OpenAI’s large language models (sometimes referred to as GPT’s) process text using tokens, which are common sequences of characters found in a set of text. The models learn to understand the statistical relationships between these tokens, and excel at producing the next token in a sequence of tokens.
You can think of tokens as the “letters” that make up the “words” and “sentences” that AI systems use to communicate.
A helpful rule of thumb is that one token generally corresponds to ~4 characters of text for common English text. This translates to roughly ¾ of a word (so 100 tokens ~= 75 words).
The process of breaking text down into tokens is called tokenization. This allows the AI to analyze and “digest” human language into a form it can understand. Tokens become the data used to train, improve, and run the AI systems.
Why Do Tokens Matter? There are two main reasons tokens are important to understand:
Strategies for Managing Tokens
Because tokens are central to how LLMs work, it’s important to learn strategies to make the most of them:
We’ve benchmarked Stable Diffusion, a popular AI image generator, on the 45 of the latest Nvidia, AMD, and Intel GPUs to see how they stack up. We’ve been poking at Stable Diffusion for over a year now, and while earlier iterations were more difficult to get running — never mind running well — things have improved substantially. Not all AI projects have received the same level of effort as Stable Diffusion, but this should at least provide a fairly insightful look at what the various GPU architectures can manage with AI workloads given proper tuning and effort.
The easiest way to get Stable Diffusion running is via the Automatic1111 webui project. Except, that’s not the full story. Getting things to run on Nvidia GPUs is as simple as downloading, extracting, and running the contents of a single Zip file. But there are still additional steps required to extract improved performance, using the latest TensorRT extensions. Instructions are at that link, and we’ve previous tested Stable Diffusion TensorRT performance against the base model without tuning if you want to see how things have improved over time. Now we’re adding results from all the RTX GPUs, from the RTX 2060 all the way up to the RTX 4090, using the TensorRT optimizations.
For AMD and Intel GPUs, there are forks of the A1111 webui available that focus on DirectML and OpenVINO, respectively. We used these webui OpenVINO instructions to get Arc GPUs running, and these webui DirectML instructions for AMD GPUs. Our understanding, incidentally, is that all three companies have worked with the community in order to tune and improve performance and features.
Physics has empiricism. If your physical theory doesn’t make a testable prediction, physicists will make fun of you. Those that do make a prediction are tested and adopted or refuted based on the evidence. Physics is trying to describe things that exist in the physical universe, so physicists have the luxury of just looking at stuff and seeing how it behaves.
Mathematics has rigor. If your mathematical claim can’t be broken down into the language of first order logic or a similar system with clearly defined axioms, mathematicians will make fun of you. Those that can be broken down into their fundamentals are then verified step by step, with no opportunity for sloppy thinking to creep in. Mathematics deals with ontologically simple entities, so it has no need to rely on human intuition or fuzzy high-level concepts in language.
Philosophy has neither of these advantages. That doesn’t mean it’s unimportant; on the contrary, philosophy is what created science in the first place! But without any way of grounding itself in reality, it’s easy for an unscrupulous philosopher to go off the rails. As a result, much of philosophy ends up being people finding justifications for what they already wanted to believe anyway, rather than any serious attempt to derive new knowledge from first principles. (Notice how broad the spread of disagreement is among philosophers on basically every aspect of their field, compared to mathematicians and physicists.)
This is not a big deal when philosophy is a purely academic exercise, but it becomes a problem when people are turning to philosophers for practical advice. In the field of artificial intelligence, things are moving quickly, and people want guidance about what’s to come. Should we consider AI to be a moral patient? Does moral realism imply that advanced AI will automatically prioritize humans’ best interests, or does the is-ought problem prevent that? What do concepts like «intelligence» and «values» actually mean?
Lanzada con poco bombo durante los primeros días de la pandemia, la serie de ocho capítulos DEVS es una de las ficciones de los últimos años con más posibilidades de convertirse en un futuro clásico de culto. Ambientada en un Silicon Valley crepuscular profundamente melancólico, esta historia del cineasta británico Alex Garland no trata directamente sobre la inteligencia artificial: su argumento traza una fábula sobre la computación cuántica, el destino frente al libre albedrío, y la posibilidad de reconstruir cada momento único de la experiencia humana. A Jorge Luis Borges probablemente le habría entusiasmado.
La historia se ha contado mil veces. Si tuviésemos que explicar los orígenes del ideario intelectual de la industria tecnológica –de lo que Richard Barbrook y Andy Cameron llamaron “la ideología californiana”– sus componentes fundamentales son el improbable encuentro hace seis décadas al sur de San Francisco entre hippies e ingenieros informáticos; entre una visión tecnocrática heredada del complejo industrial-militar de la guerra fría, y los deseos de emancipación colectiva y liberación de la consciencia de la contracultura. El legendario Whole Earth Catalog de Stewart Brand (la publicación seminal de la cultura digital), las propuestas del visionario arquitecto Buckminster Fuller, los experimentos de convivencia planteados en comunas como Drop City… fueron caldo de cultivo para emprendedores que como Steve Jobs imaginaron un futuro cercano en que el PC era tanto un acelerador de la eficiencia como una herramienta para la realización personal y la autonomía creativa. Una prótesis intelectual, una “bicicleta de la mente” que nos permitiría llegar a donde no seríamos capaces como especie exclusivamente biológica.
La industria tecnológica se sitúa hoy en su momento más existencial desde al menos los años 90, con la emergencia de la Internet comercial. El movimiento pro ética de la IA cree que los posibles riesgos del Deep Learning y las redes neuronales requieren de un desarrollo controlado y cuidadoso que permita su introducción paulatina en todos los aspectos de la vida cotidiana. Los aceleracionistas defienden que estos miedos son conservadores y que el inevitable desarrollo de la IA traerá consigo una nueva era de prosperidad humana y crecimiento, soluciones al cambio climático y a enfermedades incurables.
…antes que una herramienta de trascendencia espiritual la IA será otro sistema de concentración de poder en un mundo en desigualdad creciente, si no cambiamos algunas de sus reglas fundamentales.
Las sociedades humanas no están maduras para asimilar una tecnología como la IA generativa en el contexto de las relaciones personales, simplemente porque no hemos pasado por una etapa de educación que permita a las personas entender de verdad con qué están conversando. Por lo general, el ser humano tiende a otorgar una cierta «autoridad» al algoritmo, le adscribe una supuesta capacidad de consulta y síntesis de información prácticamente ilimitada, y tiende a prácticamente subcontratar su pensamiento crítico a las respuestas a las que accede a través de medios tecnológicos. El desconocimiento de la tecnología, como bien decía Arthur C. Clarke, hace que se convierta en indistinguible de la magia. Y algo así, sin duda, puede tener efectos enormemente nocivos en las sociedades humanas: desde trastornos de percepción de la realidad hasta auténticas alienaciones y problemas psicológicos.
https://twitter.com/JaimeObregon/status/1645138600537858048?s=20
Whatever your thoughts on AI bots, you may want to take action on your own website to block ChatGPT from crawling, indexing, and using your website content and data.
Fashion brand Levi Strauss & Co has announced a partnership with digital fashion studio Lalaland.ai to make custom artificial intelligence (AI) generated avatars in what it says will increase diversity among its models.
WordPress plugin developers are adopting AI-powered tech and building it into their products, such as RankMath’s AI-generated suggestions for creating SEO-friendly content, WordPress.com’s experimental blocks for AI-generated images and content, and a Setary’s plugin that uses AI to write and bulk edit WooCommerce product descriptions. The wpfrontpage site is tracking these plugins but WordPress.org also lists dozens of plugins with AI, many of them created to write content or generate images.
Get an abbreviated description made by an AI of you based on your latest tweets. What do other users perceive from what you tweet?
YOLOv4 installation has for a while been very tricky to install…until today. I will show you how to install YOLOv4 TensorFlow running on video in under 5 minutes.
You can run this either on CPU or CUDA Supported GPU (Nvidia Only). I achieved 3 FPS on CPU and 30 FPS on GPU (1080Ti)
Feministai.net is an ongoing effort, work in progress debate that seeks to contribute to the development of a feminist framework to question algorithmic decisions making systems that are being deployed by the public sector. Our ultimate goal is to develop arguments and build bridges for advocacy with different human rights groups, from women’s rights defenders to LGBTIQ + groups, especially in Latin America to address the trend in which governments are adopting A.I. systems to deploy social policies, not only without considering human rights implications but also in disregard of oppressive power relations that might be amplified through a false sense of neutrality brought by automation. Automation of the status quo, pertained by inequalities and discrimination.
The current debate of A.I. principles and frameworks is mostly focused on “how to fix it?”, instead of to “why we actually need it?” and “for whose benefit”. Therefore, the first tool of our toolkit to question A.I. systems is the scheme of Oppressive A.I. that we drafted based on both, empirical analysis of cases from Latin America and bibliographic review of critical literature. Is a particular A.I system based on surveilling the poor? Is it automating neoliberal policies? Is it based on precarious labor and colonial extractivism of data bodies and resources from our territories? Who develops it is part of the group targeted by it or its likely to restate structural inequalities of race, gender, sexuality? Can the wider community have enough transparency to check by themselves the accuracy in the answers to the previous questions?
What is a feminist approach to consent? How can it be applied to technologies? Those simple questions were able to shed light on how limited is the individualistic notion of consent proposed in data protection frameworks.
Miquela nace en 2016. Un perfil de Instagram daba comienzo a su historia: una joven hispano-brasileña, residiendo en Los Ángeles y proyectando identidad de IT-Girl comenzaba su rastro digital suscitando todo tipo de especulaciones (como que era una campaña para promocionar el juego Los Sims). Después de tres años ya sabemos un poco de qué va la historia: «un estudio transmedia que crea universos narrativos y personajes digitales». Esto es lo que puede leerse en la escueta web (es un Google Doc en realidad) de presentación de Brud.
The Amazon Echo as an anatomical map of human labor, data and planetary resources
Yet in the twenty-first century, power will be determined not by one’s nuclear arsenal, but by a wider spectrum of technological capabilities based on digitization. Those who aren’t at the forefront of artificial intelligence (AI) and Big Data will inexorably become dependent on, and ultimately controlled by, other powers. Data and technological sovereignty, not nuclear warheads, will determine the global distribution of power and wealth in this century. And in open societies, the same factors will also decide the future of democracy.
The most important issue facing the new European Commission, then, is Europe’s lack of digital sovereignty. Europe’s command of AI, Big Data, and related technologies will determine its overall competitiveness in the twenty-first century. But Europeans must decide who will own the data needed to achieve digital sovereignty, and what conditions should govern its collection and use.
In the broadest sense, AI refers to machines that can learn, reason, and act for themselves. They can make their own decisions when faced with new situations, in the same way that humans and animals can.
As it currently stands, the vast majority of the AI advancements and applications you hear about refer to a category of algorithms known as machine learning. These algorithms use statistics to find patterns in massive amounts of data. They then use those patterns to make predictions on things like what shows you might like on Netflix, what you’re saying when you speak to Alexa, or whether you have cancer based on your MRI.
Google is reportedly working on an A.I.-based health and wellness coach.
Thanks to its spectrum of hardware products, Google would have a notable advantage over existing wellness coaching apps. While its coach, as reported, would primarily exist on smartwatches to start, Android Police noted that the company could include a smartphone counterpart as well. The company could also eventually spread it to Google Home or Android TV. The latter is unchartered territory for these kinds of apps, which are typically limited to smartphones and wearables. With availability in the home, lifestyle coaching recommendations could become increasingly contextual and less obtrusive. If you ask for a chicken parmesan dinner recipe, it could offer a healthier alternative instead; or if you’re streaming music at 10 p.m. and have set a goal to get more sleep, perhaps it could interrupt your music playback to remind you start getting ready for bed. A smartwatch or phone could do this too, of course, but by linking up its product ecosystem, Google could deliver helpful notifications in the context that makes the most sense.
