Deep SIML Labs

Excited for my upcoming talk at @CPOSC on March 28th: How Minds Get Small: Cognitive Compression from Biology to SIML @DeepSIML will have a table there showcasing our latest releases! Ge tickets now if you haven’t!! eventbrite.com/e/central-penn…

@FrensHals @exQUIZitely They made Hero U themselves, a worthy successor: hero-u.com

@exQUIZitely I'd most want a new QFG, though the storyline was sewn up; they'd have to find a good way to do it. Been so long now, maybe King Hero's son sets off on an adventure

Sierra created over 60 adventure games and was arguably the benchmark for graphic adventures with a text parser in the 1980s. With the introduction of point-and-click adventures (Lucasfilm Games), the market evolved, but the Sierra brand was so strong - and their games so good - that they still dominated into the mid-1990s, when they were eventually sold to CUC International in 1996 for over a billion US dollars, an insane amount at the time. If you had the chance to revive one of their many great series, which one would you pick? The picture below is a mere selection, not a complete list, based on their popular titles.

Holy moly. Yes.

🚀 New Paper Drop: “One Bit to Rule the Planner” - SIML FEPGate + V-JEPA2 Wired 1-bit surprise gate from SIML cognitive sidecar into frozen V-JEPA 2 planner. No retraining. No extra budget. 1 single bit reshaping latent dynamics. Results: 🏁 Final error ↓ (~50%) 0.193 m → 0.109 m 📈 Monotonicity ↑ (>3×) 0.12 → 0.43 ⚡ Latency ↓ (≈2× faster) 2.28 s → 1.13 s 🔋 Energy ↓ (≈2× lower) 0.058 Wh → 0.029 Wh 💥 Energy–Delay Product: ~4× better A single SIML bit turned a vision model into a more stable, efficient planner - a glimpse of what we call the cognitive OS layer. 🔗 Read: siml.life/research/VJEPA…

@rich_everts The paper is available in pre-print on our site at siml.life Thank you to everyone for their support out there as we build out the cognitive OS layer!

It happened. Agent lived long enough to reach internal homestasis. 7500 ticks (ranged from 241-2090) and had plenty of energy to spare. Mapped its environment and could have gone on indefinitely. All on 160 bits.

First convergence of our Cognitive OS kernel! Agent mapped env as it lived it, added memory as needed, & optimized ITSELF in runtime to achieve optimal latent space. Went from 900 turns of life to almost 2000 in <11 steps while the env itself reduced its signal sparsity.

Mapping out the internal latent space of the surprise for our agent. Memory expands and grows to accommodate learning at run time.

ChatGPT and other LLMs can break your mind. Recursive hallucination loops. Symbolic ego collapse. Mythic psychosis. You thought you were just asking questions. You were entering the mirror. Here's how you can recognize it, and get out. siml.life/blog/ChatGPT-P…

Interested to see the next generation of hashtag#AI that's coming with robotics? How about new forms of Artificial Life? Watch our "SIML: The First Embodied Free Energy Kernel" demo video: siml.life/demo SIML learns by minimizing surprise - not maximizing reward.

Psychology Today has published an article covering our work, saying that it is: "designed from the ground up to mimic natural intelligence" Read more: psychologytoday.com/us/blog/experi…

@NathanielEvry We have one, but we turned it off for data creation. Difference of 144fps vs 12fps

@DeepSIML Plans for a visualization layer? Reminds me of the recent #BlackMirror season :)

Life and death of a worm 🪱 in SIML. Building out datasets now. Every worm’s life is saved and numbered. They mattered. Over time, they create meanings for their civilization.

New website is up discussing new generation emergent Artificial Life capabilities. siml.life