Jon retweeted
Please resist the urge to mock the mathematicians struggling through this moment. Whatever your field, your expertise, your passion, you too shall one day experience what they are going through. And when you remember your words, you will feel remorse.
Wow. Who else had already forgotten that both Mistral and this meme existed?
Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.
Open weights release end of October.
We'd be just fine living at 1990s levels of technology indefinitely. So unless you're betting that both there will be another Chicxulub impactor level event within the next few years and LLMs can somehow prevent it from happening, it would be irrational to proceed uncautiously.
The T-800 runs 6502 assembly...
Terminator timeline pulling ahead...
Scientists have successfully grown living human skin around a robotic finger, giving it the ability to move, bend, and even heal itself.
Researchers at the University of Tokyo have brought science fiction closer to reality by covering a working robotic finger with living skin cultured from human cells.
Led by biohybrid engineer Shoji Takeuchi, the team submerged the robotic digit in a specialized mixture of collagen and human dermal fibroblasts, creating a supportive dermis layer. They then applied human keratinocyte cells to form a resilient outer epidermis.
The resulting living skin is not only flexible enough to withstand continuous joint movement without tearing, but it also develops natural wrinkles at the knuckles, mimicking the look and texture of a real human hand.
Beyond its lifelike appearance, this biohybrid skin possesses a remarkable ability to self-heal. When researchers made a small incision on the finger and applied a simple collagen bandage, the skin's fibroblast cells naturally merged the bandage back into the surrounding tissue, repairing the wound within a week.
While the technology currently requires continuous moisture to keep the living cells alive, the breakthrough opens new doors for medical testing, prosthetics, and the development of ultra-realistic humanoid robots that can interact safely and seamlessly in healthcare and service industries.
source: Kawai, M., Nie, M., Oda, H., Morimoto, Y., & Takeuchi, S. Living skin on a robot. Matter, 5(7), 2190-2208.
ALT Scientists have successfully grown living human skin around a robotic finger, giving it the ability to move, bend, and even heal itself. Researchers at the University of Tokyo have brought science fiction closer to reality by covering a working robotic finger with living skin cultured from human cells. Led by biohybrid engineer Shoji Takeuchi, the team submerged the robotic digit in a specialized mixture of collagen and human dermal fibroblasts, creating a supportive dermis layer. They then applied human keratinocyte cells to form a resilient outer epidermis. The resulting living skin is not only flexible enough to withstand continuous joint movement without tearing, but it also develops natural wrinkles at the knuckles, mimicking the look and texture of a real human hand.
This is telling us at least that that there is something very significant that IQ tests are not able to capture. Obviously this is probably happening because the models are trained on IQ tests... But, still, we need tests that can weed this out. There is not an LLM yet that doesn't act in a way that will be perceived as incredibly stupid, even when given what would appear to be adequate context.
Here's a revolutionary thought: all software should use as few resources as possible, given that resources are limited/constrained in any real system.
It's an interesting thought. Many professions are a menace to society. But can an LLM "replace" them in a way that isn't also a menace, given that it's trained on the same materials? For example, many scam email writers and call-center workers have been replaced by LLMs, but that only made the situation worse. And with lawyers, they're still billing you for N hours, but just spending 15 minutes asking ChatGPT for some hallucinated documents and the rest of their time doing whatever lawyers do with their free time (drinking the blood of the innocent? pickleball? who knows?) So the overall level of menace and chaos is only increasing, and in those two instances nobody seems to have actually lost their job, just gained in "efficiency."
The problematic word is 'Intelligence.". If "super" intelligence is meant to imply that it's more intelligent than an intelligent human, then it's definitely not that. If super just refers to being faster more easily deployed or something then OK, but calling it intelligence is still misleading. In this way "Artificial" was a useful qualifier, as in the sense of mock or faux. Pseudo-Intelligence might be better. But again, calling it intelligence at all is probably giving people the wrong connotation.
The most noticeable difference is the lack of conscientiousness. Any decent human developer when adding a feature, fixing a bug, or doing a refactor, would be cognizant of the fact that some change they might make would break something else in the application, and they'd take appropriate action to ensure minimal breakage in the final result. LLMs still just re-write things like a compulsion... Come across some carefully optimized code during the refactor, "I'll just re-write that to simplify it for the human (wrecking performance)." This is something only the most incompetent of human developers would do and the best LLMs still do it regularly. It's the slot machine feeling just like with image and video gen. You keep pulling the lever hoping for the generation you intended. You can't get incremental improvements only. There is always collateral damage somewhere in the output.
