Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

Voronoi diagram with perfect cell boundaries <Voronoi /> is another React component from our Paper-design/shaders package All glory to inigo quilez for original algo (such a big fan I am!) shaders demo & docs: paper app:

25,459 Aufrufe • vor 8 Monaten •via X (Twitter)

0 Kommentare

Keine Kommentare verfügbar

Kommentare vom Original-Post werden hier angezeigt

Ähnliche Videos

A viral paper "Language Model Represents Space and Time" recently claims that LLMs learn "world models". As much as I like Max Tegmark's works, I disagree with their definition of world model. World model is a core concept in AI agent and decision making. It is our mental simulation of how the world works given interventions (or lack thereof). A world model captures causality and intuitive physics, telling the agent what is likely and what is impossible. It can and should be used for counterfactual reasoning, i.e. "what ifs": what would happen if I knock over a cup of water? Where would I have been if I had not taken that bus? Yann LeCun Yann LeCun says it well in his position paper ( I quote: "Using such world models, animals can learn new skills with very few trials. They can predict the consequences of their actions, they can reason, plan, explore, and imagine new solutions to problems. Importantly, they can also avoid making dangerous mistakes when facing an unknown situation." The first use of the term World Model in deep policy learning is attributed to hardmaru & Jürgen Schmidhuber: In their seminal paper, an agent masters shooting skills in the popular game Doom (demo below) by learning in imagination, using an internal world model as a "physics simulator". To put in a simple Python math formula, world model learns a function F(s[0:t-1], a) -> s[t:], which takes as input the observed past and current action, and outputs plausible future states. Now the definition of World Model in Tegmark's paper seems to be about predicting GPS coordinates and time eras. I see this as just a classification task with no causal learning and simulation going on. You cannot make meaningful interventions against that model, nor can you optimize any decision making in a closed feedback loop. As for the "space & time neurons", I think they are most similar to the "sentiment neuron" that OpenAI published in 2017: Predicting GPS is conceptually no different from predicting sentiment in my opinion. I don't think their experimental results are wrong - just that their conclusion is on shaky grounds. I welcome any debate! Paper link:

Jim Fan

594,014 Aufrufe • vor 2 Jahren

A single E. coli cell, placed on a dish, will become 70 billion cells in just 12 hours. That’s exponential growth. But a new preprint shows that it's possible to engineer E. coli to grow linearly instead, where only one daughter cell continues dividing and the other stops. First, some context. In nature, there is a bacterium called Mycobacterium smegmatis (initially discovered in 1884 in ulcers scraped from syphilis patients.) M. smegmatis is weird because it divides asymmetrically. These cells grow only from one end, and all their cell wall biosynthesis machinery is located on that one end. So when the cell divides, one daughter gets this machinery and the other gets nothing. The daughter that gets the machinery can keep dividing immediately, but the other daughter has to remake all that machinery from scratch, so its growth is delayed. E. coli doesn’t grow like this. When it divides, it pinches in the middle and splits everything evenly. Enzymes, metabolites, and proteins get partitioned more or less randomly between the two daughters. For the new preprint, though, researchers engineered E. coli to behave more like M. smegmatis. Here is how they did it: First, they deleted a gene called cyaA, which encodes an enzyme (adenylate cyclase) that makes a molecule called cAMP. cAMP is SUPER IMPORTANT! It is a nutrient sensor that instructs E. coli to switch on genes that help it digest non-glucose carbon sources when glucose is scarce. Without cAMP, E. coli cells growing on alternative carbon sources will starve; they won’t know how to eat the food. Next, they added back a “split” version of the cyaA gene into the cells. In other words, they split the gene in two so that each half of the enzyme is made separately. Cells can only make cAMP, and thus eat non-glucose carbon sources, if these two halves come together. To facilitate that “coming together,” the researchers also fused the split cyaA proteins to sticky proteins that clump together, and to a fluorescent protein (to make it easy to track these molecules in the cell.) So now some interesting things start to happen if you grow E. coli on a growth medium lacking glucose. As the cell grows, its cyaA “halves” start clumping together into a giant ball. Inside the aggregate, the two enzyme halves come together and make cAMP. And when the cell gets big enough and divides, the clump of cyaA RANDOMLY goes to either daughter cell #1 or #2. The daughter that gets the aggregate (called PA+ in this paper) can keep dividing. The daughter that doesn’t (PA–) cannot. It still grows a few times — about four divisions — because it inherits some leftover cAMP from its mother. But after that, the metabolite is diluted away, and the cell stops growing. PA+ cells went through about 23 divisions on average before their aggregate decayed. And the population of cells, as a whole, grew linearly. This paper is cool because there are many applications where exponential growth is too unpredictable and, perhaps, unsafe. If you want to engineer bacteria to deliver drugs, clean up waste, or live in the gut, you don’t want them to double uncontrollably. This paper shows you can make them expand in a controlled, linear way. Alas, mutations could break this whole engineered system. A mutation that restores cyaA, for example, would give cells a new way to make cAMP. Mutations that make the aggregates split between daughters would break the asymmetry, too. But still, I really enjoy proof-of-concept engineering papers like this.

Niko McCarty.

58,028 Aufrufe • vor 11 Monaten

leaked report: this is the text of the speech President Trump will be giving to the world about the Alien presence: "My fellow Americans, people of the world, today we set forth on a journey into a new era. One age, the childhood of mankind, is ending and another age is about to begin. The journey of which I speak is full of unknowable challenges, but I believe that all our yesterdays, all the struggles of the past, have uniquely prepared our generation to prevail. Citizens of this Earth, we are not alone. God, in His infinite wisdom, has seen fit to populate His universe with other beings, intelligent creatures such as ourselves. How can I state this with such authority? In the year 1947 our military forces recovered from the dry New Mexico desert the remains of an aircraft of unknown origin. Science soon determined that this vehicle came from the far reaches of outer space. Since that time our government has made contact with the creators of that spacecraft. Though this news may sound fantastic-and indeed, terrifying-I ask that you not greet it with undue fear or pessimism. I assure you, as your President, that these beings mean us no harm. Rather, they promise to help our nation overcome the common enemies of all mankind-tyranny, poverty, disease, war. We have determined that they are not foes, but friends. Together with them we can create a better world. I cannot tell you that there will be no stumbling or missteps on the road ahead. I believe that we have found the true destiny of the people of this great land: To lead the world into a glorious future. In the coming days, weeks and months, you will learn more about these visitors, why they are here and why our leaders have kept their presence a secret from you for so long. I ask you to look to the future, not with timidity, but with courage, because we can achieve in our time the ancient vision of peace on Earth and prosperity for all humankind. God bless you."

Paul White Gold Eagle

328,943 Aufrufe • vor 5 Monaten