AI For Students Anthropic

AI For Students Anthropic — independent reviews, comparisons, pricing and step-by-step guides on Aizhi.

Intelligent agent

In artificial intelligence, an intelligent agent is an entity that perceives its environment, takes actions autonomously to achieve goals, and may improve its performance through machine learning or by acquiring knowledge. AI textbooks define artificial intelligence as the "study and design of intelligent agents," emphasizing that goal-directed behavior is central to intelligence. A specialized subset of intelligent agents, agentic AI (also known as an AI agent or simply agent), expands this concept by proactively pursuing goals, making decisions, and taking actions over extended periods. Intelligent agents can range from simple to highly complex. A basic thermostat or control system is considered an intelligent agent, as is a human being, or any other system that meets the same criteria—such as a firm, a state, or a biome. Intelligent agents operate based on an objective function, which encapsulates their goals. They are designed to create and execute plans that maximize the expected value of this function upon completion. For example, a reinforcement learning agent has a reward function, which allows programmers to shape its desired behavior. Similarly, an evolutionary algorithm's behavior is guided by a fitness function. Intelligent agents in artificial intelligence are closely related to agents in economics, and versions of the intelligent agent paradigm are studied in cognitive science, ethics, and the philosophy of practical reason, as well as in many interdisciplinary socio-cognitive modeling and computer social simulations. Intelligent agents are often described schematically as abstract functional systems similar to computer programs . To distinguish theoretical models from real-world implementations, abstract descriptions of intelligent agents are called abstract intelligent agents. Intelligent agents are also closely related to software agents—autonomous computer programs that carry out tasks on behalf of users. They are also referred to using a term borrowed from economics: a "rational agent". == Intelligent agents as the foundation of AI == The concept of intelligent agents provides a foundational lens through which to define and understand artificial intelligence. For instance, the influential textbook Artificial Intelligence: A Modern Approach (Russell & Norvig) describes: Agent: Anything that perceives its environment (using sensors) and acts upon it (using actuators). E.g., a robot with cameras and wheels, or a software program that reads data and makes recommendations. Rational Agent: An agent that strives to achieve the best possible outcome based on its knowledge and past experiences. "Best" is defined by a performance measure – a way of evaluating how well the agent is doing. Artificial Intelligence (as a field): The study and creation of these rational agents. Other researchers and definitions build upon this foundation. Padgham & Winikoff emphasize that intelligent agents should react to changes in their environment in a timely way, proactively pursue goals, and be flexible and robust (able to handle unexpected situations). Some also suggest that ideal agents should be "rational" in the economic sense (making optimal choices) and capable of complex reasoning, like having beliefs, desires, and intentions (BDI model). Kaplan and Haenlein offer a similar definition, focusing on a system's ability to understand external data, learn from that data, and use what is learned to achieve goals through flexible adaptation. Defining AI in terms of intelligent agents offers several key advantages: Avoids Philosophical Debates: It sidesteps arguments about whether AI is "truly" intelligent or conscious, like those raised by the Turing test or Searle's Chinese Room. It focuses on behavior and goal achievement, not on replicating human thought. Objective Testing: It provides a clear, scientific way to evaluate AI systems. Researchers can compare different approaches by measuring how well they maximize a specific "goal function" (or objective function). This allows for direct comparison and combination of techniques. Interdisciplinary Communication: It creates a common language for AI researchers to collaborate with other fields like mathematical optimization and economics, which also use concepts like "goals" and "rational agents." == Objective function == An objective function (or goal function) specifies the goals of an intelligent agent. An agent is deemed more intelligent if it consistently selects actions that yield outcomes better aligned with its objective function. In effect, the objective function serves as a measure of success. The objective function may be: Simple: For example, in a game of Go, the objective function might assign a value of 1 for a win and 0 for a loss. Complex: It might require the agent to evaluate and learn from past actions, adapting its behavior based on patterns that have proven effective. The objective function encapsulates all of the goals the agent is designed to achieve. For rational agents, it also incorporates the trade-offs between potentially conflicting goals. For instance, a self-driving car's objective function might balance factors such as safety, speed, and passenger comfort. Different terms are used to describe this concept, depending on the context. These include: Utility function: Often used in economics and decision theory, representing the desirability of a state. Objective function: A general term used in optimization. Loss function: Typically used in machine learning, where the goal is to minimize the loss (error). Reward Function: Used in reinforcement learning. Fitness Function: Used in evolutionary systems. Goals, and therefore the objective function, can be: Explicitly defined: Programmed directly into the agent. Induced: Learned or evolved over time. In reinforcement learning, a "reward function" provides feedback, encouraging desired behaviors and discouraging undesirable ones. The agent learns to maximize its cumulative reward. In evolutionary systems, a "fitness function" determines which agents are more likely to reproduce. This is analogous to natural selection, where organisms evolve to maximize their chances of survival and reproduction. Some AI systems, such as nearest-neighbor, reason by analogy rather than being explicitly goal-driven. However, even these systems can have goals implicitly defined within their training data. Such systems can still be benchmarked by framing the non-goal system as one whose "goal" is to accomplish its narrow classification task. Systems not traditionally considered agents, like knowledge-representation systems, are sometimes included in the paradigm by framing them as agents with a goal of, for example, answering questions accurately. Here, the concept of an "action" is extended to encompass the "act" of providing an answer. As a further extension, mimicry-driven systems can be framed as agents optimizing a "goal function" based on how closely the agent mimics the desired behavior. In generative adversarial networks (GANs) of the 2010s, an "encoder"/"generator" component attempts to mimic and improvise human text composition. The generator tries to maximize a function representing how well it can fool an antagonistic "predictor"/"discriminator" component. While symbolic AI systems often use an explicit goal function, the paradigm also applies to neural networks and evolutionary computing. Reinforcement learning can generate intelligent agents that appear to act in ways intended to maximize a "reward function". Sometimes, instead of setting the reward function directly equal to the desired benchmark evaluation function, machine learning programmers use reward shaping to initially give the machine rewards for incremental progress. Yann LeCun stated in 2018, "Most of the learning algorithms that people have come up with essentially consist of minimizing some objective function." AlphaZero chess had a simple objective function: +1 point for each win, and -1 point for each loss. A self-driving car's objective function would be more complex. Evolutionary computing can evolve intelligent agents that appear to act in ways intended to maximize a "fitness function" influencing how many descendants each agent is allowed to leave. The mathematical formalism of AIXI was proposed as a maximally intelligent agent in this paradigm. However, AIXI is uncomputable. In the real world, an intelligent agent is constrained by finite time and hardware resources, and scientists compete to produce algorithms that achieve progressively higher scores on benchmark tests with existing hardware. == Agent function == An intelligent agent's behavior can be described mathematically by an agent function. This function determines what the agent does based on what it has seen. A percept refers to the agent's sensory inputs at a single point in time. For example, a self-driving car's percepts might include camera images, lidar data, GPS coordinates, and speed r
Read more →
Blockhead (thought experiment)

Blockhead is a theoretical computer system invented as part of a thought experiment by philosopher Ned Block, which appeared in a paper titled "Psychologism and Behaviorism". Block did not personally name the computer in the paper. == Overview == In "Psychologism and Behaviorism", Block argues that the internal mechanism of a system is important in determining whether that system is intelligent and claims to show that a non-intelligent system could pass the Turing test. Block asks the reader to imagine a conversation lasting any given amount of time. He states that given the nature of language, there are a finite number of syntactically and grammatically correct sentences that can be used to start a conversation. Consequently, there is a limit to how many "sensible" responses can be made to the first sentence, then to the second sentence, and so on until the conversation ends. Block then asks the reader to imagine a computer which had been programmed with all the sentences in theory, if not in practice. Block argues that such a machine could continue a conversation with a person on any topic because the computer would be programmed with every sentence that it was possible to use so the computer would be able to pass the Turing test despite the fact that—according to Block—it was not intelligent. Block says that this does not show that there is only one correct internal structure for generating intelligence but simply that some internal structures do not generate intelligence. The argument is related to John Searle's Chinese room.
Read more →
OntoWiki

OntoWiki was a free and open-source semantic wiki application, meant to serve as an ontology editor and a knowledge acquisition system. It is a web-based application written in PHP and using either a MySQL database or a Virtuoso triple store. OntoWiki is form-based rather than syntax-based, and thus tries to hide as much of the complexity of knowledge representation formalisms from users as possible. OntoWiki is mainly being developed by the Agile Knowledge Engineering and Semantic Web (AKSW) research group at the University of Leipzig, a group also known for the DBpedia project among others, in collaboration with volunteers around the world. In 2009 the AKSW research group got a budget of €425,000 from the Federal Ministry of Education and Research of Germany for the development of the OntoWiki. In 2010 OntoWiki became part of the technology stack supporting the LOD2 (linked open data) project. Leipzig University is one of the consortium members of the project, which is funded by a €6.5m EU grant. The development ended in 2016 due to the lack of capacity migrating from PHP 5 to 7 including the required Zend Framework from version 1 to 2.
Read more →
Maia and Marco

Maia and Marco are artificial intelligence used by GMA Network. Unveiled in 2023, they are used to fulfill the role of sports newscasters. == Background == Maia and Marco are artificial intelligence (AI) which take the form of three-dimensional human avatars. Maia makes use of a female avatar while Marco uses a male likeness. They have aesthetic features that are typical to Filipino showbusiness personalities. Among the technologies used in making and operating the AI include image generation, text-to-speech AI voice synthesis/generation, and deep learning face animation. They are also demonstrated to be bilingual, being able to speak in English and Tagalog (Filipino). == Use == The AI pair was unveiled by GMA Network on September 24, 2023, for their coverage of Season 99 of the National Collegiate Athletic Association (NCAA). Fulfilling the role of sports newscasters, Maia and Marco would join GMA's courtside human reporters. The AI pair are scheduled to appear four times a month on GMA's digital media platforms. They will not appear in traditional television broadcast. == Reception == The launch of the Maia and Marco was met with strong reactions. Various journalists and other personalities across the Philippine media industry expressed concern that their employment be at risk with the introduction of AI. The quality of the AI ability to emulate human behavior was characterized by critics as "soulless". GMA responding to concerns has stated that the AI would complement rather than replace its live human journalists including sportscasters. The National Union of Journalists of the Philippines urged dialogue among its peers in the newsroom on policy on how to use AI, which the group acknowledge as "inevitable".
Read more →
Native cloud application

A native cloud application (NCA) is a type of computer software that natively utilizes services and infrastructure from cloud computing providers such as Amazon EC2, Force.com, or Microsoft Azure. NCAs exhibit a combined usage of the three fundamental technologies: Computational grid - loosely, e.g. MapReduce Data grids (e.g. distributed in-memory data caches) Auto-scaling on any managed infrastructure
Read more →
Mind map

A mind map is a diagram used to visually organize information into a hierarchy, showing relationships among pieces of the whole. It is often based on a single concept, drawn as an image in the center of a blank page, to which associated representations of ideas such as images, words and parts of words are added. Major ideas are connected directly to the central concept, and other ideas branch out from those major ideas. Mind maps can also be drawn by hand, either as "notes" during a lecture, meeting or planning session, for example, or as higher quality pictures when more time is available. Mind maps are considered to be a type of spider diagram. == Origin == Although the term "mind map" was first popularized by British popular psychology author and television personality Tony Buzan, the use of diagrams that visually "map" information using branching and radial maps traces back centuries. These pictorial methods record knowledge and model systems, and have a long history in learning, brainstorming, memory, visual thinking, and problem solving by educators, engineers, psychologists, and others. Some of the earliest examples of such graphical records were developed by Porphyry of Tyros, a noted thinker of the 3rd century, as he graphically visualized the concept categories of Aristotle. Philosopher Ramon Llull (1235–1315) also used such techniques. Buzan's specific approach, and the introduction of the term "mind map", started with a 1974 BBC TV series he hosted, called Use Your Head. In this show, and companion book series, Buzan promoted his conception of radial tree, diagramming key words in a colorful, radiant, tree-like structure. == Differences from other visualizations == Concept maps: Mind maps differ from concept maps in that mind maps are based on a radial hierarchy (tree structure) denoting relationships with a central concept, whereas concept maps can be more free-form, based on connections between concepts in more diverse patterns. Also, concept maps typically have text labels on the links between nodes. However, either can be part of a larger personal knowledge base system. Modeling graphs or graphical modeling languages: There is no rigorous right or wrong with mind maps, which rely on the arbitrariness of mnemonic associations to aid people's information organization and memory. In contrast, a modeling graph such as a UML diagram structures elements using a precise standardized iconography to aid the design of systems. == Research == === Effectiveness === Cunningham (2005) conducted a user study in which 80% of the students thought "mindmapping helped them understand concepts and ideas in science". Other studies also report some subjective positive effects of the use of mind maps. Positive opinions on their effectiveness, however, were much more prominent among students of art and design than in students of computer and information technology, with 62.5% vs 34% (respectively) agreeing that they were able to understand concepts better with mind mapping software. Farrand, Hussain, and Hennessy (2002) found that spider diagrams (similar to concept maps) had limited, but significant, impact on memory recall in undergraduate students (a 10% increase over baseline for a 600-word text only) as compared to preferred study methods (a 6% increase over baseline). This improvement was only robust after a week for those in the diagram group and there was a significant decrease in motivation compared to the subjects' preferred methods of note taking. A meta study about concept mapping concluded that concept mapping is more effective than "reading text passages, attending lectures, and participating in class discussions". The same study also concluded that concept mapping is slightly more effective "than other constructive activities such as writing summaries and outlines". However, results were inconsistent, with the authors noting "significant heterogeneity was found in most subsets". In addition, they concluded that low-ability students may benefit more from mind mapping than high-ability students. === Features === Joeran Beel and Stefan Langer conducted a comprehensive analysis of the content of mind maps. They analysed 19,379 mind maps from 11,179 users of the mind mapping applications SciPlore MindMapping (now Docear) and MindMeister. Results include that average users create only a few mind maps (mean=2.7), average mind maps are rather small (31 nodes) with each node containing about three words (median). However, there were exceptions. One user created more than 200 mind maps, the largest mind map consisted of more than 50,000 nodes and the largest node contained ~7,500 words. The study also showed that between different mind mapping applications (Docear vs MindMeister) significant differences exist related to how users create mind maps. === Automatic creation === There have been some attempts to create mind maps automatically. Brucks & Schommer created mind maps automatically from full-text streams. Rothenberger et al. extracted the main story of a text and presented it as mind map. There is also a patent application about automatically creating sub-topics in mind maps. == Tools == Mind-mapping software can be used to organize large amounts of information, combining spatial organization, dynamic hierarchical structuring and node folding.Software packages can extend the concept of mind-mapping by allowing individuals to map more than thoughts and ideas with information on their computers and the Internet, like spreadsheets, documents, Internet sites, images and videos. It has been suggested that mind-mapping can improve learning/study efficiency up to 15% over conventional note-taking. == Gallery == The following dozen examples of mind maps show the range of styles that a mind map may take, from hand-drawn to computer-generated and from mostly text to highly illustrated. Despite their stylistic differences, all of the examples share a tree structure that hierarchically connects sub-topics to a main topic.
Read more →
Noam Shazeer

Noam Shazeer (born 1975 or 1976) is an American computer scientist and entrepreneur known for his contributions to the field of artificial intelligence and deep learning, particularly in the development of transformer models and natural language processing. He lives in Palo Alto, California. == Career == Noam Shazeer joined Google in 2000. One of his first major achievements was improving the spelling corrector of Google's search engine. In 2017, Shazeer was one of the lead authors of the seminal paper "Attention Is All You Need", which introduced the transformer architecture. At Google, Shazeer and his colleague Daniel de Freitas built a chatbot named Meena. Following the refusal of Google to release the chatbot to the public, Shazeer and Freitas left the company in 2021 to found Character.AI. In September 2023, Time Magazine chose Shazeer as one of the 100 most influential people in the AI world. In August 2024, it was reported that Shazeer would be returning to Google to co-lead the Gemini AI project. Shazeer was appointed as technical lead on Gemini, along with Jeff Dean and Oriol Vinyals. It was part of a $2.7 billion deal for Google to license Character's technology. Since he owns 30-40% of the company, it is estimated he netted $750 million-$1 billion. In 2026, he was elected a member of the National Academy of Engineering. == Views == Shazeer said about artificial general intelligence that he doesn't "particularly care about AGI in the sense of wanting something that can do absolutely everything a person can do”. When asked in 2023 if he is afraid that AGI will destroy the world, he said: "No. Not yet. [...] We’re going to work on it as the technology improves". When asked why do large language models work he answered: "My best guess is divine benevolence [...] Nobody really understands what’s going on. This is a very experimental science [...] It’s more like alchemy or whatever chemistry was in the Middle Ages.” Shazeer has stated, "I do not believe that humans have an attribute called gender... I do not believe that G-d puts people in the wrong bodies. I do not believe that it is okay to sterilize children." == Personal life == Shazeer is an orthodox Jew. His grandparents escaped the Holocaust into the Soviet Union and later lived some time in Israel before emigrating to the USA. His father, Dov Shazeer, was a math teacher who became an engineer and his mother was a homemaker. His sister was ordained as a rabbi by Hebrew College. Shazeer was born in Philadelphia, attended grade school at Cohen Hillel Academy in Marblehead, Massachusetts, and attended Swampscott High School in Swampscott, Massachusetts. He won a gold medal with perfect score at International Mathematical Olympiad 1994 as a member of the USA team. He went on to study math and computer science at Duke University in Durham, North Carolina from 1994 to 1998. At Duke he was a recipient of the Angier B. Duke Memorial Scholarship, and, as part of the Duke math team, won prizes in several math tournaments. He started studying in a graduate program in Berkeley but did not finish it. He is a father of three and is married to Yael Shacham Shazeer
Read more →
Preferential entailment

Preferential entailment is a non-monotonic logic based on selecting only models that are considered the most plausible. The plausibility of models is expressed by an ordering among models called a preference relation, hence the name preference entailment. Formally, given a propositional formula F {\displaystyle F} and an ordering over propositional models ≤ {\displaystyle \leq } , preferential entailment selects only the models of F {\displaystyle F} that are minimal according to ≤ {\displaystyle \leq } . This selection leads to a non-monotonic inference relation: F ⊨ pref G {\displaystyle F\models _{\text{pref}}G} holds if and only if all minimal models of F {\displaystyle F} according to ≤ {\displaystyle \leq } are also models of G {\displaystyle G} . Circumscription can be seen as the particular case of preferential entailment when the ordering is based on containment of the sets of variables assigned to true (in the propositional case) or containment of the extensions of predicates (in the first-order logic case).
Read more →
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
Read more →
AutoGPT

AutoGPT is an open-source autonomous software agent that uses OpenAI's large language models, such as GPT-4, to attempt to achieve a goal specified by a user in natural language. Unlike chatbots that require continuous user commands, AutoGPT works autonomously by breaking the main goal into smaller sub-tasks and using tools like web browsing and file management to complete them. Released in March 2023, the project quickly gained popularity on GitHub and social media, with users creating agents for tasks like software development, market research, and content creation. One notable experiment, ChaosGPT, was tasked with destroying humanity, which brought mainstream attention to the technology's potential. However, AutoGPT is known for significant limitations, including a tendency to get stuck in loops, hallucinate information, and incur high operational costs due to its reliance on paid APIs. == Background == AutoGPT was released on March 30, 2023, by Toran Bruce Richards, the founder of video game company Significant Gravitas Ltd. It was one of the first widely accessible applications to showcase the autonomous capabilities of GPT-4, which had been released weeks earlier. Richards's goal was to create a model that could respond to real-time feedback and pursue objectives with a long-term outlook without needing constant human intervention. The application operates by prompting a user to define an agent's name, role, and main objective, including up to five sub-goals to achieve it. AutoGPT then works independently to reach its objective. The project is publicly available on GitHub but requires users to install it in a development environment like Docker and have a paid OpenAI account to obtain the necessary API key. In October 2023, the project's parent company, Significant Gravitas Ltd., raised $12 million in venture funding to support further development. == Capabilities == The overarching capability of AutoGPT is the breaking down of a large task into various sub-tasks without the need for user input. These sub-tasks are then chained together and performed sequentially to yield a larger result as originally laid out by the user input. One of the distinguishing features of AutoGPT is its ability to connect to the internet. This allows for up-to-date information retrieval to help complete tasks. In addition, AutoGPT maintains short-term memory for the current task, which allows it to provide context to subsequent sub-tasks needed to achieve the larger goal. Another feature is its ability to store and organize files so users can better structure their data for future analysis and extension. AutoGPT is also multimodal, which means that it can take in both text and images as input. With these features, AutoGPT is claimed to be capable of automating workflows, analyzing data, and coming up with new suggestions. == Applications == === Software === AutoGPT can be used to develop software applications from scratch. AutoGPT can also debug code and generate test cases. Observers suggest that AutoGPT's ability to write, debug, test, and edit code may extend to AutoGPT's own source code, enabling self-improvement. === Business === AutoGPT can be used to do market research, analyze investments, research products and write product reviews, create a business plan or improve operations, and create content such as a blog or podcast. One user has used AutoGPT to conduct product research and write a summary on the best headphones. Another user has used AutoGPT to summarize recent news events and prepare an outline for a podcast. === Other === AutoGPT was used to create ChefGPT, an AI agent able to independently explore the internet to generate and save unique recipes. AutoGPT was also used to create ChaosGPT, an AI agent tasked to “destroy humanity, establish global dominance, cause chaos and destruction, control humanity through manipulation, and attain immortality”. ChaosGPT reportedly researched nuclear weapons and tweeted disparagingly about humankind. == Limitations == AutoGPT is susceptible to frequent mistakes, primarily because it relies on its own feedback, which can compound errors. In contrast, non-autonomous models can be corrected by users overseeing their outputs. Furthermore, AutoGPT has a tendency to hallucinate or to present false or misleading information as fact when responding. AutoGPT can be constrained by the cost associated with running it as its recursive nature requires it to continually call the OpenAI API on which it is built. Every step required in one of AutoGPT's tasks requires a corresponding call to GPT-4 at a cost of at least about $0.03 for every 1000 tokens used for inputs and $0.06 for every 1000 tokens for output when choosing the cheapest option. For reference, 1000 tokens roughly result in 750 words. Another limitation is AutoGPT's tendency to get stuck in infinite loops. Developers believe that this is a result of AutoGPT's inability to remember, as it is unaware of what it has already done and repeatedly attempts the same subtask without end. Andrej Karpathy, co-founder of OpenAI which creates GPT-4, further explains that it is AutoGPT's “finite context window” that can limit its performance and cause it to “go off the rails”. Like other autonomous agents, AutoGPT is prone to distraction and unable to focus on its objective due to its lack of long-term memory, leading to unpredictable and unintended behavior. == Reception == AutoGPT became the top trending repository on GitHub after its release and has since repeatedly trended on Twitter. In April 2023, Avram Piltch wrote for Tom's Hardware that AutoGPT 'might be too autonomous to be useful,' as it did not ask questions to clarify requirements or allow corrective interventions by users. Piltch nonetheless noted that such tools have "a ton of potential" and should improve with better language models and further development. Malcolm McMillan from Tom's Guide mentioned that AutoGPT may not be better than ChatGPT for tasks involving conversation, as ChatGPT is well-suited for situations in which advice, rather than task completion, is sought. Will Knight from Wired wrote that AutoGPT is not a foolproof task-completion tool. When given a test task of finding a public figure's email address, he noted that it was not able to accurately find the email address. Clara Shih, Salesforce Service Cloud CEO commented that "AutoGPT illustrates the power and unknown risks of generative AI," and that due to usage risks, enterprises should include a human in the loop when using such technologies. Performance is reportedly enhanced when using AutoGPT with GPT-4 compared to GPT-3.5. For example, one reviewer who tested it on a task of finding the best laptops on the market with pros and cons found that AutoGPT with GPT-4 created a more comprehensive report than one by GPT 3.5.
Read more →
Responsible AI Safety and Education Act

The Responsible AI Safety and Education Act (RAISE Act) is a New York State law that imposes transparency, safety, and reporting requirements on developers of large frontier artificial intelligence models. The law was signed by Governor Kathy Hochul on December 19, 2025. It was sponsored by State Senator Andrew Gounardes and Assemblymember Alex Bores. The RAISE Act is the second U.S. state law to regulate frontier AI model developers, following California's Transparency in Frontier Artificial Intelligence Act (TFAIA), which was signed in September 2025. Hochul signed the bill on the condition that the legislature would pass chapter amendments to bring the law closer to the California model. The amending bills (A9449/S8828) were introduced in January 2026; as of February 2026 they remain in committee, though the Governor's office and legal commentators treat the agreed-upon amendments as representing the final form of the law. == Provisions == The following describes the RAISE Act as it is expected to operate after the agreed-upon chapter amendments take effect. The law is expected to take effect on January 1, 2027. === Scope === The law applies to "large frontier developers," defined as companies with annual revenues exceeding $500 million that develop "frontier models," which are foundation models trained using more than 1026 floating-point operations (FLOPs). The version passed by the legislature in June 2025 had instead defined large developers based on having spent over $100 million in aggregate compute costs, and also included a provision prohibiting deployment of frontier models posing "unreasonable risk of critical harm"; both were removed as part of the negotiations between Hochul and the legislature. Accredited colleges and universities engaged in academic research are exempt, as is the state's Empire AI consortium. === Safety and transparency framework === Large frontier developers must write, implement, and publicly publish a "frontier AI framework" describing how they assess and mitigate catastrophic risks, secure unreleased model weights against unauthorized access, use third-party evaluators, govern internal use of frontier models, and respond to safety incidents. The framework must describe these measures "in detail," a requirement that goes beyond the California TFAIA's requirement to describe a developer's "approach." The framework must be reviewed at least annually, and material modifications must be published with justification within 30 days. Before or concurrently with deploying a new or substantially modified frontier model, developers must publish a transparency report including the model's release date, supported languages and output modalities, intended uses, and any restrictions on use. Large frontier developers must additionally include summaries of catastrophic risk assessments and the extent of third-party involvement. === Catastrophic risk and incident reporting === The law defines "catastrophic risk" as a foreseeable and material risk that a frontier model will contribute to the death of or serious injury to more than 50 people, or more than $1 billion in property damage, arising from a frontier model providing expert-level assistance in creating chemical, biological, radiological, or nuclear weapons; engaging in cyberattacks or conduct equivalent to crimes such as murder, assault, or theft without meaningful human oversight; or evading the control of its developer or user. Loss of equity value is explicitly excluded from the definition of property damage. "Critical safety incidents" include unauthorized access to model weights resulting in death or injury, materialization of a catastrophic risk, loss of control of a frontier model causing death or injury, and a model using deceptive techniques to subvert developer controls outside of an evaluation context in a manner that increases catastrophic risk. Frontier developers must report critical safety incidents within 72 hours, or within 24 hours if the incident poses an imminent risk of death or serious physical injury. === Enforcement === The chapter amendments establish a new office within the New York State Department of Financial Services to oversee compliance, receive incident reports, and publish annual reports on AI safety beginning in 2028. Large frontier developers must file disclosure statements with this office and pay pro rata assessments to fund its operations. The New York Attorney General may bring civil actions, with penalties of up to $1 million for a first violation and $3 million for subsequent violations. The version passed by the legislature in June 2025 had set penalties at up to $10 million and $30 million respectively. The law does not create a private right of action. == Legislative history == The bill was introduced in the Assembly on March 5, 2025, by Assemblymember Alex Bores, and in the Senate on March 27, 2025, by Senator Andrew Gounardes. After a series of amendments, the legislature passed the bill in June 2025. Governor Hochul did not immediately sign the bill, using nearly all the time available under New York law before acting; had she not signed by the end of 2025, the bill would have been pocket vetoed. The tech industry lobbied against the bill during this period, and Hochul initially proposed a near-complete rewrite modeled on California's TFAIA. Legislators resisted the extent of the changes, and the two sides ultimately agreed on a version that used the California law as a base but preserved several provisions that went beyond it, including the 72-hour incident reporting timeline and the creation of a dedicated enforcement office. Hochul signed the original bill (S6953-B/A6453-B) on December 19, 2025, with the legislature committing to pass chapter amendments formalizing the agreed changes in the January 2026 session. The amending bills (A9449 in the Assembly, S8828 in the Senate) were introduced on January 6 and January 8, 2026. OpenAI and Anthropic expressed support for the law. Anthropic's head of external affairs Sarah Heck said the two state laws "should inspire Congress to build on them." The super PAC network Leading the Future, backed by Andreessen Horowitz and OpenAI president Greg Brockman, subsequently announced plans to challenge Bores in a future election. == Federal preemption debate == Hochul signed the RAISE Act eight days after President Donald Trump issued an executive order on December 11, 2025, directing the Department of Justice to challenge state AI laws deemed to conflict with a "minimally burdensome" national AI policy. On January 9, 2026, the Department of Justice announced the establishment of an AI Litigation Task Force as called for by the executive order. The executive order also threatened states with loss of certain federal broadband funding if their AI laws were found to be onerous. Legal commentators have noted several potential avenues for federal challenge, including arguments that the law constitutes compelled speech, violates the dormant Commerce Clause by creating a patchwork of state regulations, or is preempted by federal AI policy. == Comparison with California's TFAIA == The RAISE Act was designed to align with California's Transparency in Frontier Artificial Intelligence Act, signed on September 29, 2025. Both laws use the same 1026 FLOP threshold to define frontier models and the same $500 million revenue threshold to define large developers. Both require public safety frameworks, transparency reports, and incident reporting. The RAISE Act's 72-hour incident reporting window is stricter than the TFAIA's 15-day window, though both require faster reporting for incidents posing imminent physical risk (24 hours under the RAISE Act, immediate under the TFAIA). The RAISE Act establishes a dedicated enforcement office within the Department of Financial Services, whereas California routes reports through the Office of Emergency Services. The RAISE Act requires developers to describe their safety measures "in detail" and how they "handle" various risks, whereas the TFAIA requires developers to describe their "approach."
Read more →
InRule Technology

InRule Technology is a software company that offers Business Rule Management System (BRMS) enterprise software products. == History == InRule Technology's Chief Executive Officer Rik Chomko and Chief Technology Officer Loren Goodman founded InRule Technology in Chicago in 2002. Paul Hessinger joined InRule Technology in 2004 as chief executive officer and chairman of the board and served until his retirement in 2015. They work with companies in several markets, including financial services, public sector, healthcare, and insurance. In 2007, InRule Technology became a charter member of the Microsoft Business Process Alliance. In August 2019, InRule was acquired by Open Gate Capital. == Products == On October 29, 2012, InRule Technology launched InRule for Microsoft Dynamics CRM. The program provides components to enable creation and update of rules within Microsoft Dynamics CRM, InRule for Microsoft Dynamics CRM provides a platform for shops that prefer to work with Microsoft's platforms. With the availability of InRule 4.6 in 2014, the company introduced deployment of InRule through REST services and allowed REST services to be called from InRule. This enables access to data exposed as a REST service and to package up a rule service for RESTful access. The product launch reflected the move of the company's core audience to use a broader array of technologies despite an earlier focus on .NET. In 2017, InRule introduced InRule for the Salesforce Platform, as well as a technology partnership with Work-Relay, a Business Process Management (BPM) application built on the Salesforce Platform. One year earlier the company introduced InRule for JavaScript, allowing enterprises to run rules on the client-side, server-side or both. The software architecture includes multiple components, including irAuthor, the primary authoring tool for creating and maintaining rules; irVerify, a real-time test environment to run and debug rule applications; and irSDK, a set of APIs that allows developers to integrate inRule into their applications. Additionally, irSOA allows users to access the InRule rule engine as a service. irSOA is now called the irServer Execution Service.
Read more →
Maximum inner-product search

Maximum inner-product search (MIPS) is a search problem, with a corresponding class of search algorithms which attempt to maximise the inner product between a query and the data items to be retrieved. MIPS algorithms are used in a wide variety of big data applications, including recommendation algorithms and machine learning. Formally, for a database of vectors x i {\displaystyle x_{i}} defined over a set of labels S {\displaystyle S} in an inner product space with an inner product ⟨ ⋅ , ⋅ ⟩ {\displaystyle \langle \cdot ,\cdot \rangle } defined on it, MIPS search can be defined as the problem of determining a r g m a x i ∈ S ⟨ x i , q ⟩ {\displaystyle {\underset {i\in S}{\operatorname {arg\,max} }}\ \langle x_{i},q\rangle } for a given query q {\displaystyle q} . Although there is an obvious linear-time implementation, it is generally too slow to be used on practical problems. However, efficient algorithms exist to speed up MIPS search. Under the assumption of all vectors in the set having constant norm, MIPS can be viewed as equivalent to a nearest neighbor search (NNS) problem in which maximizing the inner product is equivalent to minimizing the corresponding distance metric in the NNS problem. Like other forms of NNS, MIPS algorithms may be approximate or exact. MIPS search is used as part of DeepMind's RETRO algorithm.
Read more →
Information Processing Language

Information Processing Language (IPL) is a programming language created by Allen Newell, Cliff Shaw, and Herbert A. Simon at RAND Corporation and the Carnegie Institute of Technology about 1956. Newell had the job of language specifier-application programmer, Shaw was the system programmer, and Simon had the job of application programmer-user. IPL included features to facilitate AI programming, specifically problem solving. such as lists, dynamic memory allocation, data types, recursion, functions as arguments, generators, and cooperative multitasking. IPL also introduced the concepts of symbol processing and list processing. Unfortunately, all of these innovations were cast in a difficult assembly-language style. Nonetheless, IPL-V (the only public version of IPL) ran on many computers through the mid 1960s. == Basics of IPL == An IPL computer has: A set of symbols. All symbols are addresses, and name cells. Unlike symbols in later languages, symbols consist of a character followed by a number, and are written H1, A29, 9–7, 9–100. Cell names beginning with a letter are regional, and are absolute addresses. Cell names beginning with "9-" are local, and are meaningful within the context of a single list. One list's 9-1 is independent of another list's 9–1. Other symbols (e.g., pure numbers) are internal. A set of cells. Lists are made from several cells including mutual references. Cells have several fields: P, a 3-bit field used for an operation code when the cell is used as an instruction, and unused when the cell is data. Q, a 3-valued field used for indirect reference when the cell is used as an instruction, and unused when the cell is data. SYMB, a symbol used as the value in the cell. A set of primitive processes, which would be termed primitive functions in modern languages. The data structure of IPL is the list, but lists are more intricate structures than in many languages. A list consists of a singly linked sequence of symbols, as might be expected—plus some description lists, which are subsidiary singly linked lists interpreted as alternating attribute names and values. IPL provides primitives to access and mutate attribute value by name. The description lists are given local names (of the form 9–1). So, a list named L1 containing the symbols S4 and S5, and described by associating value V1 to attribute A1 and V2 to A2, would be stored as follows. 0 indicates the end of a list; the cell names 100, 101, etc. are automatically generated internal symbols whose values are irrelevant. These cells can be scattered throughout memory; only L1, which uses a regional name that must be globally known, needs to reside in a specific place. IPL is an assembly language for manipulating lists. It has a few cells which are used as special-purpose registers. H1, for example, is the program counter. The SYMB field of H1 is the name of the current instruction. However, H1 is interpreted as a list; the LINK of H1 is, in modern terms, a pointer to the beginning of the call stack. For example, subroutine calls push the SYMB of H1 onto this stack. H2 is the free-list. Procedures which need to allocate memory grab cells off of H2; procedures which are finished with memory put it on H2. On entry to a function, the list of parameters is given in H0; on exit, the results should be returned in H0. Many procedures return a Boolean result indicating success or failure, which is put in H5. Ten cells, W0-W9, are reserved for public working storage. Procedures are "morally bound" (to quote the CACM article) to save and restore the values of these cells. There are eight instructions, based on the values of P: subroutine call, push/pop S to H0; push/pop the symbol in S to the list attached to S; copy value to S; conditional branch. In these instructions, S is the target. S is either the value of the SYMB field if Q=0, the symbol in the cell named by SYMB if Q=1, or the symbol in the cell named by the symbol in the cell named by SYMB if Q=2. In all cases but conditional branch, the LINK field of the cell tells which instruction to execute next. IPL has a library of some 150 basic operations. These include such operations as: Test symbols for equality Find, set, or erase an attribute of a list Locate the next symbol in a list; insert a symbol in a list; erase or copy an entire list Arithmetic operations (on symbol names) Manipulation of symbols; e.g., test if a symbol denotes an integer, or make a symbol local I/O operations "Generators", which correspond to iterators and filters in functional programming. For example, a generator may accept a list of numbers and produce the list of their squares. Generators could accept suitably designed functions—strictly, the addresses of code of suitably designed functions—as arguments. == History == IPL was first utilized to demonstrate that the theorems in Principia Mathematica which were proven laboriously by hand, by Bertrand Russell and Alfred North Whitehead, could in fact be proven by computation. According to Simon's autobiography Models of My Life, this application was originally developed first by hand simulation, using his children as the computing elements, while writing on and holding up note cards as the registers which contained the state variables of the program. IPL was used to implement several early artificial intelligence programs, also by the same authors: the Logic Theorist (1956), the General Problem Solver (1957), and their computer chess program NSS (1958). Several versions of IPL were created: IPL-I (never implemented), IPL-II (1957 for JOHNNIAC), IPL-III (existed briefly), IPL-IV, IPL-V (1958, for IBM 650, IBM 704, IBM 7090, Philco model 212, many others. Widely used). IPL-VI was a proposal for an IPL hardware. A co-processor “IPL-VC” for the CDC 3600 at Argonne National Libraries was developed which could run IPL-V commands. It was used to implement another checker-playing program. This hardware implementation did not improve running times sufficiently to “compete favorably with a language more directly oriented to the structure of present-day machines”. IPL was soon displaced by Lisp, which had much more powerful features, a simpler syntax, and the benefit of automatic garbage collection. == Legacy to computer programming == IPL arguably introduced several programming language features: List manipulation—but only lists of atoms, not general lists Property lists—but only when attached to other lists Higher-order functions—while assembly programming had always allowed computing with the addresses of functions, IPL was an early attempt to generalize this property of assembly language in a principled way Computation with symbols—though symbols have a restricted form in IPL (letter followed by number) Virtual machine Many of these features were generalized, rationalized, and incorporated into Lisp and from there into many other programming languages during the next several decades.
Read more →
Alexey Chervonenkis

Alexey Yakovlevich Chervonenkis (Russian: Алексей Яковлевич Червоненкис; 7 September 1938 – 22 September 2014) was a Soviet and Russian mathematician. Along with Vladimir Vapnik, he was one of the main developers of the Vapnik–Chervonenkis theory, also known as the "fundamental theory of learning", an important part of computational learning theory. Chervonenkis held joint appointments with the Russian Academy of Sciences and Royal Holloway, University of London. Alexey Chervonenkis got lost in Losiny Ostrov National Park on 22 September 2014, and later during a search operation was found dead near Mytishchi, a suburb of Moscow. He had died of hypothermia.
Read more →