Chinchilla is a family of large language models (LLMs) developed by the research team at Google DeepMind, presented in March 2022. == Models == It is named "chinchilla" because it is a further development over a previous model family named Gopher. Both model families were trained in order to investigate the scaling laws of large language models. It claimed to outperform GPT-3. It considerably simplifies downstream utilization because it requires much less computer power for inference and fine-tuning. Based on the training of previously employed language models, it has been determined that if one doubles the model size, one must also have twice the number of training tokens. This hypothesis has been used to train Chinchilla by DeepMind. Similar to Gopher in terms of cost, Chinchilla has 70B parameters and four times as much data. Chinchilla has an average accuracy of 67.5% on the Measuring Massive Multitask Language Understanding (MMLU) benchmark, which is 7% higher than Gopher's performance. Chinchilla was still in the testing phase as of January 12, 2023. Chinchilla contributes to developing an effective training paradigm for large autoregressive language models with limited compute resources. The Chinchilla team recommends that the number of training tokens is twice for every model size doubling, meaning that using larger, higher-quality training datasets can lead to better results on downstream tasks. It has been used for the Flamingo vision-language model. == Architecture == Both the Gopher family and Chinchilla family are families of transformer models. In particular, they are essentially the same as GPT-2, with different sizes and minor modifications. Gopher family uses RMSNorm instead of LayerNorm; relative positional encoding rather than absolute positional encoding. The Chinchilla family is the same as the Gopher family, but trained with AdamW instead of Adam optimizer. The Gopher family contains six models of increasing size, from 44 million parameters to 280 billion parameters. They refer to the largest one as "Gopher" by default. Similar naming conventions apply for the Chinchilla family. Table 1 of shows the entire Gopher family: Table 4 of compares the 70-billion-parameter Chinchilla with Gopher 280B.
Schema-agnostic databases
Schema-agnostic databases or vocabulary-independent databases aim at supporting users to be abstracted from the representation of the data, supporting the automatic semantic matching between queries and databases. Schema-agnosticism is the property of a database of mapping a query issued with the user terminology and structure, automatically mapping it to the dataset vocabulary. The increase in the size and in the semantic heterogeneity of database schemas bring new requirements for users querying and searching structured data. At this scale it can become unfeasible for data consumers to be familiar with the representation of the data in order to query it. At the center of this discussion is the semantic gap between users and databases, which becomes more central as the scale and complexity of the data grows. == Description == The evolution of data environments towards the consumption of data from multiple data sources and the growth in the schema size, complexity, dynamicity and decentralisation (SCoDD) of schemas increases the complexity of contemporary data management. The SCoDD trend emerges as a central data management concern in Big Data scenarios, where users and applications have a demand for more complete data, produced by independent data sources, under different semantic assumptions and contexts of use, which is the typical scenario for Semantic Web Data applications. The evolution of databases in the direction of heterogeneous data environments strongly impacts the usability, semiotics and semantic assumptions behind existing data accessibility methods such as structured queries, keyword-based search and visual query systems. With schema-less databases containing potentially millions of dynamically changing attributes, it becomes unfeasible for some users to become aware of the 'schema' or vocabulary in order to query the database. At this scale, the effort in understanding the schema in order to build a structured query can become prohibitive. == Schema-agnostic queries == Schema-agnostic queries can be defined as query approaches over structured databases which allow users satisfying complex information needs without the understanding of the representation (schema) of the database. Similarly, Tran et al. defines it as "search approaches, which do not require users to know the schema underlying the data". Approaches such as keyword-based search over databases allow users to query databases without employing structured queries. However, as discussed by Tran et al.: "From these points, users however have to do further navigation and exploration to address complex information needs. Unlike keyword search used on the Web, which focuses on simple needs, the keyword search elaborated here is used to obtain more complex results. Instead of a single set of resources, the goal is to compute complex sets of resources and their relations." The development of approaches to support natural language interfaces (NLI) over databases have aimed towards the goal of schema-agnostic queries. Complementarily, some approaches based on keyword search have targeted keyword-based queries which express more complex information needs. Other approaches have explored the construction of structured queries over databases where schema constraints can be relaxed. All these approaches (natural language, keyword-based search and structured queries) have targeted different degrees of sophistication in addressing the problem of supporting a flexible semantic matching between queries and data, which vary from the completely absence of the semantic concern to more principled semantic models. While the demand for schema-agnosticism has been an implicit requirement across semantic search and natural language query systems over structured data, it is not sufficiently individuated as a concept and as a necessary requirement for contemporary database management systems. Recent works have started to define and model the semantic aspects involved on schema-agnostic queries. === Schema-agnostic structured queries === Consist of schema-agnostic queries following the syntax of a structured standard (for example SQL, SPARQL). The syntax and semantics of operators are maintained, while different terminologies are used. ==== Example 1 ==== SELECT ?y { BillClinton hasDaughter ?x . ?x marriedTo ?y . } which maps to the following SPARQL query in the dataset vocabulary: ==== Example 2 ==== which maps to the following SPARQL query in the dataset vocabulary: === Schema-agnostic keyword queries === Consist of schema-agnostic queries using keyword queries. In this case the syntax and semantics of operators are different from the structured query syntax. ==== Example ==== "Bill Clinton daughter married to" "Books by William Goldman with more than 300 pages" == Semantic complexity == As of 2016 the concept of schema-agnostic queries has been developed primarily in academia. Most of schema-agnostic query systems have been investigated in the context of Natural Language Interfaces over databases or over the Semantic Web. These works explore the application of semantic parsing techniques over large, heterogeneous and schema-less databases. More recently, the individuation of the concept of schema-agnostic query systems and databases have appeared more explicitly within the literature. Freitas et al. provide a probabilistic model on the semantic complexity of mapping schema-agnostic queries.
Type–token distinction
The type–token distinction is the difference between a type of objects (analogous to a class) and the individual tokens of that type (analogous to instances). Since each type may be instantiated by multiple tokens, there are generally more tokens than types of an object. For example, the sentence "A rose is a rose is a rose" contains three word types: three word tokens of the type a, two word tokens of the type is, and three word tokens of the type rose. The distinction is important in disciplines such as logic, linguistics, metalogic, typography, and computer programming. == Overview == The type–token distinction separates types (abstract descriptive concepts) from tokens (objects that instantiate concepts). For example, in the sentence "the bicycle is becoming more popular" the word bicycle represents the abstract concept of bicycles and this abstract concept is a type, whereas in the sentence "the bicycle is in the garage", it represents a particular object and this particular object is a token. Similarly, the word type 'letter' uses only four letter types: L, E, T and R. Nevertheless, it uses both E and T twice. One can say that the word type 'letter' has six letter tokens, with two tokens each of the letter types E and T. Whenever a word type is inscribed, the number of letter tokens created equals the number of letter occurrences in the word type. Some logicians consider a word type to be the class of its tokens. Other logicians counter that the word type has a permanence and constancy not found in the class of its tokens. The type remains the same while the class of its tokens is continually gaining new members and losing old members. == Typography == In typography, the type–token distinction is used to determine the presence of a text printed by movable type: The defining criteria which a typographic print has to fulfill is that of the type identity of the various letter forms which make up the printed text. In other words: each letter form which appears in the text has to be shown as a particular instance ("token") of one and the same type which contains a reverse image of the printed letter. == Charles Sanders Peirce == The distinctions between using words as types or tokens were first made by American logician and philosopher Charles Sanders Peirce in 1906 using terminology that he established. Peirce's type–token distinction applies to words, sentences, paragraphs and so on: to anything in a universe of discourse of character-string theory, or concatenation theory. Peirce's original words are the following: A common mode of estimating the amount of matter in a ... printed book is to count the number of words. There will ordinarily be about twenty 'thes' on a page, and, of course, they count as twenty words. In another sense of the word 'word,' however, there is but one word 'the' in the English language; and it is impossible that this word should lie visibly on a page, or be heard in any voice .... Such a ... Form, I propose to term a Type. A Single ... Object ... such as this or that word on a single line of a single page of a single copy of a book, I will venture to call a Token. .... In order that a Type may be used, it has to be embodied in a Token which shall be a sign of the Type, and thereby of the object the Type signifies.
Aidan Gomez
Aidan Gomez is a British-Canadian computer scientist working in the field of artificial intelligence, with a focus on natural language processing. He is the co-founder and CEO of the technology company Cohere. == Early life and education == Gomez grew up in Brighton, Ontario. He graduated from the University of Toronto with a bachelor's degree in computer science and mathematics. He was pursuing a PhD in computer science from the University of Oxford. He paused his studies to launch Cohere. He was granted the PhD in 2024. == Career == In 2017, as a 20 year-old intern at Google Brain, Gomez was one of eight authors of the research paper "Attention Is All You Need", which is credited with changing the AI industry and helping lead to the creation of ChatGPT. The paper proposed a novel deep learning architecture called the transformer, that enables machine learning models to analyze large amounts of data for patterns, and then use those patterns to make predictions while leveraging GPU parallelization. It has been commonly adopted for training large language models and in the development of generative AI. In the same year, Gomez founded FOR.ai, a program to help researchers learn machine learning techniques in a collaborative format. An outgrowth of this project was Cohere For AI (now Cohere Labs), which released Aya, an open-source multilingual LLM. As a PhD student, Gomez worked as a machine learning researcher at Google Brain. At that time, he co-authored the paper "One Model to Learn Them All" about multi-task learning by a single neural network. In 2019, Gomez left Google Brain to launch Cohere, an enterprise-focused company that helps businesses implement AI into chatbots, search engines, and other products. As of Sept 2025, Cohere has raised about US$1.6 billion at valuation north of $7 billion, as Gomez leads the company as its CEO. Gomez was named to the 2023 Time 100/AI list of the most influential people in the field of artificial intelligence. He and his fellow Cohere founders Ivan Zhang and Nick Frosst were named number 1 on 2023 Maclean's AI Trailblazers Power List. In April 2025, Gomez was elected to the board of Rivian. == Views on AI == Gomez has stated that warnings regarding the existential risk from artificial intelligence are overblown, and that real risks involve the automated spread of misinformation on social media. He said that the United States would win the AI arms race over China.
Procedural reasoning system
In artificial intelligence, a procedural reasoning system (PRS) is a framework for constructing real-time reasoning systems that can perform complex tasks in dynamic environments. It is based on the notion of a rational agent or intelligent agent using the belief–desire–intention software model. A user application is predominately defined, and provided to a PRS system is a set of knowledge areas. Each knowledge area is a piece of procedural knowledge that specifies how to do something, e.g., how to navigate down a corridor, or how to plan a path (in contrast with robotic architectures where the programmer just provides a model of what the states of the world are and how the agent's primitive actions affect them). Such a program, together with a PRS interpreter, is used to control the agent. The interpreter is responsible for maintaining beliefs about the world state, choosing which goals to attempt to achieve next, and choosing which knowledge area to apply in the current situation. How exactly these operations are performed might depend on domain-specific meta-level knowledge areas. Unlike traditional AI planning systems that generate a complete plan at the beginning, and replan if unexpected things happen, PRS interleaves planning and doing actions in the world. At any point, the system might only have a partially specified plan for the future. PRS is based on the BDI or belief–desire–intention framework for intelligent agents. Beliefs consist of what the agent believes to be true about the current state of the world, desires consist of the agent's goals, and intentions consist of the agent's current plans for achieving those goals. Furthermore, each of these three components is typically explicitly represented somewhere within the memory of the PRS agent at runtime, which is in contrast to purely reactive systems, such as the subsumption architecture. == History == The PRS concept was developed by the Artificial Intelligence Center at SRI International during the 1980s, by many workers including Michael Georgeff, Amy L. Lansky, and François Félix Ingrand. Their framework was responsible for exploiting and popularizing the BDI model in software for control of an intelligent agent. The seminal application of the framework was a fault detection system for the reaction control system of the NASA Space Shuttle Discovery. Development on this PRS continued at the Australian Artificial Intelligence Institute through to the late 1990s, which led to the development of a C++ implementation and extension called dMARS. == Architecture == The system architecture of SRI's PRS includes the following components: Database for beliefs about the world, represented using first order predicate calculus. Goals to be realized by the system as conditions over an interval of time on internal and external state descriptions (desires). Knowledge areas (KAs) or plans that define sequences of low-level actions toward achieving a goal in specific situations. Intentions that include those KAs that have been selected for current and eventual execution. Interpreter or inference mechanism that manages the system. == Features == SRI's PRS was developed for embedded application in dynamic and real-time environments. As such it specifically addressed the limitations of other contemporary control and reasoning architectures like expert systems and the blackboard system. The following define the general requirements for the development of their PRS: asynchronous event handling guaranteed reaction and response types procedural representation of knowledge handling of multiple problems reactive and goal-directed behavior focus of attention reflective reasoning capabilities continuous embedded operation handling of incomplete or inaccurate data handling of transients modeling delayed feedback operator control == Applications == The seminal application of SRI's PRS was a monitoring and fault detection system for the reaction control system (RCS) on the NASA space shuttle. The RCS provides propulsive forces from a collection of jet thrusters and controls altitude of the space shuttle. A PRS-based fault diagnostic system was developed and tested using a simulator. It included over 100 KAs and over 25 meta level KAs. RCS specific KAs were written by space shuttle mission controllers. It was implemented on the Symbolics 3600 Series LISP machine and used multiple communicating instances of PRS. The system maintained over 1000 facts about the RCS, over 650 facts for the forward RCS alone and half of which are updated continuously during the mission. A version of the PRS was used to monitor the reaction control system on the Space Shuttle Discovery. PRS was tested on Shakey the robot including navigational and simulated jet malfunction scenarios based on the space shuttle. Later applications included a network management monitor called the Interactive Real-time Telecommunications Network Management System (IRTNMS) for Telecom Australia. == Extensions == The following list the major implementations and extensions of the PRS architecture. UM-PRS OpenPRS (formerly C-PRS and Propice) AgentSpeak Distributed multi-agent reasoning system (dMARS) GORITE JAM JACK Intelligent Agents SRI Procedural Agent Realization Kit (SPARK) PRS-CL
PagedAttention
PagedAttention is an attention algorithm for efficient serving of large language models (LLMs). It was introduced in 2023 by Woosuk Kwon and colleagues in the paper Efficient Memory Management for Large Language Model Serving with PagedAttention, alongside the vLLM serving engine. The method stores the key–value cache used during autoregressive decoding in fixed-size blocks that can be mapped to non-contiguous physical memory, borrowing ideas from virtual memory, paging, and operating system design. == Background == In transformer inference, the key–value cache grows with sequence length and the number of concurrent requests. Kwon et al. argued that earlier serving systems typically reserved contiguous cache regions in advance, which caused reserved space, internal fragmentation, and external fragmentation. In their experiments, the paper reported that the effective memory utilization of previous systems could fall as low as 20.4%. == Description == PagedAttention partitions the cache of each sequence into fixed-size KV blocks. A request's cache is represented as a sequence of logical blocks, while a block table maps those logical blocks to physical GPU-memory blocks. As a result, neighboring logical blocks do not need to be contiguous in physical memory, and new blocks can be allocated on demand as generation proceeds. The design also makes it easier to share cache state across related decoding paths. In vLLM, physical blocks can be reference-counted and shared among requests or branches, with block-granularity copy-on-write used when a shared block must be modified. The original paper applied this design to parallel sampling, beam search, and prompts with shared prefixes. == Mathematical formulation == For a query token i {\displaystyle i} in causal self-attention, the standard attention output can be written as a i j = exp ( q i ⊤ k j / d ) ∑ t = 1 i exp ( q i ⊤ k t / d ) , o i = ∑ j = 1 i a i j v j {\displaystyle a_{ij}={\frac {\exp(\mathbf {q} _{i}^{\top }\mathbf {k} _{j}/{\sqrt {d}})}{\sum _{t=1}^{i}\exp(\mathbf {q} _{i}^{\top }\mathbf {k} _{t}/{\sqrt {d}})}},\;\mathbf {o} _{i}=\sum _{j=1}^{i}a_{ij}\mathbf {v} _{j}} where q i {\displaystyle \mathbf {q} _{i}} , k j {\displaystyle \mathbf {k} _{j}} , and v j {\displaystyle \mathbf {v} _{j}} are the query, key, and value vectors, and d {\displaystyle d} is the attention dimension. If the cache is partitioned into blocks of size B {\displaystyle B} , the key and value blocks may be written as K j = ( k ( j − 1 ) B + 1 , … , k j B ) , V j = ( v ( j − 1 ) B + 1 , … , v j B ) {\displaystyle \mathbf {K} _{j}=(\mathbf {k} _{(j-1)B+1},\ldots ,\mathbf {k} _{jB}),\;\mathbf {V} _{j}=(\mathbf {v} _{(j-1)B+1},\ldots ,\mathbf {v} _{jB})} PagedAttention then performs the computation blockwise: A i j = exp ( q i ⊤ K j / d ) ∑ t = 1 ⌈ i / B ⌉ exp ( q i ⊤ K t / d ) , o i = ∑ j = 1 ⌈ i / B ⌉ V j A i j ⊤ {\displaystyle \mathbf {A} _{ij}={\frac {\exp(\mathbf {q} _{i}^{\top }\mathbf {K} _{j}/{\sqrt {d}})}{\sum _{t=1}^{\lceil i/B\rceil }\exp(\mathbf {q} _{i}^{\top }\mathbf {K} _{t}/{\sqrt {d}})}},\;\mathbf {o} _{i}=\sum _{j=1}^{\lceil i/B\rceil }\mathbf {V} _{j}\mathbf {A} _{ij}^{\top }} where A i j {\displaystyle \mathbf {A} _{ij}} is the vector of attention scores for the j {\displaystyle j} -th KV block. In the formulation given by Kwon et al., this preserves the causal attention calculation while allowing the key and value blocks to reside in non-contiguous physical memory. == Performance and use == The vLLM paper reported that, on its evaluated workloads, the use of PagedAttention and the associated memory-management design improved serving throughput by 2–4× over the compared baselines, including FasterTransformer and Orca, while preserving model outputs. In experiments on OPT-13B with the Alpaca trace, the paper also reported memory savings of 6.1–9.8% for parallel sampling and 37.6–55.2% for beam search through KV-block sharing. A 2024 survey of LLM serving systems described PagedAttention as having become an industry norm in LLM serving frameworks, citing support in TGI, vLLM, and TensorRT-LLM. == Limitations and alternatives == Subsequent work has described trade-offs in the approach. The 2025 vAttention paper argued that PagedAttention requires attention kernels to be rewritten to support paging and increases software complexity, portability issues, redundancy, and execution overhead, proposing instead a memory manager that keeps the cache contiguous in virtual memory while relying on demand paging for physical allocation. === vAttention === Unlike PagedAttention, vAttention does not introduce a different attention rule; it retains the standard attention computation Attention ( q i , K , V ) = softmax ( q i K ⊤ s c a l e ) V . {\displaystyle \operatorname {Attention} (q_{i},K,V)=\operatorname {softmax} \left({\frac {q_{i}K^{\top }}{\mathrm {scale} }}\right)V.} In the notation of Prabhu et al., the key and value tensors for a request seen so far are K , V ∈ R L ′ × ( H × D ) {\displaystyle K,V\in \mathbb {R} ^{L'\times (H\times D)}} , where L ′ {\displaystyle L'} is the context length seen so far, H {\displaystyle H} is the number of KV heads on a worker, and D {\displaystyle D} is the dimension of each KV head. In systems prior to PagedAttention, the K cache (or V cache) at each layer of a worker is typically allocated as a 4D tensor of shape [ B , L , H , D ] , {\displaystyle [B,L,H,D],} where B {\displaystyle B} is batch size and L {\displaystyle L} is the maximum context length supported by the model. vAttention preserves this contiguous virtual-memory view while deferring physical-memory allocation to runtime. A serving framework maintains separate K and V tensors for each layer, so vAttention reserves 2 N {\displaystyle 2N} virtual-memory buffers on a worker, where N {\displaystyle N} is the number of layers managed by that worker. The maximum size of one virtual-memory buffer is B S = B × S , {\displaystyle BS=B\times S,} where S {\displaystyle S} is the maximum size of a single request's per-layer K cache (or V cache) on a worker. The paper defines S = L × H × D × P , {\displaystyle S=L\times H\times D\times P,} where P {\displaystyle P} is the number of bytes needed to store one element. In this formulation, vAttention keeps the KV cache contiguous in virtual memory and relies on demand paging for physical allocation, rather than modifying the attention kernel to operate over non-contiguous KV-cache blocks.
Diffbot
Diffbot is a developer of machine learning and computer vision algorithms and public APIs for extracting data from web pages / web scraping to create a knowledge base. == Overview == The company has gained interest from its application of computer vision technology to web pages, wherein it visually parses a web page for important elements and returns them in a structured format. In 2015 Diffbot announced it was working on its version of an automated "knowledge graph" by crawling the web and using its automatic web page extraction to build a large database of structured web data. In 2019 Diffbot released their Knowledge Graph which has since grown to include over two billion entities (corporations, people, articles, products, discussions, and more), and ten trillion "facts." == Features == The company's products allow software developers to analyze web home pages and article pages, and extract the "important information" while ignoring elements deemed not core to the primary content. In August 2012 the company released its Page Classifier API, which automatically categorizes web pages into specific "page types". As part of this, Diffbot analyzed 750,000 web pages shared on the social media service Twitter and revealed that photos, followed by articles and videos, are the predominant web media shared on the social network. In September 2020 the company released a Natural Language Processing API for automatically building Knowledge Graphs from text. The company raised $2 million in funding in May 2012 from investors including Andy Bechtolsheim and Sky Dayton. Diffbot's customers include Adobe, AOL, Cisco, DuckDuckGo, eBay, Instapaper, Microsoft, Onswipe and Springpad.