The Webby Awards (colloquially referred to as the Webbys) are awards for excellence on the Internet presented annually by the International Academy of Digital Arts and Sciences, a judging body composed of over three thousand industry experts and technology innovators. Categories include websites, advertising and media, online film and video, mobile sites and apps, and social. Two winners are selected in each category, one by members of The International Academy of Digital Arts and Sciences, and one by the public who cast their votes during Webby People's Voice voting. Each winner presents a five-word acceptance speech, a trademark of the annual awards show. In its early years, the award was hailed as the "Internet's highest honor" and was associated with the phrase "The Oscars of the Internet." == History == In its early years, the organization was one of several vying to be the premiere internet awards show. Both shows would compare themselves to the Oscars, as did media outlets such as The New York Times to Canada's Globe & Mail. The winners of the First Annual Webby Awards in 1995 were presented by John Brancato and Michael Ferris, writers for Columbia Pictures. It was held at the Hollywood Roosevelt Hotel. The televised Webby Awards were sponsored by the Academy of Web Design and Cool Site of the Day. The first Webby Awards were produced by Kay Dangaard at the Hollywood Roosevelt Hotel as a nod to the first site of the Academy of Motion Picture Arts and Sciences (Oscars). That first year, they were called "Webbie" Awards. The first "Site of the Year" winner was the pioneer webisodic serial The Spot. The modern Webby Awards were co-founded by Tiffany Shlain, a filmmaker, when she was hired by The Web Magazine to re-establish them, and were first held in San Francisco in 1997. They quickly became known for its requirement that winners give their acceptance speeches in five words. After this, the awards became more successful than the magazine and IDG closed the publication. Shlain and co-founder Maya Draisin Farrah continued to run The Webby Awards until 2004. The International Academy of Digital Arts and Sciences, which selects the winners of The Webby Awards, was established in 1998 by co-founders Tiffany Shlain, Spencer Ante and Maya Draisin. Members of the Academy include Kevin Spacey, Grimes, Questlove, Internet inventor Vint Cerf, Instagram's Head of Fashion Partnerships Eva Chen, comedian Jimmy Kimmel, Twitter founder Biz Stone, Vice Media co-founder and CEO Shane Smith, Tumblr's David Karp, Director of Harvard's Berkman Klein Center for Internet & Society Susan P. Crawford, Refinery29's Executive Creative Director Piera Gelardi, and CEO and co-founder of Gimlet Media Alex Blumberg. The Webby Awards is owned and operated by the Webby Media Group, a division of Recognition Media, which also owns and produces the Lovie Awards in Europe and Netted by the Webbys, a daily email publication launched in 2009. David-Michel Davies, CEO of Webby Media Group, current Executive Director of the Webby Awards and co-founder of Internet Week New York, was named Executive Director of the Webby Awards in 2005. In 2009, the 13th Annual Webby Awards received nearly 10,000 entries from all 50 US states and over 60 countries. That same year, more than 500,000 votes were cast in The Webby People's Voice Awards. In 2012, the 16th Annual Webby awards received 1.5 million votes from more than 200 countries for the People's Voice awards. In 2015, the 19th Annual Webby Awards received nearly 13,000 entries from all 50 U.S. states and over 60 countries worldwide. == Nomination process == The 2000 awards began the transition to nominee submissions. Previously, nominees had been selected by an internal committee. As early as 2017, organizations wanting to nominate themselves were charged $395 for a single entry. An "ad campaign entry" would cost $595. By 2024, those fees had risen to $495 and $675, respectively. Executive Academy Members with category-specific expertise evaluate the shortlisted entries based on the appropriate Website, Advertising & Media, Online Film & Video, Mobile Sites & Apps, and Social category criteria, and cast ballots to determine Webby Honorees, Nominees and Webby Winners. Deloitte provides vote tabulation consulting for the Webby Awards. In addition to the award given in each category by the International Academy of Digital Arts and Sciences, another winner is selected in each category as determined by the general public during People's Voice voting. Winners of both the Academy-selected and People's Voice-selected awards are invited to the Webbys. == Awards granted == The Webby Awards are presented in over a hundred categories among all four types of entries. A website can be entered in multiple categories and receive multiple awards. In each category, two awards are handed out: a Webby Award selected by The International Academy of Digital Arts and Sciences, and a People's Voice Award selected by the general public. == Ceremony == Between 2005 and 2019, the Webby Awards were presented in New York City. Many of the ceremony hosts are comedians and comedic actors. Comedian Rob Corddry hosted the ceremony from 2005 to 2007. Seth Meyers of Saturday Night Live hosted in 2008 and 2009, B.J. Novak of the sitcom The Office in 2010, and Lisa Kudrow in 2011. Comedian, actor, and writer Patton Oswalt hosted from 2012 to 2014. Comedian Hannibal Buress hosted in 2015. The Webbys are famous for limiting recipients to five-word speeches, which are often humorous, although some exceed the limit. In 2005 when accepting his Lifetime Achievement Award, former Vice President Al Gore's speech was "Please don't recount this vote." He was introduced by Vint Cerf who used the same format to state, "We all invented the Internet." In 2013, the creator of the Graphics Interchange Format (GIF), Steve Wilhite, accepted his Webby and delivered his now famous five-word speech, "It's pronounced 'Jif' not 'Gif'." == Criticism == The Webbys have been criticized for their pay-to-enter and pay-to-attend policy (winners and nominees also have to pay to attend the award ceremony), and thus for not taking most websites into consideration before distributing their awards. Gawker, its Valleywag column, and others, have called the awards a scam, with Valleywag saying, "...somewhere along the way, the organizers figured out that this goofy charade could be milked for profit." In response, Webby Awards executive director David-Michel Davies told the Wall Street Journal that entry fees "provide the best and most sustainable model for ensuring that our judging process remains consistent and rigorous and is not dependent on things like sponsorships that can fluctuate from year to year." == Anthem Awards == In 2021, the Webby organization started a new line of awards, the Anthem Awards, to honor the purpose and mission-driven work of people, companies and organizations worldwide. The finalists and winners are selected by the International Academy of Digital Arts and Sciences.
Acoustic model
An acoustic model is used in automatic speech recognition to represent the relationship between an audio signal and the phonemes or other linguistic units that make up speech. The model is learned from a set of audio recordings and their corresponding transcripts. It is created by taking audio recordings of speech, and their text transcriptions, and using software to create statistical representations of the sounds that make up each word. == Background == Modern speech recognition systems use both an acoustic model and a language model to represent the statistical properties of speech. The acoustic model models the relationship between the audio signal and the phonetic units in the language. The language model is responsible for modeling the word sequences in the language. These two models are combined to get the top-ranked word sequences corresponding to a given audio segment. Most modern speech recognition systems operate on the audio in small chunks known as frames with an approximate duration of 10ms per frame. The raw audio signal from each frame can be transformed by applying the mel-frequency cepstrum. The coefficients from this transformation are commonly known as mel-frequency cepstral coefficients (MFCCs) and are used as an input to the acoustic model along with other features. Recently, the use of convolutional neural networks has led to major improvements in acoustic modeling. == Speech audio characteristics == Audio can be encoded at different sampling rates (i.e. samples per second – the most common being: 8, 16, 32, 44.1, 48, and 96 kHz), and different bits per sample (the most common being: 8-bits, 16-bits, 24-bits or 32-bits). Speech recognition engines work best if the acoustic model they use was trained with speech audio which was recorded at the same sampling rate/bits per sample as the speech being recognized. == Telephony-based speech recognition == The limiting factor for telephony based speech recognition is the bandwidth at which speech can be transmitted. For example, a standard land-line telephone only has a bandwidth of 64 kbit/s at a sampling rate of 8 kHz and 8-bits per sample (8000 samples per second 8-bits per sample = 64000 bit/s). Therefore, for telephony based speech recognition, acoustic models should be trained with 8 kHz/8-bit speech audio files. In the case of voice over IP, the codec determines the sampling rate/bits per sample of speech transmission. Codecs with a higher sampling rate/bits per sample for speech transmission (which improve the sound quality) necessitate acoustic models trained with audio data that matches that sampling rate/bits per sample. == Desktop-based speech recognition == For speech recognition on a standard desktop PC, the limiting factor is the sound card. Most sound cards today can record at sampling rates of between 16–48 kHz of audio, with bit rates of 8- to 16-bits per sample, and playback at up to 96 kHz. As a general rule, a speech recognition engine works better with acoustic models trained with speech audio data recorded at higher sampling rates/bits per sample. But using audio with too high a sampling rate/bits per sample can slow the recognition engine down. A compromise is needed. Thus for desktop speech recognition, the current standard is acoustic models trained with speech audio data recorded at sampling rates of 16 kHz/16 bits per sample.
Coreu
COREU (French: Correspondance Européenne – Telex network of European correspondents, also EUKOR-Netzwerk in Austria) is a communication network of the European Union for the communication of the Council of the European Union, the European correspondents of the foreign ministries of the EU member states, permanent representatives of member states in Brussels, the European Commission, and the General Secretariat of the Council of the European Union. The European Parliament is not among the participants. COREU is the European equivalent of the American Secret Internet Protocol Router Network (SIPRNet, also known as Intelink-S). COREU's official aim is fast communication in case of crisis. The network enables a closer cooperation in matters regarding foreign affairs. In actuality the system's function exceeds that of mere communication, it also enables decision-making. COREU's first goal is to enable the exchange of information before and after decisions. Relaying upfront negotiations in preparation of meetings is the second goal. In addition, the system also allows the editing of documents and the decision-making, especially if there is little time. While the first two goals are preparatory measures for a shared foreign policy, the third is a methodical variant marked by practise that is defining for the image of the Common Foreign and Security Policy. == Members == (The following information dates from 2013): There is one representative in each of the capital cities in the EU.(since 1973) In Germany for example, this is the European correspondent (EU-KOR) from the Foreign Office. In Austria it is the European correspondent from the Referat II.1.a in the Federal Ministry for Europe, Integration and Foreign Affairs They are the correspondents (since 1982) for the European Commission They comprise the secretariat for the European Council They also make up the European External Action Service (EEAS) (responsible for foreign policy issues, since 1987) == Data volume and technical details == COREU functions as a spoke-hub distribution paradigm system with the hub in Brussels. The network is operated by the European Union Intelligence and Situation Centre (formerly Joint Situation Center, JSC). The technical infrastructure is located in a building of the European Council. COREU may be described as an advanced telex system with encrypted messages via dedicated terminals. Once a message has reached the destination, it is then redistributed via the local media. In contrast, messages of governments are transmitted via local media to the correspondents and from there delivered point-to-point to Brussels via COREU. In 2010, approximately 8500 communications had been distributed over this network. == History == A telex-based communication system under the name COREU was established in 1973. Originally, only the ministries of Foreign Affairs in the European capitals were connected to it. This telex system was replaced in 1997 by the mail system CORTESY (COREU Terminal Equipment System). The name was retained despite the technical innovation. COREU was reportedly compromised by hackers working for the People's Liberation Army Strategic Support Force, allowing for the theft of thousands of low-classified documents and diplomatic cables.
ISO 15765-2
ISO 15765-2, or ISO-TP (Transport Layer), is an international standard for sending data packets over a CAN bus. The protocol allows for the transport of messages that exceed the eight byte maximum payload of CAN frames. ISO-TP segments longer messages into multiple frames, adding metadata (CAN-TP Header) that allows the interpretation of individual frames and reassembly into a complete message packet by the recipient. It can carry up to 232-1 (4294967295) bytes of payload per message packet starting from the 2016 version. Prior versions were limited to a maximum payload size of 4095 bytes. In the OSI model, ISO-TP covers the layer 3 (network layer) and 4 (transport layer). The most common application for ISO-TP is the transfer of diagnostic messages with OBD-II equipped vehicles using KWP2000 and UDS, but is used broadly in other application-specific CAN implementations where one might need to send messages longer than what the CAN protocol physical layer allows (eight bytes for CAN, 64 bytes for CAN FD, and 2048 bytes for CAN-XL). ISO-TP can be operated with its own addressing as so-called Extended Addressing or without address using only the CAN ID (so-called Normal Addressing). Extended addressing uses the first data byte of each frame as an additional element of the address, reducing the application payload by one byte. For clarity the protocol description below is based on Normal Addressing with eight byte CAN frames. In total, six types of addressing are allowed by the ISO 15765-2 Protocol. ISO-TP prepends one or more metadata bytes to the payload data in the eight byte CAN frame, reducing the payload to seven or fewer bytes per frame. The metadata is called the Protocol Control Information, or PCI. The PCI is one, two or three bytes. The initial field is four bits indicating the frame type, and implicitly describing the PCI length. ISO 15765-2 is a part of ISO 15765 (headlined Road vehicles — Diagnostic communication over Controller Area Network (DoCAN)), which has the following parts: ISO 15765-1 Part 1: General information and use case definition ISO 15765-2 Part 2: Transport protocol and network layer services ISO 15765-3 Part 3: Implementation of unified diagnostic services (UDS on CAN) – replaced by ISO 14229-3 Road vehicles — Unified diagnostic services ISO 15765-4 Part 4: Requirements for emissions-related systems == List of protocol control information (PCI) field types == The ISO-TP defines four frame types: A message of seven bytes or less is sent in a single frame, with the initial byte containing the type (0) and payload length (1-7 bytes). With the 0 in the type field, this can also pass as a simpler protocol with a length-data format and is often misinterpreted as such. A message longer than 7 bytes requires segmenting the message packet over multiple frames. A segmented transfer starts with a First Frame. The PCI is two bytes in this case, with the first 4 bit field the type (type 1) and the following 12 bits the message length (excluding the type and length bytes). The recipient confirms the transfer with a flow control frame. The flow control frame has three PCI bytes specifying the interval between subsequent frames and how many consecutive frames may be sent (Block Size). For CAN FD, the ISO 15765-2 protocol has been extended for Single and First frame, to allow larger size values, but still backwards compatible with traditional ISO 15765. See CAN FD. The initial byte contains the type (type = 3) in the first four bits, and a flag in the next four bits indicating if the transfer is allowed (0 = Continue To Send, 1 = Wait, 2 = Overflow/abort). The next byte is the block size, the count of frames that may be sent before waiting for the next flow control frame. A value of zero allows the remaining frames to be sent without flow control or delay. The third byte is the minimum Separation Time (STmin), the minimum delay time between frames. STmin values up to 127 (0x7F) specify the minimum number of milliseconds to delay between frames, while values in the range 241 (0xF1) to 249 (0xF9) specify delays increasing from 100 to 900 microseconds. Note that the Separation Time is defined as the minimum time between the end of one frame to the beginning of the next. Robust implementations should be prepared to accept frames from a sender that misinterprets this as the frame repetition rate i.e. from start-of-frame to start-of-frame. Even careful implementations may fail to account for the minor effect of bit-stuffing in the physical layer. The sender transmits the rest of the message using Consecutive Frames. Each Consecutive Frame has a one byte PCI, with a four bit type (type = 2) followed by a 4-bit sequence number. The sequence number starts at 1 and increments with each frame sent (1, 2,..., F, 0, 1,...), with which lost or discarded frames can be detected. Each consecutive frame starts at 0, initially for the first set of data in the first frame will be considered as 0th data. So the first set of CF(Consecutive frames) start from 0x1. There afterwards when it reaches 0x2F, will be started from 0x20 (e.g. 0x21, 0x22, 0x23...0x2F, 0x20, 0x21...). The 12-bit length field (as indicated in the First Frame) allows up to 4095 bytes of user data in a segmented message, but in practice the typical application-specific limit is considerably lower because of receive buffer or hardware limitations. == Timing parameters == Timing parameters, such as P1 and P2 timers, have to be mentioned. == Standards == ISO 15765-2:2016 Road vehicles -- Diagnostic communication over Controller Area Network (DoCAN) -- Part 2: Transport protocol and network layer services
IBM remote batch terminals
The IBM 2780 and the IBM 3780 are devices developed by IBM for performing remote job entry (RJE) and other batch functions over telephone lines; they communicate with the mainframe via Binary Synchronous Communications (BSC or Bisync) and replaced older terminals using synchronous transmit-receive (STR). In addition, IBM has developed workstation programs for the 1130, 360/20, 2922, System/360 other than 360/20, System/370 and System/3. == 2780 Data Transmission Terminal == The 2780 Data Transmission Terminal first shipped in 1967. It consists of: A line printer similar to the IBM 1443 that can print up to 240 lines per minute (lpm), or 300 lpm using an extremely restricted character set. A card reader/punch unit, similar to an IBM 1442, that can read up to 400 cards per minute (cpm) and can punch up to 355 cpm. A line buffer that stores data received or to be transmitted over the communications line. A binary synchronous adapter which controls the flow of data over the communications line. The 2780 is capable of local (offline) card to print operation. It comes in four models: Model 1: Can read punched cards and transmit the data to a remote host computer, and can receive and print data sent by the host. Model 2: Same as Model 1 but adds the ability to punch card data received from the host. Model 3: Can only print data received from the host, but not send data to it. Model 4: Can read and punch card data, but has no printing capabilities. The 2780 uses a dedicated communication line at speeds of 1200, 2000, 2400 or 4800 bits per second. It is a half duplex device, although full duplex lines can be used with some increase in throughput. It can communicate in Transcode (a 6-bit code), 8-bit EBCDIC, or 7-bit ASCII. == 2770 Data Communication System == The 2770, announced in 1969, "was said to surpass all other IBM terminals in the variety of available input-output devices." The 2770 was developed by the IBM General Products Division (GPD) in Rochester, MN. It comes standard with a desktop terminal with keyboard. The printer and other devices (any two in any combination) can be attached to the 2772 Multi-Purpose Control unit. Possible devices include: 50 Magnetic Data Inscriber 545 Card Punch Model 3 (non-printing) or Model 4 (printing) 1017 Paper Tape Reader 1018 Paper Tape Punch 1053 Printer Model 1 1255 Magnetic Character Reader Models 1, 2 or 3 2203 Printer Model A1 or A2 2213 Printer Model 1 or 2 2265 Display Station Model 2 2502 Card Reader Model A1 or A2 5496 Data Recorder == 3780 Data Communications Terminal == In May 1972, IBM announced the IBM 3780, an enhanced version of the 2780. The 3780 was developed by IBM's Data Processing Division (DPD). There is one model, with an optional card punch. The 3780 drops Transcode support and incorporates several performance enhancements. It supports compression of blank fields in data using run-length encoding. It provides the ability to interleave data between devices, introduces double buffering, and adds support for the Wait-before-transmit ACKnowledgement (WACK) and Temporary Text Delay (TTD) Binary Synchronous control characters. The integrated punched card unit can read cards at 600 cards per minute. The integrated printer is rated at 300, 350 or 425 lines per minute based on characters set (63, 52 or 39 characters). The 3781 Card Punch is an optional feature. It punches 160 columns per second, or 91 cards per minute if all 80 columns are punched. The IBM 2780 and 3780 were later emulated on various types of equipment, including eventually the personal computer. A notable early emulation was the DN60, by Digital Equipment Corporation in the late 1970s. == 3770 Data Communications System == In 1974 IBM Data Processing Division (DPD) offered a successor to the 3780, called the 3770 Data Communications System, supporting SDLC, BSC, BSC Multi-leaving and SNA, depending on the configuration. The 3770 is a family of desk console style terminals that offers a variety of keyboard and printer combinations as well as I/O equipment attachment and communications features. The terminals come built into a desk and include the following models: 3771 Communication Terminal (optional card reader, optional card punch, wire matrix printer) Models 1 (40 cps printer), 2 (80 cps printer), and 3 (120 cps printer). 3773 Communication Terminal (diskette, wire matrix printer) Models 1 (40 cps printer), 2 (80 cps printer), and 3 (120 cps printer). Each model has a P version which adds some programming features. 3774 Communication Terminal (optional card reader, optional card punch, optional belt printer, wire matrix printer) Models 1 (80 cps printer), and 2 (120 cps printer). Each model has a P version which adds some programming features, a 480-character display and a non-removable diskette. 3775 Communication Terminal (optional card reader, optional card punch, optional diskette, belt printer) Model 1 (120 lpm printer). The model P1 adds some programming features, a 480-character display and a non-removable diskette. 3776 Communication Terminal (optional card reader, optional card punch, optional diskette, belt printer) Models 1 (300 lpm printer) and 2 (400 lpm printer). Models 3 and 4 are similar to models 1 and 2. 3777 Communication Terminal (optional card reader, optional diskette, train printer) Model 1 (up to 1000 lpm printer depending on character set). Model 2 adds an optional card punch, model 3 adds an optional magnetic tape drive and model 4 replaces the train printer with a slower model called the IBM 3262. The model 4 also allows a second, optional, 3262. The following I/O devices can be attached to a 3770 terminal: IBM 2502 Card Reader: Models A1 (up to 150 card per minute), A2 (up to 300 cards per minute) or A3 (up to 400 cards per minute) IBM 3203 Printer Model 3: 1000 LPM using 48 character set IBM 3501 Card Reader: Up to 50 cards per minute desktop unit IBM 3521 Card Punch: Up to 50 cards per minute IBM 3782 Card Attachment unit, which allows the 2502 or 3521 to be attached to any terminal except the 3777 IBM 3784 Line Printer, can be attached to a 3774 as a second printer. Up to 155 LPM with 48 characters set print belt. == Workstation programs == IBM distributes workstation programs with systems software including OS/360 Attached Support Processor (ASP) Houston Automatic Spooling Priority (HASP and HASP II) Operating System/Virtual Storage 1 (OS/VS1) Operating System/Virtual Storage 2 (OS/VS2 MVS) Release 2 through 3.8 MVS versions from MVS/SP Version 1 through z/OS Priority Output Writers, Execution processors and input Readers (POWER) Remote Spooling Communications Subsystem (RSCS) Except for the RJE workstation programs in OS/360, these programs use a variation of BSC known as Multi-leaving. In addition, IBM provides separately ordered workstation programs using BSC. Systems Network Architecture (SNA) and TCP/IP. Workstation programs are available from IBM and third-party vendors to support all of these protocols: 2770/3770 2780/3780 Multileaving Network Job Entry (NJE) OS/360 RJE SNA TCP/IP
Compute (machine learning)
In machine learning and deep learning, compute is the amount of computing power or computational resources required to train machine learning models and large language models. More broadly, compute is the computational power or resources necessary for a computer or computer program to function. == Definition == Compute is commonly defined as the amount of computing power or computational resources required to train machine learning and large language models. The term "compute" has also been more broadly applied to cloud computing, referencing processing power, memory, networking, storage, and other resources required for the computation of any program. Compute is measured in petaflop/s-days and is used to document AI training. A petaflop/s-day (pfs-day) consists of performing 1015 neural net operations per second for one day, or a total of about 1020 operations. The compute-time product serves as a mental convenience, similar to kilowatt-hour for energy. An amount of compute is meant to give an idea of the number of actual operations performed. == History == In a 2018 analysis titled "AI and compute", artificial intelligence company OpenAI introduced the concept of compute. OpenAI identified two eras of training AI systems in terms of compute-usage. From 1959 to 2012, compute roughly followed Moore’s law. Between 2012 and 2018, the amount of compute used in the largest AI training runs increased exponentially, growing by more than 300,000 times — roughly doubling every 3.4 months. By comparison, Moore’s Law doubled every two years over the same period. One of the largest models, released in 2020, used 600,000 times more computing power than the 2012 model. After 2020, compute growth began to slow down, with the compute needed for the largest AI models continuing to slow down in 2023. The notion of compute has become increasingly used from the mid-2020s onwards. == Compute growth and AI progress == Larger AI models trained on more data and using more computational resources, tend to perform better. This happens even if the algorithms themselves remain unchanged. As early as 2018, OpenAI noted the exponential increase in compute to be have a key role in AI progress. OpenAI considers three factors drive the advance of AI: algorithmic innovation, data, and the amount of compute available for training. AI models with more compute not only improve in the tasks they were trained on but can develop emergent abilities. Incremental improvements can lead to more abrupt leaps in capabilities. AI provider SpaceXAI said in 2026 that their AI progress is driven by compute and used it a key metric in the AI training of its supercomputer Colossus, the which contains 1 million GPUs. Anthropic has a contract of $1.25 billion per month with SpaceXAI to buy all the compute capacity at Colossus 1 data center. === Criticism and policy === Increasing, promoting or constraining progress in artificial intelligence has often be done via controlling the amount of compute. Policymarkers have enacted policies and provided support to make compute resources more accessible to domestic AI researchers. In a January 2022 report, the Center for Security and Emerging Technology (CSET) suggested to institutions that increasingly powerful and generalizable AI (AGI) will likely require other strategies than maximizing compute. Some AI researchers are also concerned that government might exclusively focus on scaling compute instead of other strategies. The CSET has reported on the various bottlenecks which could explain why deep learning needs for compute have slow down: training is expensive and training extremely large models generates traffic jams across many processors that are difficult to manage. there is a limited supply of AI chips (see AI chip memory shortage). CSET advances that the main resource is human capital, specifically talented researchers — according to a 2023 published survey of more than 400 AI researchers, academic and private sector workers. The survey found that AI researchers are not primarily or exclusively constrained by compute access. However, both academic and industry AI researchers equally report concerns that insufficient compute could prevent them from contributing meaningfully to AI research in the future. High compute users are more concerned about compute access. When asked about which resource provided by the government would be the most useful to them, some AI researchers select compute, other prefer grant funding. For this goal, CSET advised policymakers to ensure that even researchers with smaller budgets could effectively contribute to AI research. Other proposed strategies include using contemporary AI algorithms, managing modern AI infrastructure or focusing on interdisciplinary work between the AI field and other fields of computer science. A 2024 study on compute access found that academic-only AI research teams often have less compute intensive research topics, especially foundation models, compared to industry AI labs. As a consequence, academia is likely to play a smaller role in advancing such techniques. The researchers suggest nationally-sponsored computing infrastructure as well as open science initiatives to boost academic compute access. === Data === A 2022 study found that current large language models are significantly under-trained, a consequence of focusing on scaling language models whilst keeping the amount of training data constant. By training over 400 language models of various parameter and token size, they found that "for compute-optimal training", the model size and the number of training tokens should ideally be scaled equally: for every doubling of model size the number of training tokens should also be doubled.
Plaintext
In cryptography, plaintext usually means unencrypted information pending input into cryptographic algorithms, usually encryption algorithms. This usually refers to data that is transmitted or stored unencrypted. == Overview == With the advent of computing, the term plaintext expanded beyond human-readable documents to mean any data, including binary files, in a form that can be viewed or used without requiring a key or other decryption device. Information—a message, document, file, etc.—if to be communicated or stored in an unencrypted form is referred to as plaintext. Plaintext is used as input to an encryption algorithm; the output is usually termed ciphertext, particularly when the algorithm is a cipher. Codetext is less often used, and almost always only when the algorithm involved is actually a code. Some systems use multiple layers of encryption, with the output of one encryption algorithm becoming "plaintext" input for the next. == Secure handling == Insecure handling of plaintext can introduce weaknesses into a cryptosystem by letting an attacker bypass the cryptography altogether. Plaintext is vulnerable in use and in storage, whether in electronic or paper format. Physical security means the securing of information and its storage media from physical, attack—for instance by someone entering a building to access papers, storage media, or computers. Discarded material, if not disposed of securely, may be a security risk. Even shredded documents and erased magnetic media might be reconstructed with sufficient effort. If plaintext is stored in a computer file, the storage media, the computer and its components, and all backups must be secure. Sensitive data is sometimes processed on computers whose mass storage is removable, in which case physical security of the removed disk is vital. In the case of securing a computer, useful (as opposed to handwaving) security must be physical (e.g., against burglary, brazen removal under cover of supposed repair, installation of covert monitoring devices, etc.), as well as virtual (e.g., operating system modification, illicit network access, Trojan programs). Wide availability of keydrives, which can plug into most modern computers and store large quantities of data, poses another severe security headache. A spy (perhaps posing as a cleaning person) could easily conceal one, and even swallow it if necessary. Discarded computers, disk drives and media are also a potential source of plaintexts. Most operating systems do not actually erase anything— they simply mark the disk space occupied by a deleted file as 'available for use', and remove its entry from the file system directory. The information in a file deleted in this way remains fully present until overwritten at some later time when the operating system reuses the disk space. With even low-end computers commonly sold with many gigabytes of disk space and rising monthly, this 'later time' may be months later, or never. Even overwriting the portion of a disk surface occupied by a deleted file is insufficient in many cases. Peter Gutmann of the University of Auckland wrote a celebrated 1996 paper on the recovery of overwritten information from magnetic disks; areal storage densities have gotten much higher since then, so this sort of recovery is likely to be more difficult than it was when Gutmann wrote. Modern hard drives automatically remap failing sectors, moving data to good sectors. This process makes information on those failing, excluded sectors invisible to the file system and normal applications. Special software, however, can still extract information from them. Some government agencies (e.g., US NSA) require that personnel physically pulverize discarded disk drives and, in some cases, treat them with chemical corrosives. This practice is not widespread outside government, however. Garfinkel and Shelat (2003) analyzed 158 second-hand hard drives they acquired at garage sales and the like, and found that less than 10% had been sufficiently sanitized. The others contained a wide variety of readable personal and confidential information. See data remanence. Physical loss is a serious problem. The US State Department, Department of Defense, and the British Secret Service have all had laptops with secret information, including in plaintext, lost or stolen. Appropriate disk encryption techniques can safeguard data on misappropriated computers or media. On occasion, even when data on host systems is encrypted, media that personnel use to transfer data between systems is plaintext because of poorly designed data policy. For example, in October 2007, HM Revenue and Customs lost CDs that contained the unencrypted records of 25 million child benefit recipients in the United Kingdom. Modern cryptographic systems resist known plaintext or even chosen plaintext attacks, and so may not be entirely compromised when plaintext is lost or stolen. Older systems resisted the effects of plaintext data loss on security with less effective techniques—such as padding and Russian copulation to obscure information in plaintext that could be easily guessed.