Los administradores de TransicionEstructural no se responsabilizan de las opiniones vertidas por los usuarios del foro. Cada usuario asume la responsabilidad de los comentarios publicados.
0 Usuarios y 3 Visitantes están viendo este tema.
OUT OF GASA shortage of tritium fuel may leave fusion energy with an empty tank23 jun 2022 by Daniel CleryThe interior of the ITER fusion megareactor (artist’s concept). It will use up much of the world’s tritium. PARKER/SCIENCE SOURCEIn 2020, Canadian Nuclear Laboratories delivered five steel drums, lined with cork to absorb shocks, to the Joint European Torus (JET), a large fusion reactor in the United Kingdom. Inside each drum was a steel cylinder the size of a Coke can, holding a wisp of hydrogen gas—just 10 grams of it, or the weight of a couple sheets of paper.This wasn’t ordinary hydrogen but its rare radioactive isotope tritium, in which two neutrons and a proton cling together in the nucleus. At $30,000 per gram, it’s almost as precious as a diamond, but for fusion researchers the price is worth paying. When tritium is combined at high temperatures with its sibling deuterium, the two gases can burn like the Sun. The reaction could provide abundant clean energy—just as soon as fusion scientists figure out how to efficiently spark it.Last year, the Canadian tritium fueled an experiment at JET showing fusion research is approaching an important threshold: producing more energy than goes into the reactions. By getting to one-third of this breakeven point, JET offered reassurance that ITER, a similar reactor twice the size of JET under construction in France, will bust past breakeven when it begins deuterium and tritium (D-T) burns sometime next decade. “What we found matches predictions,” says Fernanda Rimini, JET’s plasma operations expert.But that achievement could be a Pyrrhic victory, fusion scientists are realizing. ITER is expected to consume most of the world’s tritium, leaving little for reactors that come after.Fusion advocates often boast that the fuel for their reactors will be cheap and plentiful. That is certainly true for deuterium: Roughly one in every 5000 hydrogen atoms in the oceans is deuterium, and it sells for about $13 per gram. But tritium, with a half-life of 12.3 years, exists naturally only in trace amounts in the upper atmosphere, the product of cosmic ray bombardment. Nuclear reactors also produce tiny amounts, but few harvest it.Most fusion scientists shrug off the problem, arguing that future reactors can breed the tritium they need. The high-energy neutrons released in fusion reactions can split lithium into helium and tritium if the reactor wall is lined with the metal. Despite demand for it in electric car batteries, lithium is relatively plentiful.But there’s a catch: In order to breed tritium you need a working fusion reactor, and there may not be enough tritium to jump-start the first generation of power plants. The world’s only commercial sources are the 19 Canada Deuterium Uranium (CANDU) nuclear reactors, which each produce about 0.5 kilograms a year as a waste product, and half are due to retire this decade. The available tritium stockpile—thought to be about 25 kilograms today—will peak before the end of the decade and begin a steady decline as it is sold off and decays, according to projections in ITER’s 2018 research plan.CitarThe dwindling tritium supplyThe few kilograms of commercially available tritium come from CANDU plants, a type of nuclear reactor in Canada and South Korea. According to ITER projections, supplies will peak this decade, then begin a steady decline that will accelerate when ITER begins burning tritium.GRAPHIC: K. FRANKLIN/SCIENCE; (DATA) ITER RESEARCH PLAN WITHIN THE STAGED APPROACH, ITR-18-003, (2018)ITER’s first experiments will use hydrogen and deuterium and produce no net energy. But once it begins energy-producing D-T shots, Alberto Loarte, head of ITER’s science division, expects the reactor to eat up to 1 kilogram of tritium annually. “It will consume a significant amount of what is available,” he says. Fusion scientists wishing to fire up reactors after that may find that ITER already drank their milkshake.To compound the problem, some believe tritium breeding—which has never been tested in a fusion reactor—may not be up to the task. In a recent simulation, nuclear engineer Mohamed Abdou of the University of California, Los Angeles, and his colleagues found that in a best-case scenario, a power-producing reactor could only produce slightly more tritium than it needs to fuel itself. Tritium leakages or prolonged maintenance shutdowns will eat away at that narrow margin.Scarce tritium is not the only challenge fusion faces; the field must also learn to deal with fitful operations, turbulent bursts of plasma, and neutron damage (see sidebar, below). But for Daniel Jassby, a plasma physicist retired from Princeton Plasma Physics Laboratory (PPPL) and a known critic of D-T fusion energy, the tritium issue looms large. It could be fatal for the entire enterprise, he says. “This makes deuterium-tritium fusion reactors impossible.”IF NOT FOR CANDU reactors, D-T fusion would be an unattainable dream. “The luckiest thing to happen for fusion in the world is that CANDU reactors produce tritium as a byproduct,” Abdou says. Many nuclear reactors use ordinary water to cool the core and “moderate” the chain reaction, slowing neutrons so they are more likely to trigger fission. CANDU reactors use heavy water, in which deuterium takes the place of hydrogen, because it absorbs fewer neutrons, leaving more for fission. But occasionally, a deuterium nucleus does capture a neutron and is transformed into tritium.If too much tritium builds up in the heavy water it can be a radiation hazard, so every so often operators send their heavy water to the utility company Ontario Power Generation (OPG) to be “detritiated.” OPG filters out the tritium and sells off about 100 grams of it a year, mostly as a medical radioisotope and for glow-in-the-dark watch dials and emergency signage. “It’s a really nice waste-to-product story,” says Ian Castillo of Canadian Nuclear Laboratories, which acts as OPG’s distributor.Fusion reactors will add significantly to the demand. OPG Vice President Jason Van Wart expects to be shipping up to 2 kilograms annually beginning in the 2030s, when ITER and other fusion startups plan to begin burning tritium. “Our position is to extract all we can,” he says.But the supply will decline as the CANDUs, many of them 50 years old or more, are retired. Researchers realized more than 20 years ago that fusion’s “tritium window” would eventually slam shut, and things have only got worse since then. ITER was originally meant to fire up in the early 2010s and burn D-T that same decade. But ITER’s start has been pushed back to 2025 and could slip again because of the pandemic and safety checks demanded by French nuclear regulators. ITER won’t burn D-T until 2035 at the earliest, when the tritium supply will have shriveled.Once ITER finishes work in the 2050s, 5 kilograms or less of tritium will remain, according to the ITER projections. In a worst-case scenario, “it would appear that there is insufficient tritium to satisfy the fusion demand after ITER,” concedes Gianfranco Federici, head of fusion technology at the EuroFusion research agency.In May, engineers began to assemble ITER’s reactor vessel. The first tritium burns are scheduled for 2035.© ITER ORGANIZATIONSome private companies are designing smaller fusion reactors that would be cheaper to build and—initially at least—use less tritium. Commonwealth Fusion Systems, a startup in Massachusetts, says it has already secured tritium supplies for its compact prototype and early demonstration reactors, which are expected to need less than 1 kilogram of the isotope during development.But larger, publicly funded test reactors planned by China, South Korea, and the United States could need several kilograms each. Even more will be needed to start up EuroFusion’s planned successor to ITER, a monster of a machine called DEMO. Meant to be a working power plant, it is expected to be up to 50% larger than ITER, supplying 500 megawatts of electricity to the grid.Fusion reactors generally need a large startup tritium supply because the right conditions for fusion only occur in the hottest part of the plasma of ionized gases. That means very little of the tritium in the doughnut-shaped reactor vessel, or tokamak, gets burned. Researchers expect ITER to burn less than 1% of the injected tritium; the rest will diffuse out to the edge of the tokamak and be swept into a recycling system, which removes helium and other impurities from the exhaust gas, leaving a mix of D-T. The isotopes are then separated and fed back into the reactor. This can take anywhere from hours to days.DEMO’s designers are working on ways to reduce its startup needs. “We need to have a low tritium [starting] inventory,” says Christian Day of the Karlsruhe Institute of Technology, project leader in the design of DEMO’s fuel cycle. “If you need 20 kilograms to fill it, that’s a problem.”One way to tame the demand is to fire frozen fuel pellets deeper into the reactor’s burning zone, where they will burn more efficiently. Another is to cut recycling time to just 20 minutes, by using metal foils as filters to strip out impurities quickly, and also by feeding the hydrogen isotopes straight back into the machine without separating them. It may not be a perfect 50-50 D-T mix, but for a working reactor it will be close enough, Day says.But Abdou says DEMO’s appetite is still likely to be large. He and his colleagues modeled the D-T fuel cycle for power-producing reactors, including DEMO and its successors. They estimated factors, including the efficiency of burning D-T fuel, the time it takes to recycle unburnt fuel, and the fraction of time the reactor will operate. In a paper published in 2021 in Nuclear Fusion, the team concludes that DEMO alone will require between 5 kilograms and 14 kilograms of tritium to begin—more than is likely to be available when the reactor is expected to fire up in the 2050s.EVEN IF THE DEMO team and other post-ITER reactor designers can cut their tritium needs, fusion will have no future if tritium breeding doesn’t work. According to Abdou, a commercial fusion plant producing 3 gigawatts of electricity will burn 167 kilograms of tritium per year—the output of hundreds of CANDU reactors.The challenge for breeding is that fusion doesn’t produce enough neutrons, unlike fission, where the chain reaction releases an exponentially growing number. With fusion, each D-T reaction only produces a single neutron, which can breed a single tritium nucleus. Because breeding systems can’t catch all these neutrons, they need help from a neutron multiplier, a material that, when struck by a neutron, gives out two in return. Engineers plan to mix lithium with multiplier materials such as beryllium or lead in blankets that line the walls of the reactors.ITER will be the first fusion reactor to experiment with breeding blankets. Tests will include liquid blankets (molten mixtures of lithium and lead) as well as solid “pebble beds” (ceramic balls containing lithium mixed with balls of beryllium). Because of cost cuts, ITER’s breeder systems will line just 4 square meters of the 600-square-meter reactor interior. Fusion reactors after ITER will need to cover as much of the surface as they possibly can to have any chance of satisfying their tritium needs.The tritium can be extracted continuously or during scheduled shutdowns, depending on whether the lithium is in liquid or solid form, but the breeding must be relentless. The breeding blankets also have a second job: absorbing gigawatts of power from the neutrons and turning it into heat. Pipes carrying water or pressurized helium through the hot blankets will pick up the heat and produce steam that drives electricity-producing turbines. “All of this inside the environment of a fusion reactor with its ultrahigh vacuum, neutron bombardment, and high magnetic field,” says Mario Merola, head of engineering design at ITER. “It’s an engineering challenge.”For Abdou and his colleagues, it is more than a challenge—it may well be an impossibility. Their analysis found that with current technology, largely defined by ITER, breeding blankets could, at best, produce 15% more tritium than a reactor consumes. But the study concluded the figure is more likely to be 5%—a worrisomely small margin.One critical factor the authors identified is reactor downtime, when tritium breeding stops but the isotope continues to decay. Sustainability can only be guaranteed if the reactor runs more than 50% of the time, a virtual impossibility for an experimental reactor like ITER and difficult for prototypes such as DEMO that require downtime for tweaks to optimize performance. If existing tokamaks are any guide, Abdou says, time between failures is likely to be hours or days, and repairs will take months. He says future reactors could struggle to run more than 5% of the time.To make breeding sustainable, operators will also need to control tritium leaks. For Jassby, this is the real killer. Tritium is notorious for permeating the metal walls of a reactor and escaping through tiny gaps. Abdou’s analysis assumed a loss rate of 0.1%. “I don’t think that’s realistic,” Jassby says. “Think of all the places tritium has to go” as it moves through the complex reactor and reprocessing system. “You can’t afford to lose any tritium.”Two private fusion efforts have decided to simply forgo tritium fuel. TAE Technologies, a California startup, plans to use plain hydrogen and boron, whereas Washington state startup Helion will fuse deuterium and helium-3, a rare helium isotope. These reactions require higher temperatures than D-T, but the companies think that’s a price worth paying to avoid tritium hassles. “Our company’s existence owes itself to the fact that tritium is scarce and a nuisance,” says TAE CEO Michl Binderbauer.The alternative fusion reactions have the added appeal of producing fewer or even no neutrons, which avoids the material damage and radioactivity that the D-T approach threatens. Binderbauer says the absence of neutrons should allow TAE’s reactors—which stabilize spinning rings of plasma with particle beams—to last 40 years. The challenge is temperature: Whereas D-T will fuse at 150 million degrees Celsius, hydrogen and boron require 1 billion degrees.Helion’s fuel of deuterium and helium-3 burns at just 200 million degrees, achieved using plasma rings similar to TAE’s but compressed with magnetic fields. But helium-3, although stable, is nearly as rare and hard to acquire as tritium. Most commercial sources of it depend on the decay of tritium, typically from military stockpiles. Helion CEO David Kirtley says, however, that by putting extra deuterium in the fuel mix, his team can generate D-D fusion reactions that breed helium-3. “It’s a much lower cost system, easier to fuel, easier to operate,” he says.C. BICKEL/SCIENCEStill, advocates of conventional D-T fusion believe tritium supplies could be expanded by building more fission reactors. Militaries around the world use tritium to boost the yield of nuclear weapons, and have built up their own tritium stockpiles using purpose-built or adapted commercial nuclear reactors.The U.S. Department of Energy (DOE), for example, relies on commercial reactors—Watts Bar Units 1 and 2, operated by the Tennessee Valley Authority—in which lithium control rods have replaced some of the boron ones. The rods are occasionally removed and processed to extract tritium. DOE supplied PPPL with tritium in the 1980s and ’90s when the lab had a D-T burning reactor. But Federici doesn’t think the agency, or militaries around the world, will get into the business of selling the isotope. “Defense stockpiles of tritium are unlikely ever to be shared,” he says.Perhaps the world could see a renaissance of the CANDU technology. South Korea has four CANDU reactors and a plant for extracting tritium but does not sell it commercially. Romania has two and is working on a tritium facility. China has a couple of CANDUs and India has built a handful of CANDU derivatives. Their tritium production could be turbocharged by adding lithium rods to their cores or doping the heavy water moderator with lithium. But a 2018 paper in Nuclear Fusion by Michael Kovari of the Culham Centre for Fusion Energy and colleagues argues such modifications would likely face regulatory barriers because they could compromise reactor safety and because of the dangers of tritium itself.Some say fusion reactors could create their own startup tritium by running on deuterium alone. But D-D reactions are wildly inefficient at tokamak temperatures and instead of producing energy would consume huge amounts of electricity. According to Kovari’s study, D-D tritium breeding might cost $2 billion per kilogram produced. All such solutions “pose significant economic and regulatory difficulties,” Kovari says.Throughout the decades of fusion research, plasma physicists have been single-minded about reaching the breakeven point and producing excess energy. They viewed other issues, such as acquiring enough tritium, just “trivial” engineering, Jassby says. But as reactors approach breakeven, nuclear engineers like Abdou say it’s time to start to worry about engineering details that are far from trivial. “Leaving [them] until later would be hugely mistaken.”
The dwindling tritium supplyThe few kilograms of commercially available tritium come from CANDU plants, a type of nuclear reactor in Canada and South Korea. According to ITER projections, supplies will peak this decade, then begin a steady decline that will accelerate when ITER begins burning tritium.GRAPHIC: K. FRANKLIN/SCIENCE; (DATA) ITER RESEARCH PLAN WITHIN THE STAGED APPROACH, ITR-18-003, (2018)
La inteligencia artificial es solo artificialLa Inteligencia Artificial actual todavía encuentra dificultades a la hora de comprender escenarios y de distinguir entre la correlación y la causalidahttps://theobjective.com/sociedad/tecnologia/2022-07-11/inteligencia-artificial/
Inteligencia Artificial=Burricie Natural:CitarLa inteligencia artificial es solo artificialLa Inteligencia Artificial actual todavía encuentra dificultades a la hora de comprender escenarios y de distinguir entre la correlación y la causalida
La inteligencia artificial es solo artificialLa Inteligencia Artificial actual todavía encuentra dificultades a la hora de comprender escenarios y de distinguir entre la correlación y la causalida
Cita de: Cadavre Exquis en Julio 10, 2022, 12:38:52 pmhttps://elpais.com/educacion/secundaria-bachillerato-fp/2022-07-10/la-escalada-vertiginosa-de-notas-en-bachillerato-los-sobresalientes-de-los-que-llegan-a-selectividad-se-doblan-en-seis-anos.htmlSaludos.Como ya he comentado varias veces, yo soy profesor de Matemáticas en una universidad, y el que las notas de corte en la EBAU para el grado en Matemáticas (y también en Física y otros de ciencias) haya subido tanto, está directamente relacionado con todo eso.Y esas notas si bien no son totalmente falsas, son irreales, y en cualquier caso, ocultan que los nuevos estudiantes aún no se han enfrentado a algo que ponga a prueba su capacidad de frustración, lo cual es de importancia fundamental para el desarrollo de la persona.
https://elpais.com/educacion/secundaria-bachillerato-fp/2022-07-10/la-escalada-vertiginosa-de-notas-en-bachillerato-los-sobresalientes-de-los-que-llegan-a-selectividad-se-doblan-en-seis-anos.htmlSaludos.
Cita de: wanderer en Julio 10, 2022, 13:06:42 pmCita de: Cadavre Exquis en Julio 10, 2022, 12:38:52 pmhttps://elpais.com/educacion/secundaria-bachillerato-fp/2022-07-10/la-escalada-vertiginosa-de-notas-en-bachillerato-los-sobresalientes-de-los-que-llegan-a-selectividad-se-doblan-en-seis-anos.htmlSaludos.Como ya he comentado varias veces, yo soy profesor de Matemáticas en una universidad, y el que las notas de corte en la EBAU para el grado en Matemáticas (y también en Física y otros de ciencias) haya subido tanto, está directamente relacionado con todo eso.Y esas notas si bien no son totalmente falsas, son irreales, y en cualquier caso, ocultan que los nuevos estudiantes aún no se han enfrentado a algo que ponga a prueba su capacidad de frustración, lo cual es de importancia fundamental para el desarrollo de la persona.Muevo mejor aquí la cita. Tengo curiosidad por una cosa, aunque probablemente no tienes tanta experiencia como para contrastar .Debió de haber un bajón tremendo con la LOGSE. Yo me libré por unos pocos años, pero pude ver algo del destrozo. En segundo de BUP ya se daba toda la trigonometría y las primeras integrales. Cuando llegábamos a selectividad ya llevábamos tres cursos haciendo integrales. Y aunque eso no nos evitó la bofetada del primer año de universidad, al menos teníamos base suficiente.No me imagino cómo estará ahora, que sólo se hacen integrales un curso. Y sí, creo también que cosas como pelearse con una integral no inmediata es básico. Por no hablar de los exámenes de selectividad de ahora, los de matemáticas me parecen bastante asequibles.
Muevo mejor aquí la cita. Tengo curiosidad por una cosa, aunque probablemente no tienes tanta experiencia como para contrastar .Debió de haber un bajón tremendo con la LOGSE. Yo me libré por unos pocos años, pero pude ver algo del destrozo. En segundo de BUP ya se daba toda la trigonometría y las primeras integrales. Cuando llegábamos a selectividad ya llevábamos tres cursos haciendo integrales. Y aunque eso no nos evitó la bofetada del primer año de universidad, al menos teníamos base suficiente.No me imagino cómo estará ahora, que sólo se hacen integrales un curso. Y sí, creo también que cosas como pelearse con una integral no inmediata es básico. Por no hablar de los exámenes de selectividad de ahora, los de matemáticas me parecen bastante asequibles.
Introducing The World’s Largest Open Multilingual Language Model: BLOOMLarge language models (LLMs) have made a significant impact on AI research. These powerful, general models can take on a wide variety of new language tasks from a user’s instructions. However, academia, nonprofits and smaller companies' research labs find it difficult to create, study, or even use LLMs as only a few industrial labs with the necessary resources and exclusive rights can fully access them. Today, we release BLOOM, the first multilingual LLM trained in complete transparency, to change this status quo — the result of the largest collaboration of AI researchers ever involved in a single research project.With its 176 billion parameters, BLOOM is able to generate text in 46 natural languages and 13 programming languages. For almost all of them, such as Spanish, French and Arabic, BLOOM will be the first language model with over 100B parameters ever created. This is the culmination of a year of work involving over 1000 researchers from 70+ countries and 250+ institutions, leading to a final run of 117 days (March 11 - July 6) training the BLOOM model on the Jean Zay supercomputer in the south of Paris, France thanks to a compute grant worth an estimated €3M from French research agencies CNRS and GENCI.Researchers can now download, run and study BLOOM to investigate the performance and behavior of recently developed large language models down to their deepest internal operations. More generally, any individual or institution who agrees to the terms of the model’s Responsible AI License (developed during the BigScience project itself) can use and build upon the model on a local machine or on a cloud provider - since it's embedded in the Hugging Face ecosystem, it's as easy as importing it with transformers and running it with accelerate. In this spirit of collaboration and continuous improvement, we’re also releasing, for the first time, the intermediary checkpoints and optimizer states of the training. Don’t have 8 A100s to play with? We're finalizing an inference API for large-scale use even without dedicated hardware or engineering. In the meantime, for quick tests, prototyping, and lower-scale use, you can already play with an early version on the HF hub.This is only the beginning. BLOOM’s capabilities will continue to improve as the workshop continues to experiment and tinker with the model. We’ve started work to make it as instructable as our earlier effort T0++ was and are slated to add more languages, compress the model into a more usable version with the same level of performance, and use it as a starting point for more complex architectures… All of the experiments researchers and practitioners have always wanted to run, starting with the power of a 100+ billion parameter model, are now possible. BLOOM is the seed of a living family of models that we intend to grow, not just a one-and-done model, and we’re ready to support community efforts to expand it.
Cita de: Benzino Napaloni en Julio 16, 2022, 20:32:38 pmMuevo mejor aquí la cita. Tengo curiosidad por una cosa, aunque probablemente no tienes tanta experiencia como para contrastar .Debió de haber un bajón tremendo con la LOGSE. Yo me libré por unos pocos años, pero pude ver algo del destrozo. En segundo de BUP ya se daba toda la trigonometría y las primeras integrales. Cuando llegábamos a selectividad ya llevábamos tres cursos haciendo integrales. Y aunque eso no nos evitó la bofetada del primer año de universidad, al menos teníamos base suficiente.No me imagino cómo estará ahora, que sólo se hacen integrales un curso. Y sí, creo también que cosas como pelearse con una integral no inmediata es básico. Por no hablar de los exámenes de selectividad de ahora, los de matemáticas me parecen bastante asequibles.Buenas. Me ha llamado mucho la atención lo que cuentas, supongo que será porque los planes de la CCAA donde tú estudiaste eran muy diferentes a la mía, pero yo las integrales las vi en COU y no antes. En 2º BUP es que no dimos ni derivadas. Sí dimos bastante trigonometría y muchos, muchos límites (entonces nos explicaban la definición "estricta" de "para todo epsilon > 0 existe un delta tal que....", cosa que en la ESO creo que ni la han olido), con todas sus indeterminaciones 0/0, inf/inf, 1^inf, etc etc. En 3º BUP sí nos hinchamos de derivadas, y en COU tuvimos el atracón de integrales con todos sus cambios de variable y demás. Y la verdad es que creo que salimos de ahí con un nivel bastante decente, en mis apuntes de Matemáticas de COU creo que venía todo el cálculo de una variable que necesité en primero de carrera. Aunque yo creo que la asignatura STEM que más se ha resentido con la Logse ha sido la Física, que en COU era quizá la más temible y en la Logse es una más (yo me he preparado exámenes de física general de 1º de carrera con mis apuntes de COU).Por otra parte, aunque esto es ya algo más subjetivo, recuerdo que la madurez que teníamos en COU no tenía ya nada que ver con la de la gente del actual 2º Bachillerato, y aun así en 1º de carrera te llevabas el guantazo de realidad de hacer exámenes de 4 horas estando acostumbrado a los examencitos de 1 o 1.5 horas. Me puedo imaginar que ahora lleguen a la universidad y no vean por dónde les vienen los guantazos.
Cita de: Mad Men en Febrero 17, 2022, 09:17:28 amSí y no.Estos sistemas aprenden a base de experiencia de los usuarios, es decir, recabando información. Cuando más se use mejor se moverá, aunque si quieres ir a la finca rústica pues el aparato no sabe por dónde tirar, porque no tiene memorizado el camino de tierra. Pero ese no es el problema. Los problemas no vienen tanto de no identificar el entorno (sobre todo la parte fija) sino de "encontrar el camino". Lo cual es un poco sorprendente porque parecería que el algoritmo de "encontrar el camino" estaría totalmente dominado.Podemos ver cómo un Model3 identifica perfectamente una rotonda (o eso parece según lo que se ve en la pantalla) y de repente va e intenta cruzar la isla central por el centro. Identifica claramente cómo es una calle e identifica claramente a una furgo aparcada a un lado y aun coche aparcado al otro y se empana porque no sabe cómo hacer el pequeño slalom para sortear los dos vehículos.En otro caso vemos como identifica claramente un semáforo en rojo y se lo salta.Da la impresión de que el sistema identifica sin problemas los entornos fijos y los objetos grandes estáticos y otros vehículos pero falla al decidir qué hacer.Efectivamente, aprender como es el entorno fijo es lo más fácil. Si esta calle tiene dos carriles o uno o si en esa señal de límite de velocidad pone 25 o 28. Pero insisto, al girar en una esquina va y decide irse por el carril del tranvía. O en una carretera totalmente normal va y ocupa el carril de giro a la izquierda en lugar de mantenerse por el carril principal.Parece que el problema es encontrar el camino y no el no ver lo que está pasando. LLega a un cruce de dos calles que tiene 4 pasos de cebra uno es cruzado por un par de peatones, en otro hay uno, otros peatones parece que van por la acera acercándose a otro y una bici adelanta alegremente y se pone a circular por la intersección, el coche ve todo eso (que en realidad es muy poco para un conductor humano) y se queda empanado como si fuese un octogenario confundido.Comerse un bolardo puese ser cuestión de no saber que estaba ahí y si de verdad se lo aprende después de chocar, podría ser que el siguiente Tesla lo evitase. Pero el 90% de los fallos que se ven en los vídeos no parecen derivados de no conocer el entorno.En una ocasión sencillamente no ve a un peatón que va a empezar a cruzar un paso de cebra, el conductor se tiene que disculpar. En otro caso se empeña en incorporarse a una vía en la que hay una continuidad de vehículos de todo tipo circulando en ambos sentidos.En otro vídeo se pone a circular a 54 millas por hora en una calle limitada a 25. El conductor vuelve a pasar otro día o quizás a otra hora y el coche vuelve a hacer lo mismo.
Sí y no.Estos sistemas aprenden a base de experiencia de los usuarios, es decir, recabando información. Cuando más se use mejor se moverá, aunque si quieres ir a la finca rústica pues el aparato no sabe por dónde tirar, porque no tiene memorizado el camino de tierra.
Cita de: sudden and sharp en Febrero 17, 2022, 17:01:09 pmCita de: Saturio en Febrero 17, 2022, 16:14:20 pm[...] En otro vídeo se pone a circular a 54 millas por hora en una calle limitada a 25. El conductor vuelve a pasar otro día o quizás a otra hora y el coche vuelve a hacer lo mismo.Ni idea del asunto... pero, aparentemente, no es un problema de toma de decisiones; sino más bien, de cómo marcar según que decisión como incorrecta. Tal vez, si esos otros vehiculos, con el claxon, digamos, le hicieran ver que no, que no se empeñe... que no fue buena idea... que se lo "grabe" en la AI, que no... No sé, un poco como las AIs que juegan al ajedrez. Mate, chata... prueba otra cosa la "proxima " vez.Ir mejorando... vaya. (Tal vez sea mucho pedir... pero un disco duro grande, la nube... qué se yo...)[ Por incordiar un poco, sorry. ]Como aprenda de lo que hacemos los humanos mal vamos. Acabaría tocando el claxon cuando gana el Real Madrid, picándose con los malotes... Ya pasó hace años con un bot que se volvió racista y faltón en un foro
Cita de: Saturio en Febrero 17, 2022, 16:14:20 pm[...] En otro vídeo se pone a circular a 54 millas por hora en una calle limitada a 25. El conductor vuelve a pasar otro día o quizás a otra hora y el coche vuelve a hacer lo mismo.Ni idea del asunto... pero, aparentemente, no es un problema de toma de decisiones; sino más bien, de cómo marcar según que decisión como incorrecta. Tal vez, si esos otros vehiculos, con el claxon, digamos, le hicieran ver que no, que no se empeñe... que no fue buena idea... que se lo "grabe" en la AI, que no... No sé, un poco como las AIs que juegan al ajedrez. Mate, chata... prueba otra cosa la "proxima " vez.Ir mejorando... vaya. (Tal vez sea mucho pedir... pero un disco duro grande, la nube... qué se yo...)[ Por incordiar un poco, sorry. ]
[...] En otro vídeo se pone a circular a 54 millas por hora en una calle limitada a 25. El conductor vuelve a pasar otro día o quizás a otra hora y el coche vuelve a hacer lo mismo.
La moraleja de esto, a mi parecer, es que hay muchas cosas que cambiar en el funcionamiento de la ciencia actual, para empezar el modelo de publicación por completo y en especial el filtro de la revisión por pares.