SprayPrinter is a device that attaches to aerosol paint cans whereby users can print images via Bluetooth from a smartphone onto a wall or almost any surface. == History == The technology behind SprayPrinter was developed by Mihkel Joala. He explained in a 2016 interview with New Atlas that his idea was inspired by the modern car engine and the Nintendo Wii console. "Engines nowadays use extremely fast valves to spray fuel to [the] combustion chamber," says Joala. "I realized I can use them to shoot paint with pinpoint accuracy." As of December 2021, the company appears to be no longer selling products. == Awards and Recognitions == In 2015, SprayPrinter received €8,000 from the Estonian prototyping contest Prototron for its initial prototype. In 2016, the SprayPrinter team won the grand prize of €30,000 from the televised pitching competition Ajujaht.
Contrastive Language-Image Pre-training
Contrastive Language-Image Pre-training (CLIP) is a technique for training a pair of neural network models, one for image understanding and one for text understanding, using a contrastive objective. This method has enabled broad applications across multiple domains, including cross-modal retrieval, text-to-image generation, and aesthetic ranking. == Algorithm == The CLIP method trains a pair of models contrastively. One model takes in a piece of text as input and outputs a single vector representing its semantic content. The other model takes in an image and similarly outputs a single vector representing its visual content. The models are trained so that the vectors corresponding to semantically similar text-image pairs are close together in the shared vector space, while those corresponding to dissimilar pairs are far apart. To train a pair of CLIP models, one would start by preparing a large dataset of image-caption pairs. During training, the models are presented with batches of N {\displaystyle N} image-caption pairs. Let the outputs from the text and image models be respectively v 1 , . . . , v N , w 1 , . . . , w N {\displaystyle v_{1},...,v_{N},w_{1},...,w_{N}} . Two vectors are considered "similar" if their dot product is large. The loss incurred on this batch is the multi-class N-pair loss, which is a symmetric cross-entropy loss over similarity scores: − 1 N ∑ i ln e v i ⋅ w i / T ∑ j e v i ⋅ w j / T − 1 N ∑ j ln e v j ⋅ w j / T ∑ i e v i ⋅ w j / T {\displaystyle -{\frac {1}{N}}\sum _{i}\ln {\frac {e^{v_{i}\cdot w_{i}/T}}{\sum _{j}e^{v_{i}\cdot w_{j}/T}}}-{\frac {1}{N}}\sum _{j}\ln {\frac {e^{v_{j}\cdot w_{j}/T}}{\sum _{i}e^{v_{i}\cdot w_{j}/T}}}} In essence, this loss function encourages the dot product between matching image and text vectors ( v i ⋅ w i {\displaystyle v_{i}\cdot w_{i}} ) to be high, while discouraging high dot products between non-matching pairs. The parameter T > 0 {\displaystyle T>0} is the temperature, which is parameterized in the original CLIP model as T = e − τ {\displaystyle T=e^{-\tau }} where τ ∈ R {\displaystyle \tau \in \mathbb {R} } is a learned parameter. Other loss functions are possible. For example, Sigmoid CLIP (SigLIP) proposes the following loss function: L = 1 N ∑ i , j ∈ 1 : N f ( ( 2 δ i , j − 1 ) ( e τ w i ⋅ v j + b ) ) {\displaystyle L={\frac {1}{N}}\sum _{i,j\in 1:N}f((2\delta _{i,j}-1)(e^{\tau }w_{i}\cdot v_{j}+b))} where f ( x ) = ln ( 1 + e − x ) {\displaystyle f(x)=\ln(1+e^{-x})} is the negative log sigmoid loss, and the Dirac delta symbol δ i , j {\displaystyle \delta _{i,j}} is 1 if i = j {\displaystyle i=j} else 0. == CLIP models == While the original model was developed by OpenAI, subsequent models have been trained by other organizations as well. === Image model === The image encoding models used in CLIP are typically vision transformers (ViT). The naming convention for these models often reflects the specific ViT architecture used. For instance, "ViT-L/14" means a "vision transformer large" (compared to other models in the same series) with a patch size of 14, meaning that the image is divided into 14-by-14 pixel patches before being processed by the transformer. The size indicator ranges from B, L, H, G (base, large, huge, giant), in that order. Other than ViT, the image model is typically a convolutional neural network, such as ResNet (in the original series by OpenAI), or ConvNeXt (in the OpenCLIP model series by LAION). Since the output vectors of the image model and the text model must have exactly the same length, both the image model and the text model have fixed-length vector outputs, which in the original report is called "embedding dimension". For example, in the original OpenAI model, the ResNet models have embedding dimensions ranging from 512 to 1024, and for the ViTs, from 512 to 768. Its implementation of ViT was the same as the original one, with one modification: after position embeddings are added to the initial patch embeddings, there is a LayerNorm. Its implementation of ResNet was the same as the original one, with 3 modifications: In the start of the CNN (the "stem"), they used three stacked 3x3 convolutions instead of a single 7x7 convolution, as suggested by. There is an average pooling of stride 2 at the start of each downsampling convolutional layer (they called it rect-2 blur pooling according to the terminology of ). This has the effect of blurring images before downsampling, for antialiasing. The final convolutional layer is followed by a multiheaded attention pooling. ALIGN a model with similar capabilities, trained by researchers from Google used EfficientNet, a kind of convolutional neural network. === Text model === The text encoding models used in CLIP are typically Transformers. In the original OpenAI report, they reported using a Transformer (63M-parameter, 12-layer, 512-wide, 8 attention heads) with lower-cased byte pair encoding (BPE) with 49152 vocabulary size. Context length was capped at 76 for efficiency. Like GPT, it was decoder-only, with only causally-masked self-attention. Its architecture is the same as GPT-2. Like BERT, the text sequence is bracketed by two special tokens [SOS] and [EOS] ("start of sequence" and "end of sequence"). Take the activations of the highest layer of the transformer on the [EOS], apply LayerNorm, then a final linear map. This is the text encoding of the input sequence. The final linear map has output dimension equal to the embedding dimension of whatever image encoder it is paired with. These models all had context length 77 and vocabulary size 49408. ALIGN used BERT of various sizes. == Dataset == === WebImageText === The CLIP models released by OpenAI were trained on a dataset called "WebImageText" (WIT) containing 400 million pairs of images and their corresponding captions scraped from the internet. The total number of words in this dataset is similar in scale to the WebText dataset used for training GPT-2, which contains about 40 gigabytes of text data. The dataset contains 500,000 text-queries, with up to 20,000 (image, text) pairs per query. The text-queries were generated by starting with all words occurring at least 100 times in English Wikipedia, then extended by bigrams with high mutual information, names of all Wikipedia articles above a certain search volume, and WordNet synsets. The dataset is private and has not been released to the public, and there is no further information on it. ==== Data preprocessing ==== For the CLIP image models, the input images are preprocessed by first dividing each of the R, G, B values of an image by the maximum possible value, so that these values fall between 0 and 1, then subtracting by [0.48145466, 0.4578275, 0.40821073], and dividing by [0.26862954, 0.26130258, 0.27577711]. The rationale was that these are the mean and standard deviations of the images in the WebImageText dataset, so this preprocessing step roughly whitens the image tensor. These numbers slightly differ from the standard preprocessing for ImageNet, which uses [0.485, 0.456, 0.406] and [0.229, 0.224, 0.225]. If the input image does not have the same resolution as the native resolution (224×224 for all except ViT-L/14@336px, which has 336×336 resolution), then the input image is first scaled by bicubic interpolation, so that its shorter side is the same as the native resolution, then the central square of the image is cropped out. === Others === ALIGN used over one billion image-text pairs, obtained by extracting images and their alt-tags from online crawling. The method was described as similar to how the Conceptual Captions dataset was constructed, but instead of complex filtering, they only applied a frequency-based filtering. Later models trained by other organizations had published datasets. For example, LAION trained OpenCLIP with published datasets LAION-400M, LAION-2B, and DataComp-1B. == Training == In the original OpenAI CLIP report, they reported training 5 ResNet and 3 ViT (ViT-B/32, ViT-B/16, ViT-L/14). Each was trained for 32 epochs. The largest ResNet model took 18 days to train on 592 V100 GPUs. The largest ViT model took 12 days on 256 V100 GPUs. All ViT models were trained on 224×224 image resolution. The ViT-L/14 was then boosted to 336×336 resolution by FixRes, resulting in a model. They found this was the best-performing model. In the OpenCLIP series, the ViT-L/14 model was trained on 384 A100 GPUs on the LAION-2B dataset, for 160 epochs for a total of 32B samples seen. == Applications == === Cross-modal retrieval === CLIP's cross-modal retrieval enables the alignment of visual and textual data in a shared latent space, allowing users to retrieve images based on text descriptions and vice versa, without the need for explicit image annotations. In text-to-image retrieval, users input descriptive text, and CLIP retrieves images with matching embeddings. In image-to-text retrieval, images are used to find related text content. CLIP’s ability to connect vis
Amino (app)
Amino was a social media application originally developed by Narvii, Inc. It was originally created by Yin Wang and Ben Anderson in 2010, and then launched as an app in 2012. Amino was acquired by MediaLab AI Inc in January 2021, and the founders are no longer associated with the application. The platform ceased all operations in December 2025. == History == In 2010, Wang and Anderson came up with the idea for a convention-like community while attending an anime convention in Boston, Massachusetts. Later that year, they would release two apps revolving around K-pop and photography that allowed fans of those subjects to chat freely. That same year, Amino was officially released. === Shutdown === In early December 2025, the Amino platform abruptly stopped all operations. Users worldwide lost access to the mobile application and website, with server requests returning connection time-out errors. Parent company MediaLab AI has issued no official statement regarding the cause to date, or declared any possible cause behind it. === Final Message === According to Shawn, a member of Amino support, Amino has ceased operations as of December 19th. The message that was sent out from Shawn reads: "Hey there, Thanks for your message. Amino has ceased operations. As of December 19th, we no longer retain personal data relating to you. Accordingly, we are unable to provide a copy of your data. Kind regards, - Amino Support" This message was sent on January 4th, 2026. This was the final support message sent from the Amino Support mail. == Growth == Amino received 1.65 million dollars of seed funding in 2014, primarily from Union Ventures. Some additional seed investors include Google Ventures, SV Angel, Box Group, and other interested parties. By July 2014, Amino's apps were downloaded 500,000 times. Though only having 15 communities at that time, Amino eventually grew to have 41 communities in September 2015. Amino's apps had been downloaded 13 million times by July 2016. Fandoms had migrated from websites like Facebook and Reddit to Amino, partly because of the app's mobile-native experience. Before 2016, when a user wanted to join a new Amino, they had to download another app for the Amino they wanted to join, with each apps name beginning with "Amino for:". In 2016, Amino Apps launched a centralized portal that hosted every Amino community in one app, meaning users no longer had to download multiple apps. In July of the same year, ACM, an app that allowed users to create their own communities, was launched. This resulted in the number of communities on Amino skyrocketing to over 2.5 million as of June 2018. == Features == The main feature of Amino was communities dedicated to a certain topic that users could join. Users could also chat with other members of a community in three ways: text, voice, or screening room, which allowed users to watch videos together while voice chatting. Other features include polls, blog posts, image posts, wiki entries, stories, and quizzes. In some cases, posts that were very well-made and had been noticed by a community's administration would end up receiving a feature, making it appear on the front page along with other featured content. In 2018, a premium membership option called Amino+ was added. Amino+ comes with additional features such as exclusive stickers, the ability to make stickers, custom chat bubbles, high resolution images, and other perks. Membership can now only be purchased with money. Amino coins can be purchased or earned through enabling ads, watching ad videos, completing activities on the Offer Wall, and playing Lucky Draw when checking in, but are of little use due to the users not being able to buy Amino+ by amino coins anymore. Members can give and receive coins through props. In 2019, Amino introduced six original short-form animated series, labelled "Amino Originals," produced by independent artists from across the internet. ATJ's "Little Red," a re-imagining of Little Red Riding Hood, premiered on November 15, 2019. "Little Red" was joined by five other shows in late December. Sophie Feher's "The Reef," a comedy featuring an aspiring marine biologist meeting a merman, premiered on December 27 alongside "Princely," an LGBT fairy tale created by Matt Bruneau-Richardson of Tiny Siren Animation. "Spaced Out," an alien abduction comedy by Michael Jae, and YouTuber Alex Clark's "Wyndvania II" premiered on December 28. Mysie Pereira's fairy tale "Turned to Stone" and Marcin Pawlowski's "Stranded" premiered on December 29, 2019. == Administration == On each community, there are two types of staff members, these being ‘Leader’ and ‘Curator.’ Leaders are higher rank than curators. Curators are usually the ones who feature posts, or post important announcements for users to see. Curators are able to disable a post or public chat, delete comments or chat threads, manage featured content, manage posts in topic categories, and approve Wiki entries. Leaders have more power than curators. In addition to curator powers, leaders can submit a community to be listed, change the Amino's features, change navigation, alter the community appearance, change the Amino's privacy settings, manage the Amino's join requests, send invites, appoint or demote Curators, strike or ban members, manage flagged content, change users' custom titles, manage topics and wiki categories, and create broadcasts (notifications sent for posts). One leader will have the status of agent. An agent is the primary leader of a community; the person who created the community is automatically agent. An agent has the ability to delete their community as long as it is not too large or too active. An agent can appoint and remove both leaders and curators. Agent status can be transferred voluntarily to another leader, curator, or community member. If an agent is inactive, Team Amino may assist in transferring agent status. == Apps == === Amino Community Manager === Otherwise known as ACM, this application is what users use to create and manage their own community in Amino. This app allows moderators to customize a community's theme, icon, and categories. ACM also allows moderation to customize community descriptions, pick leaders, change language settings, create a tagline for the community, change the home page lay out, alter the side navigation menu, and more. Unlisted communities are able to change their community's title and Amino ID, but this is not an option once a community is listed. A leader can use ACM to submit a request for their community to be listed on the explore page, after which the community will be reviewed by Team Amino for approval. Communities can be deleted on ACM, but only by the agent of that community. == Guidelines == Amino has a set of guidelines that all communities must comply with. Amino does not allow harassment or hate, spam or self-promotion (including promotion of one's own Amino community), sexual/NSFW content, self harm, real graphic/gross content (fictional content is generally acceptable), unsafe/illegal content, or content that violates copyright. Communities are allowed to have additional rules so long as they do not violate Amino's rules. In addition to Amino's rules, users are required to be at least 13 years of age in the U.S. and 16 years of age in European Union countries. While sexual imagery is not allowed in any community and text based sexual content is not allowed in public areas, some private communities are allowed to discuss sexual themes. However, they are not exempt from Amino's rules on NSFW content. If guidelines are broken, a leader may disable content or impose a warning, strike, or ban, depending on the severity of the infringement. A warning is a message informing the user that they have violated a guideline and may face further punishment unless they change their behaviour. A strike will put the user in read-only mode for up to 24 hours; this mode prevents the user from posting, chatting, or interacting with posts in that community. A ban removes the user from the community. Team Amino can separately issue users with strikes or bans across the entire platform. == Controversies == In 2017, organizations in Argentina for the protection of minors reported inappropriate material on the app, ranging from pornography to material promoting suicide to underage users. In 2019, Abilene police in Texas released a statement that sexual predators were using Amino chat rooms to approach minors. In 2020, authorities from the Christian County in the state of Kentucky alerted parents about possible sexual predators on Amino. In 2025, the British Police identified Amino as one of several platforms used by a child exploitation network that had previously extorted minors in different countries in Europe and North America. Several families reported to the National Society for the Prevention of Cruelty to Children that pedophiles were using the app for the purpose of sexual role-playing with minors, c
Content strategy
Content strategy guides the planning, development, and management of content. It is a recognized field in user experience design, and it also draws from adjacent disciplines such as information architecture, content management, business analysis, digital marketing, and technical communication. == Definitions == Content strategy has been described as planning for "the creation, publication, and governance of useful, usable content." It has also been called "a repeatable system that defines the entire editorial content development process for a website development project." In a 2007 article titled "Content Strategy: The Philosophy of Data," Rachel Lovinger describes the goal of content strategy as using "words and data to create unambiguous content that supports meaningful, interactive experiences." Here, she also provided the analogy that "content strategy is to copywriting as information architecture is to design." She encourages content strategists and collaborators to engage in early discussions about content meaning, models, and tools, to make sure strategy is integrated from the start rather than as an afterthought. The Content Strategy Alliance combines Kevin Nichols' definition with Kristina Halvorson's and defines content strategy as "getting the right content to the right user at the right time through strategic planning of content creation, delivery, and governance." == Practitioners == Content strategists are often familiar with a wide range of approaches, techniques, and tools. The perspectives that content strategists bring also depend heavily on their professional training and education. For instance, some specialize in "front-end strategy," which includes developing personas, journey mapping the user experience, aligning business strategy and user needs, developing a brand strategy, exploring different channels, and creating style guidelines and search engine optimization (SEO) guidelines. Others specialize in "back-end strategy," which includes creating content models, planning taxonomies and metadata, structuring content management systems, and building systems to support content reuse. Both roles involve addressing workflow and governance issues. Many organizations and individuals tend to confuse content strategists with editors. However, content strategy is "about more than just the written word," according to Washington State University associate professor Brett Atwood. For example, Atwood indicates that a practitioner needs to also "consider how content might be re-distributed and/or re-purposed in other channels of delivery." It has also been proposed that the content strategist performs the role of a curator. Just as a museum curator sifts through a collection of content and identifies key pieces that can be juxtaposed against each other to create meaning and spur excitement, a content strategist "must approach a business’s content as a medium that needs to be strategically selected and placed to engage the audience, convey a message, and inspire action."
Electric Literature
Electric Literature is an American literary magazine. == History == Founded by Andy Hunter and Scott Lindenbaum in 2009 as a print quarterly journal, Electric Literature transitioned to a daily website in 2012 under the helm of Halimah Marcus and Benjamin Samuel. Electric Literature publishes essays, reading lists, interviews, fiction, poetry, graphic narratives, humor, and book news, all available to read online for free without a paywall. It launched the first fiction magazine on the iPhone and iPad. Work published has been recognized by Best American Short Stories, Essays, Poetry, and Comics, the Pushcart Prize, Best Canadian Short Stories, The Best of the Small Presses, and the O. Henry Prize. in 2014, Electric Literature became a registered non-profit. In 2016, Halimah Marcus was appointed the first executive director of Electric Literature. She has been with the magazine since 2010. In 2021, Denne Michele Norris became editor-in-chief of Electric Literature, the first Black and openly trans editor-in-chief of a major U.S. literary publication. In 2022, Electric Literature was the Digital Prize Winner of the Whiting Literary Magazine Prizes. In 2023, Electric Literature partnered with Banned Books USA to offer free banned and challenged books to residents of Florida. In 2025, Electric Literature published their first book, edited by Norris and published by HarperOne: Both/And: Essays by Trans and Gender-Nonconforming Writers of Color. It builds on a prior essay series that Electric Literature sponsored for trans writers of color. Both/And became a finalist for the 2026 Lambda Literary Award for Transgender Nonfiction. == Recommended reading == In May 2012, Electric Literature launched Recommended Reading, a weekly fiction magazine. Each issue is curated by a well known editor or writer. == The Commuter == The Commuter, a weekly magazine for poetry, flash, graphic, or experimental narrative, debuted in January 2018, helmed by writer Kelly Luce.
Label noise
Label noise refers to errors or inaccuracies in the class labels of data instances. This is a widespread issue in machine learning datasets, arising from human annotator mistakes, unclear labeling instructions, automated labeling methods, or adversarial attacks in supervised learning. Label noise can be roughly divided into random noise, where labels are flipped independently of input features, and systematic noise, where mislabeling is dependent on certain patterns or biases in the data. Label noise can be damaging to model performance, especially for complex models that may overfit to noisy labels rather than generalizable patterns. Many approaches have been proposed to deal with the effects of label noise, including robust loss functions, noise-tolerant algorithms, data cleaning methods, and semi-supervised learning approaches. To reduce the impact of wrong labels during training, techniques like label smoothing, sample reweighting and using trusted validation sets are used. The role of noise-robust training paradigms and curriculum learning strategies to improve resilience against mislabeled data is also explored in recent research.
Control communications
In telecommunications, control communications is the branch of technology devoted to the design, development, and application of communications facilities used specifically for control purposes, such as for controlling (a) industrial processes, (b) movement of resources, (c) electric power generation, distribution, and utilization, (d) communications networks, and (e) transportation systems.