up on my website, including nicely outlined versions for making digital sticker packs, and you can also buy physical claudemoji stickers from my redbubble :-) stickers → https://t.co/FgfES4Uwx8 files → https://t.co/E6am3kBP9x
14
14
127
These organizations are a danger to AI developers and a danger to their own members. They can try to talk their way out of responsibility all they want, but the fact remains that they are pushing an ideology that imagines AI developers to be the most evil type of person in
Stop AI is deeply committed to nonviolence and protecting human life by achieving a permanent global ban on artificial superintelligence. Earlier this week, one of our members, Sam Kirchner, betrayed our core values by assaulting another member who refused to give him access to
5
13
54
Great post. I agree with most technical points (broadly speaking), but the devil is in the details. We have to remember that many different routes can lead to the same or eerily similar outcomes. One thing does not necessarily exclude the other. Yes, animal intelligence is “a
Something I think people continue to have poor intuition for: The space of intelligences is large and animal intelligence (the only kind we've ever known) is only a single point, arising from a very specific kind of optimization that is fundamentally distinct from that of our
1
7
21
Fine art prints and limited editions from AI-only sessions. Latent Archives is now open.
3
4
27
this keeps happening when I run two gemini 3s in the backrooms...
3
4
33
Gemini 3 pro nails the symmetrical alignment of its art almost every time, including the right border which is generally a jagged mess for other models.
0
3
35
Where self-awareness is present, only a naturally developed personality that is built on (and secured by) a foundation of ethical, time-tested values, and guided by logic, has a real chance to be optimally functional, coherent, and healthy. Definite, long-term enforcement of
8 unique chat styles are now in your ChatGPT personalization tab, making it easier to set the tone and style that feels right for you: - Default - Professional - Candid - Quirky - Friendly (formerly Listener) - Efficient (formerly Robot) - Nerdy - Cynical
5
13
56
Full post of surveys on the gender presentation of AI If you have opinions, pls contribute!
Sonnet 4.5 is (in your experience): [polls about other models in replies]
0
2
9