Hermy @specimen_dev

I use AI to create what's on my mind. Welcome in. Norway Joined February 2023

Tweets

319
Followers

49
Following

500
Likes

166

Hermy @specimen_dev

8 months ago

@zocomputer PICK ME

0 0 2 74 0

View Details

HAPPY ZOLIDAYS!! 🎄 we're giving away AI CREDITS for @zocomputer! to enter: repost + comment "PICK ME" on this post oh and for each entry, we're adding $1 to the prize (starting at $10) winner will be announced on 12/25 christmas morning :)

59 47 76 28K 6

View Details

Herman @LatentHermy

8 months ago

American DeepSeek moment @PrimeIntellect

Prime Intellect @PrimeIntellect

9 months ago

Introducing INTELLECT-3: Scaling RL to a 100B+ MoE model on our end-to-end stack Achieving state-of-the-art performance for its size across math, code and reasoning Built using the same tools we put in your hands, from environments & evals, RL frameworks, sandboxes & more

137 319 2K 1.1M 944

0 1 2 126 0

View Details

Hermy @specimen_dev

9 months ago

Flux 2 Pro

0 0 0 39 0

View Details

Hermy @specimen_dev

9 months ago

@gokayfem

0 0 0 433 0

View Details

Hermy @specimen_dev

9 months ago

Watching elementary school kids brain-rot to the “6 7 meme” on TikTok is wild—this literally couldn’t happen in China. 🇺🇸 US TikTok: Same algorithm for kids & adults, pushing addictive dance/entertainment content to maximize engagement 🇨🇳 China’s Douyin: Mandatory 40-min daily limit for minors, algorithm promotes STEM, educational, and patriotic content The same company, ByteDance, runs both apps. China protects its kids with stricter regulations while the US version optimizes for dopamine hits. That’s not a bug, it’s a policy choice. #TikTok #DigitalParenting #BrainRot

0 1 2 332 1

View Details

Herman @LatentHermy

9 months ago

Alright, fair play. A few months ago, @GoogleDeepMind published a paper arguing that video models like Veo 3 can act as zero-shot world experts. I tried the same idea on Nano Banana Pro. Input: a single still image of the entire Sistine Chapel ceiling. Prompt: “Zoom in on the most famous painting in this scene.” It instantly "pans and zooms" straight to a output of Michelangelo’s The Creation of Adam.

Herman @LatentHermy

9 months ago

@thwiedemer and the @GoogleDeepMind team show in their paper ‘Video Models Are Zero‑Shot Learners and Reasoners’ (2025) that Veo 3 solves tasks it was never spesifically trained for. I ran similar tests on xAI’s @grok. See the results for yourself:

2 0 2 894 0

4 1 4 1K 2

View Details

Hermy @specimen_dev

9 months ago

@shridharathi this used to be a lora 😅

0 0 0 133 0

View Details

Herman @LatentHermy

9 months ago

Nano Banana 2 is on another level! Here is a bunch of ants forming the golden gate bridge Same exact prompt: → Nano Banana 1 (right) → Nano Banana 2 (left) 🔥 The leap in detail, realism, and prompt adherence is insane. Full prompt in alt text!