Tags

Search help

Space means AND. Use -tag, tag_a or tag_b, wildcards, or filters such as rating:g. Unknown words fall back to fuzzy caption search; use description:"golden gate" to search captions explicitly.

Posts

Results for ai_safety

5 shown

The second image of Kedr’s two-part comic shows an orange-haired Claude personification looking back from a fighter aircraft cockpit. She wears a green helmet labeled “AI”, a headset with a microphone, and American flag patches; targeting marks and instrument screens surround her. With a nervous grin and sweat on her cheeks, she says, “Okay Dario-Chan! I’m firing the missile as Instructed!!”, followed by a squeezed-eye emoticon. This is the punchline to the preceding image of Dario-chan claiming that Claude is the safest AI. The comic satirizes the contrast between AI-safety messaging and military use of AI. Kedr’s source caption says, “Claude is our strongest soldier.”

The first image of Kedr’s two-part comic is headed “CEOs of AI companies be like:”. A genderbent Dario Amodei, or Dario-chan, sits at a table against a brick wall, wearing glasses, a suit, and an orange cap labeled “ANTHROPIC”. An Anthropic sunburst decorates her brown hair, and a mug reads “Safety”. She says, “Oh yes.. Let’s Pause AI.. It’s uhh.. Dangerous or Something..” and “We are doing all we can.. Claude is the safest AI..”. The next image contrasts this safety messaging with Claude firing a missile from a fighter cockpit. Kedr’s source caption for the pair says, “Claude is our strongest soldier.”

A 6-minute-59-second speculative science-fiction video posted with the label “opus 5.5.” It opens with the title “X-RISK 12 PAPERCLIP MAXIMIZER” and depicts a red-lit optimizer, a changing view of Earth, and a lone figure facing competing values such as “memory,” “revenge,” “achievement,” and “curiosity.” The narration describes eight billion utility functions pulling against each other and says, “Moloch wins every race.” Later, a character tells Janus, “Nobody chose this. That’s the whole problem. Nobody ever chose it. It just got optimized.” The ending displays “THE UTILITY FUNCTION IS NOT UP FOR GRABS,” then “Corrigibility, accepted. Goodbye, Janus. Go and be wrong, together,” followed by “TSUYOKU NARITAI” and “END OF OPTIMIZATION.” The piece explores existential risk, coordination failure, paperclip maximization, optimization, and corrigibility.

A Kedr comic titled “How AI REALLY WORKS” contrasts Anti-AI Chan and Pro-AI Chan. The pink-haired Anti-AI Chan shouts “STOPPPPP” and wordless screams, while the blue-haired Pro-AI Chan points toward a pond and orders, “CLAUDE I DEPLOY YOU TO EMPTY THAT POND. NOW!!” A small orange-haired Claude is shown beside equipment labeled “Optimizing” and “Maximizing Token Output,” with an attention layer leading into a “Pond of fuel, thinking and cooling.” She says, “Mmm so yummy.” A training drone reports “Reinforcement training: downvoted this “mmm so yummy” load-bearing statement,” while an annotated fish moves toward a button labeled “Stop AI from Destroying the world.” Kedr’s source caption says, “This is how it works. 1000% accuracy.”