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 1 Visitante están viendo este tema.
Elon Musk Announces xAI With Goal To Understand 'True Nature of the Universe'Posted by msmash on Wednesday July 12, 2023 @01:31PM from the how-about-that dept.Elon Musk announced the formation of what he's calling xAI, whose goal is to "understand the true nature of the universe." The team at xAI, led by Musk, includes individuals who have previously worked at DeepMind, OpenAI, Google Research, Microsoft Research, Tesla, and the University of Toronto."Collectively we contributed some of the most widely used methods in the field, in particular the Adam optimizer, Batch Normalization, Layer Normalization, and the discovery of adversarial examples. We further introduced innovative techniques and analyses such as Transformer-XL, Autoformalization, the Memorizing Transformer, Batch Size Scaling, and uTransfer. We have worked on and led the development of some of the largest breakthroughs in the field including AlphaStar, AlphaCode, Inception, Minerva, GPT-3.5, and GPT-4," xAI said in a blog post.
CitarElon Musk Announces xAI With Goal To Understand 'True Nature of the Universe'Posted by msmash on Wednesday July 12, 2023 @01:31PM from the how-about-that dept.Elon Musk announced the formation of what he's calling xAI, whose goal is to "understand the true nature of the universe." The team at xAI, led by Musk, includes individuals who have previously worked at DeepMind, OpenAI, Google Research, Microsoft Research, Tesla, and the University of Toronto."Collectively we contributed some of the most widely used methods in the field, in particular the Adam optimizer, Batch Normalization, Layer Normalization, and the discovery of adversarial examples. We further introduced innovative techniques and analyses such as Transformer-XL, Autoformalization, the Memorizing Transformer, Batch Size Scaling, and uTransfer. We have worked on and led the development of some of the largest breakthroughs in the field including AlphaStar, AlphaCode, Inception, Minerva, GPT-3.5, and GPT-4," xAI said in a blog post.Saludos.
Cita de: Cadavre Exquis en Julio 01, 2023, 12:23:46 pmCitarChinese Researchers Used AI To Design RISC-V CPU In Under 5 HoursPosted by BeauHD on Friday June 30, 2023 @08:45PM from the performance-is-nothing-to-brag-about-though dept.Required Snark shares a report from Tom's Hardware:CitarA group of Chinese scientists has published (PDF) a paper titled "Pushing the Limits of Machine Design: Automated CPU Design with AI." The paper details the researchers' work in designing a new industrial-scale RISC-V CPU in under 5 hours. It is claimed this AI-automated feat was about 1000x faster than a human team could have finished a comparable CPU design. However, some may poke fun at the resulting AI-designed CPU performing approximately on par with an i486.Training consisted of observing a series of CPU inputs and outputs. The scientists generated a Binary Speculation Diagram (BSD) from this I/O and leveraged principles of Monte Carlo-based expansion and Boolean functions to hone the accuracy and efficiency of the AI-based CPU design. Thus the CPU design was formed "from only external input-output observations instead of formal program code," explains the scientists. It also boasted an impressive 99.99999999999% accuracy. Using the above-outlined process, an automated AI design of a CPU was created.The taped-out RISC-V32IA instruction set CPU was fabricated at 65nm and could run at up to 300 MHz. Running the Linux (kernel 5.15) operating system and SPEC CINT 2000 on the AI-generated CPU validated its functionality. In Drystone benchmarks, the AI-generated CPU performed on par with an i486. Interestingly, it appears to be a little bit faster than an Acorn Archimedes A3010 in the same test. Though some might be unimpressed by the performance of the AI-generated CPU, the scientists also seem quite proud that their generated BSD "discovered the von Neumann architecture from scratch."Saludos.Aunque lo que han hecho es interesante, es más interesante aún las enormes diferencias entre lo que esconde el titular (podemos hacer chips con una IA mil veces más rápido) y lo que luego se desvela en los detalles (el resultado es de una potencia de hace 30 años y el diseño es regulero).Y esta tendencia crece y crece, mientras chistes (me niego a llamarlo medio) como Twitter incentivan, por su propia y absurda estructura, el quedarse con la información superficial y eliminar todo detalle y matización de cualquier asunto. Este asunto de la IA es el ejemplo perfecto. Mucha gente hablando de chorradas sensacionalistas sin saber siquiera lo más fundamental (y atribuyendo propiedades mágicas a dicha tecnología por pura ignorancia) y unos pocos que hablan con fundamento, ahogados entre todo el ruido.
CitarChinese Researchers Used AI To Design RISC-V CPU In Under 5 HoursPosted by BeauHD on Friday June 30, 2023 @08:45PM from the performance-is-nothing-to-brag-about-though dept.Required Snark shares a report from Tom's Hardware:CitarA group of Chinese scientists has published (PDF) a paper titled "Pushing the Limits of Machine Design: Automated CPU Design with AI." The paper details the researchers' work in designing a new industrial-scale RISC-V CPU in under 5 hours. It is claimed this AI-automated feat was about 1000x faster than a human team could have finished a comparable CPU design. However, some may poke fun at the resulting AI-designed CPU performing approximately on par with an i486.Training consisted of observing a series of CPU inputs and outputs. The scientists generated a Binary Speculation Diagram (BSD) from this I/O and leveraged principles of Monte Carlo-based expansion and Boolean functions to hone the accuracy and efficiency of the AI-based CPU design. Thus the CPU design was formed "from only external input-output observations instead of formal program code," explains the scientists. It also boasted an impressive 99.99999999999% accuracy. Using the above-outlined process, an automated AI design of a CPU was created.The taped-out RISC-V32IA instruction set CPU was fabricated at 65nm and could run at up to 300 MHz. Running the Linux (kernel 5.15) operating system and SPEC CINT 2000 on the AI-generated CPU validated its functionality. In Drystone benchmarks, the AI-generated CPU performed on par with an i486. Interestingly, it appears to be a little bit faster than an Acorn Archimedes A3010 in the same test. Though some might be unimpressed by the performance of the AI-generated CPU, the scientists also seem quite proud that their generated BSD "discovered the von Neumann architecture from scratch."Saludos.
Chinese Researchers Used AI To Design RISC-V CPU In Under 5 HoursPosted by BeauHD on Friday June 30, 2023 @08:45PM from the performance-is-nothing-to-brag-about-though dept.Required Snark shares a report from Tom's Hardware:CitarA group of Chinese scientists has published (PDF) a paper titled "Pushing the Limits of Machine Design: Automated CPU Design with AI." The paper details the researchers' work in designing a new industrial-scale RISC-V CPU in under 5 hours. It is claimed this AI-automated feat was about 1000x faster than a human team could have finished a comparable CPU design. However, some may poke fun at the resulting AI-designed CPU performing approximately on par with an i486.Training consisted of observing a series of CPU inputs and outputs. The scientists generated a Binary Speculation Diagram (BSD) from this I/O and leveraged principles of Monte Carlo-based expansion and Boolean functions to hone the accuracy and efficiency of the AI-based CPU design. Thus the CPU design was formed "from only external input-output observations instead of formal program code," explains the scientists. It also boasted an impressive 99.99999999999% accuracy. Using the above-outlined process, an automated AI design of a CPU was created.The taped-out RISC-V32IA instruction set CPU was fabricated at 65nm and could run at up to 300 MHz. Running the Linux (kernel 5.15) operating system and SPEC CINT 2000 on the AI-generated CPU validated its functionality. In Drystone benchmarks, the AI-generated CPU performed on par with an i486. Interestingly, it appears to be a little bit faster than an Acorn Archimedes A3010 in the same test. Though some might be unimpressed by the performance of the AI-generated CPU, the scientists also seem quite proud that their generated BSD "discovered the von Neumann architecture from scratch."
A group of Chinese scientists has published (PDF) a paper titled "Pushing the Limits of Machine Design: Automated CPU Design with AI." The paper details the researchers' work in designing a new industrial-scale RISC-V CPU in under 5 hours. It is claimed this AI-automated feat was about 1000x faster than a human team could have finished a comparable CPU design. However, some may poke fun at the resulting AI-designed CPU performing approximately on par with an i486.Training consisted of observing a series of CPU inputs and outputs. The scientists generated a Binary Speculation Diagram (BSD) from this I/O and leveraged principles of Monte Carlo-based expansion and Boolean functions to hone the accuracy and efficiency of the AI-based CPU design. Thus the CPU design was formed "from only external input-output observations instead of formal program code," explains the scientists. It also boasted an impressive 99.99999999999% accuracy. Using the above-outlined process, an automated AI design of a CPU was created.The taped-out RISC-V32IA instruction set CPU was fabricated at 65nm and could run at up to 300 MHz. Running the Linux (kernel 5.15) operating system and SPEC CINT 2000 on the AI-generated CPU validated its functionality. In Drystone benchmarks, the AI-generated CPU performed on par with an i486. Interestingly, it appears to be a little bit faster than an Acorn Archimedes A3010 in the same test. Though some might be unimpressed by the performance of the AI-generated CPU, the scientists also seem quite proud that their generated BSD "discovered the von Neumann architecture from scratch."
El tráfico a nivel mundial de la plataforma de OpenAI había experimentado un constante crecimiento desde su lanzamiento hasta el pasado junio, que ha registrado un descenso del 9,7%.ChatGPT, el 'chatbot' impulsado por Inteligencia Artificial (IA) desarrollado por OpenAI, ha perdido un 9,7 por ciento de tráfico a nivel global durante el mes de junio, el primer descenso registrado desde su lanzamiento, que se produjo en enero de este año.El 'chatbot' de OpenAI se presentó como un servicio entrenado con un modelo de lenguaje capaz de procesar y generar contenidos para mantener una conversación de texto, con la habilidad de enlazar ideas y de recordar conversaciones previas.La compañía dirigida por Sam Altman continúa implementando mejoras y nuevas capacidades para su 'chatbot', como la posibilidad de utilizar su nueva generación de modelo de lenguaje GPT-4, que presentó en marzo, para todos los desarrolladores de la API. Este nuevo modelo integra capacidad para resolver grandes problemas con precisión y ofrecer respuestas más útiles y seguras.Asimismo, ChatGPT también se ha expandido incorporándose en productos y servicios de otras marcas, como es el caso de las nuevas versiones de Bing y Edge, desarrolladas por Microsoft, que fueron presentadas en febrero.Debido a esta incorporación, el 'chatbot' ha ido experimentando un alto crecimiento exponencial tanto de tráfico, como de usuarios que descargan su 'app' o lo utilizan a través de la web.Sin embargo, durante este mes de junio, ChatGPT ha experimentado por primera vez un descenso del tráfico de un 9,7 por ciento, así como un 5,7 por ciento menos de visitantes únicos, tal y como ha registrado la compañía especializada en el análisis web Similarweb en un comunicado en su blog.Esta caída de tráfico hace referencia al tráfico web móvil y de escritorio a nivel mundial. Por otra parte, además de una disminución del número de visitantes únicos, también se ha registrado una bajada del 8,5 por ciento en la cantidad de tiempo que los visitantes pasan dentro del sitio web.Con todo ello, los resultados sugieren que ha descendido el interés por parte de los usuarios hacia ChatGPT en este último mes, en comparación a cómo ha estado creciendo su uso en los meses tras su lanzamiento. No obstante, los datos recabados por Similarweb también indican que ChatGPT sigue atrayendo "más visitantes de todo el mundo" que otros servicios. Entre ellos, Bing.Igualmente, a pesar de su descenso, ChatGPT continúa superando al segundo servicio independiente más popular de 'chatbots' de IA como es Character.AI, cuyo tráfico también ha disminuido un 32 por ciento "mes a mes", según esta compañía.
El largo camino de la inteligencia artificial y la medicinaJulián Estévez 26 de julio de 2023FuenteHace unos días se publicaba la noticia de que la tecnología de moda últimamente, la célebre inteligencia artificial generativa, estaba siendo usada en la no menos famosa Clínica Mayo de EEUU para mejorar los tratamientos médicos a los enfermos.Med-PaLM 2 de Google, una herramienta de inteligencia artificial diseñada para responder preguntas sobre información médica, ha estado en pruebas en el hospital de investigación de la Clínica Mayo desde abril, según informó The Wall Street Journal. PaLM 2 es el modelo de lenguaje en el que se basa Bard de Google, y por lo tanto, Med-PaLM 2 es una variante especializada en medicina.Google cree que su modelo actualizado puede ser particularmente útil en países con "acceso más limitado a los médicos", y que su variante médica puede ayudar a los facultativos mucho más que chatbots genéricos con el de Bing o ChatGPT.No es la primera vez que las grandes empresas tecnológicas hacen anuncios de este tipo, y tratan de captar el mercado de la medicina. Por ejemplo, el caído IBM de Watson aspiraba a eso precisamente, el famoso ordenador que ganó en Jeopardy, y cuya aplicación en la medicina fue un absoluto fracaso.En esta ocasión, Google no está inflando tanto la burbuja de las promesas, y aspiran a que los doctores puedan emplear la IA para mejorar su búsqueda de información y completar más rápido el trabajo administrativo.Para evaluar la bondad del software Med-PaLM 2, Google publicó un artículo en Nature en el que explican cómo sometieron a su inteligencia artificial al examen para lograr la licencia médica en EEUU, las cuales fueron 140 preguntas. Los resultados muestran que Med-PaLM 2 logró un 92,6% de las respuestas alineadas correctamente con el consenso médico, mientras que la versión anterior de la IA de Google, Flan-PaLM, logró un 61,9%. Eso sí, ambos siguen superados por los facultativos humanos.También es llamativo la velocidad a la que están evolucionando estos sistemas en el famoso examen médico de EEUU (USMLE):El pastel de beneficios de la IA en medicina es muy goloso, y seguro que ninguna de estas multibillonarias empresas no cejarán en su empeño fácilmente. Quizás os interese una bonita historia sobre el primer chatbot que aspiró a introducirse en los servicios médicos, el llamado Eliza, programado por Joseph Weizembaum en 1964, que aspiraba a tener conversaciones con pacientes psiquiátricos, y sobre la que hablé hace poco en el blog.Si me permitís recomendaros una lectura de verano breve sobre Eliza y otros chatbots que intentaban superar el test de Turing, os sugiero el librito The most human human, de Brian Christian. Aunque probablemente, os proponga alguna lectura de verano más en la próxima entrada.
Netflix Lists $900,000 Job Seeking AI To 'Create Great Content'Posted by msmash on Thursday July 27, 2023 @12:00AM from the closer-look dept.An anonymous reader shares a report:CitarAs Hollywood executives insist it is "just not realistic" to pay actors -- 87 percent of whom earn less than $26,000 -- more, they are spending lavishly on AI programs. While entertainment firms like Disney have declined to go into specifics about the nature of their investments in artificial intelligence, job postings and financial disclosures reviewed by The Intercept reveal new details about the extent of these companies' embrace of the technology. In one case, Netflix is offering as much as $900,000 for a single AI product manager.[...] Netflix's posting for a $900,000-a-year AI product manager job makes clear that the AI goes beyond just the algorithms that determine what shows are recommended to users. The listing points to AI's uses for content creation: "Artificial Intelligence is powering innovation in all areas of the business," including by helping them to "create great content." Netflix's AI product manager posting alludes to a sprawling effort by the business to embrace AI, referring to its "Machine Learning Platform" involving AI specialists "across Netflix."A research section on Netflix's website describes its machine learning platform, noting that while it was historically used for things like recommendations, it is now being applied to content creation. "Historically, personalization has been the most well-known area, where machine learning powers our recommendation algorithms. We're also using machine learning to help shape our catalog of movies and TV shows by learning characteristics that make content successful. We use it to optimize the production of original movies and TV shows in Netflix's rapidly growing studio."
As Hollywood executives insist it is "just not realistic" to pay actors -- 87 percent of whom earn less than $26,000 -- more, they are spending lavishly on AI programs. While entertainment firms like Disney have declined to go into specifics about the nature of their investments in artificial intelligence, job postings and financial disclosures reviewed by The Intercept reveal new details about the extent of these companies' embrace of the technology. In one case, Netflix is offering as much as $900,000 for a single AI product manager.[...] Netflix's posting for a $900,000-a-year AI product manager job makes clear that the AI goes beyond just the algorithms that determine what shows are recommended to users. The listing points to AI's uses for content creation: "Artificial Intelligence is powering innovation in all areas of the business," including by helping them to "create great content." Netflix's AI product manager posting alludes to a sprawling effort by the business to embrace AI, referring to its "Machine Learning Platform" involving AI specialists "across Netflix."A research section on Netflix's website describes its machine learning platform, noting that while it was historically used for things like recommendations, it is now being applied to content creation. "Historically, personalization has been the most well-known area, where machine learning powers our recommendation algorithms. We're also using machine learning to help shape our catalog of movies and TV shows by learning characteristics that make content successful. We use it to optimize the production of original movies and TV shows in Netflix's rapidly growing studio."
Forget Subtitles. YouTube Now Dubs (Some) Videos with AI-Generated VoicesPosted by EditorDavid on Monday July 31, 2023 @03:59AM from the lots-in-translation dept.An anonymous reader shared this report from the international tech news site Rest of World:CitarIn an open letter earlier this year, Neal Mohan, the recently appointed head of YouTube, made a pledge to creators that better translation tools were coming. Now, YouTube is delivering on that promise with Aloud — a free tool that automatically dubs videos using synthetic voices, raising creators' hopes and putting new pressure on dubbing firms that already cater to YouTubers.At the VidCon convention in late June, YouTube announced a pilot for Aloud. The tool first generates a transcription of a video's audio, which a creator can edit before selecting their preferred language and style of synthetic voice. The dub can take just minutes to generate.The pilot currently includes the option to dub videos into English, Spanish, and Portuguese. The company has said more languages are coming — likely including Bahasa Indonesia and Hindi, which are already advertised on the Aloud website. Hundreds of creators have already signed up to test the tool. "Our long-term goal is to be able to dub between any two languages, and as part of that goal we will continue to pilot and learn from dubbing content in different regions," Buddhika Kottahachchi, co-founder of Aloud and the recently appointed head of product for YouTube Dubbing, told Rest of World. "Helping a creator expand beyond their primary language can help them reach new audiences..."In the lead up to the pilot announcement, YouTube also released a new product feature that allows viewers to select between multiple dubbing tracks on a single video, similar to the current option for subtitles.Here's a video of YouTube's announcement, with five"audio tracks" (in different languages) available if you click the "gear" icon. While YouTube's top stars hire dubbing services, many smaller creators can't afford them, the article points out. "By offering Aloud for free, YouTube is setting up a new swath of creators to access dubs for the first time..."YouTube's new push into automated dubbing is a serious challenge for existing dubbing companies, which are now forced to compete with a free competitor built into the platform."
In an open letter earlier this year, Neal Mohan, the recently appointed head of YouTube, made a pledge to creators that better translation tools were coming. Now, YouTube is delivering on that promise with Aloud — a free tool that automatically dubs videos using synthetic voices, raising creators' hopes and putting new pressure on dubbing firms that already cater to YouTubers.At the VidCon convention in late June, YouTube announced a pilot for Aloud. The tool first generates a transcription of a video's audio, which a creator can edit before selecting their preferred language and style of synthetic voice. The dub can take just minutes to generate.The pilot currently includes the option to dub videos into English, Spanish, and Portuguese. The company has said more languages are coming — likely including Bahasa Indonesia and Hindi, which are already advertised on the Aloud website. Hundreds of creators have already signed up to test the tool. "Our long-term goal is to be able to dub between any two languages, and as part of that goal we will continue to pilot and learn from dubbing content in different regions," Buddhika Kottahachchi, co-founder of Aloud and the recently appointed head of product for YouTube Dubbing, told Rest of World. "Helping a creator expand beyond their primary language can help them reach new audiences..."In the lead up to the pilot announcement, YouTube also released a new product feature that allows viewers to select between multiple dubbing tracks on a single video, similar to the current option for subtitles.
— del hundimiento de Vox el 23-julio-2023...https://www.youtube.com/watch?v=6E9cgDQPSck ;
XGracia moinsdewatt