This article presents a detailed timeline of events in the history of computing from 2020 to the present. For narratives explaining the overall developments, see the history of computing.
Significant events in computing include events relating directly or indirectly to software, hardware and wetware.
Excluded (except in instances of significant functional overlap) are:
events about uses of computational tools in biotechnology and similar fields (except for improvements to the underlying computational tools) as well as events in media-psychology except when those are directly linked to computational tools
OpenAI develops a model called "GPT 4b-micro", which suggests ways that protein factors could be re-engineered to become more effective.[2]
DeepSeek releases DeepSeek-R1 on 20th January, a large language model based on DeepSeek-V3 utilising a chain-of-thought process similar to OpenAI o1.[3]
2024
AI
An AI-based study shows for the first time that fingerprints from different fingers of the same person share strong detectable similarities.[4][5]
According to the International Telecommunication Union, the global Internet population reached 5.5 billion in 2024, meaning more than two-thirds of the world's population is now online.[15][16]
Chatbot and text-generating AI, ChatGPT (released on 30 Nov 2022), a large language model, became popular, with some considering the large public's attention as unwarranted hype as potential applications are limited. Similar software such as Cleverbot existed for many years, and the software is, on the fundamental level, not structured toward accuracy – e.g. providing seemingly credible but incorrect answers to queries and operating "without a contextual understanding of the language" – but only toward essentially the authenticity of mimicked human language.[18][19][20][21][22] It was estimated that only two months after its launch, it had 100 million active users.[23] Applications may include solving or supporting school writing assignments,[24] malicious social bots (e.g. for misinformation, propaganda, and scams),[25][26] and providing inspiration (e.g. for artistic writing or in design or ideation in general).[27][28]
Google released chatbotBard due to effects of the ChatGPT release,[29] with potential for integration into its Web search and, like ChatGPT software, also as a software development helper tool.[30]DuckDuckGo released the DuckAssist feature integrated into its search engine that summarizes information from Wikipedia to answer search queries that are questions. The experimental feature was shut down without explanation on 12 April.[31][32][33] There has been further development regarding LLMs or ChatGPT as user interfaces of Wikipedia or as software using its structured knowledge by others.[34][35] It may demonstrate an alternative approach to ChatGPT whose fundamental algorithms are not designed to generate text that is true, including for example "hallucinations" and fake citations or misinformation more generally.[36][37]Elicit.org may provide a free alternative to this tool.[38][39] A broader alternative approach to the software's Q&A applications and use of text generation for assignments may be the improvement of media literacy and Web search skills in education systems.
Further LLM developments during what has been called an "AI boom" included: local or open source versions of LLaMA which was leaked in March,[40][41][42] news outlets reported on GPT4-based Auto-GPT that given natural language commands uses the Internet and other tools in attempts to understand and achieve its tasks with unclear or so-far little practicality,[43] a systematic evaluation of answers from four "generative search engines" suggested their outputs "appear informative, but frequently contain unsupported statements and inaccurate citations",[44] a multi-modal open source tool for understanding and generating speech,[45] a data scientist argued that "researchers need to collaborate to develop open-source LLMs that are transparent" and independent,[46]Stability AI launched an open source LLM.[47]
A method for editing NeRF scenes, a novel media technique from 2020,[48] with natural language commands was demonstrated by Nvidia.[49][50]
An open letter "Pause Giant AI Experiments" initiated by the Future of Life Institute called for "AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4" due to "profound risks to society and humanity".[51][52] It received substantial media attention and also contributed to speculations about perceived large LLM potential). At the time there was extensive media coverage of views that regard ChatGPT as a potential step towards AGI or sentient machines, also extending to some academic works (e.g. a popular preprint by a company).[17] The coverage focused on such views may not represent the majority expert views and, for example, some researchers noted that e.g. the ability to generate coherent text and imitations are not the same as understanding language.[53] A set of techniques under development included self-refining code or text.[54]
ChatGPT was shown to outperform human doctors in responding to online medical questions when measured on quality and empathy by "a team of licensed health care professionals",[55][56] albeit the chatbot may have previously been trained with these redditquestion and answers threads.
A novel potentially significantly more efficient text-to-image approach, as implemented in MUSE, was reported.[68][69]
A first successful autonomous long-duration operation, including simulated combat, of a modified F-16 fighter jet, X-62A, by two AI software was reported.[70][71][72]
A text-to-speech synthesizer, VALL-E, that can be trained to mimic anybody's voice with just three seconds of voice data and may produce the most natural-sounding results to date, was reported in a preprint.[73][74]
A use of world models for a wide range of domains that make decisions using e.g. different 3D worlds and reward frequencies and outperforms previous approaches, DreamerV3, was reported as a step towards general artificial intelligence in a preprint.[75][76]
A study reported the development of deep learning algorithms to identify technosignature candidates, finding 8 potential alien signals not detected earlier.[83][84]
The world's first COVID-19 drug designed by generative AI was approved for human use, with clinical trials expected to begin in China. The new drug, ISM3312, was developed by Insilico Medicine.[89]
The LLMGPT-4 was launched by OpenAI.[90][91] It and ChatGPT based on it continued to receive major global media attention.
Researchers suggested that growing influence of industry in AI research means that "public interest alternatives for important AI tools may become increasingly scarce".[92]
Google revealed PaLM-E, an embodied multimodal language model with 562 billion parameters.[93][94]
Researchers demonstrated an open source 'AI scientist' that can create models of natural phenomena from knowledge axioms and experimental data, showing the software can rediscover physical laws like "Kepler's third law of planetary motion, Einstein's relativistic time-dilation law, and Langmuir's theory of adsorption" using logical reasoning and a few data points.[96][95]
The fMRI machine used for brain-reading
Researchers demonstrated a non-invasive brain-reading method. It can translate a person's neural activity into a continuous stream of text using fMRI data and transformer machine learning. Prior training data is required for this semantic decoding. Participants listened to stories for 16 hours while their brain activity was recorded.[97]
A machine learning model is trained to recognise the key features of chemicals with senolytic activity. It finds three chemicals – ginkgetin, periplocin and oleandrin – able to remove senescent cells without damaging healthy cells.[102][103]
Articles in science outlets like Nature suggest contemporary viral concerns about hypothetical existential risk of AI "plays into the tech companies' agenda" – partly in the form of 'criti-hype'[104] – and that this "hinders effective regulation of the societal harms AI is causing right now" and in the near-future.[105]
A science writer provides an overview of "the nascent industry of AI-designed drugs".[106]
Illustration of "thought cloning"
A preprint introduces the concept of "thought cloning" by which AI use data of or imitate human thinking.[107]
Metaresearchers showed that AI trained with study-author-networks data could generate scientifically promising "alien" hypotheses that would likely not be considered otherwise.[108]
A study provides an overview and living review of open source LLMs, assessing the levels of openness of their differentiated elements and reviewing the risks of relying on proprietary software or the importance of open source AI.[109]
AI-supported mammography screening was demonstrated to have the potential to substantially reduce workload and to possibly improve cancer detection rates.[110]
A study that won an international competition demonstrated an approach that predicted 70% of earthquakes, suggesting some form of earthquake prediction may be feasible in the future.[112]
A natural language system was demonstrated that can provide explanations for the conclusion-making of machine learning models for explainable AI.[114][115]
A preprint reported some large language models have an 'extractable memorization' flaw by which training data can be extracted at affordable costs by queries.[116]
A study hypothesized mental health awareness efforts (in current forms) or glamorised and romanticised mental disorders on social media (e.g. quotes about depression on aesthetically-appealing backgrounds shared widely on certain social media)[which?] may have contributed to the recent rise in reported mental health problems – by intensifying and over-diagnosing of such – beyond e.g. increased reporting of previously under-recognised symptoms or mental health-related issues.[149][150]
Parts of Twitter's recommender algorithms became open source, welcomed and requested by many albeit with several issues related to code exclusion and verifiability.[154][155][156] Around the time, the free version of its API, which was also used for research, was shut down[157] – followed shortly thereafter by reddit,[158] proprietary verification checkmarks caused controversy,[159] parts of its source code were leaked,[160] and applications of a "state-affiliated media" label – which purportedly uses "Category:Publicly funded broadcasters" data which, like the label, does not differentiate and list the shares of funding sources – caused controversy.[161][162] In April, Twitter was warned by EU digital policy-makers after a report indicated its recent policies "boost" Russian disinformation-based propaganda.[163]
An umbrella review summarized the research on benefits and risks associated with digital media use by youths, suggesting caregivers, policymakers and researchers should continue to move away from prevailing oversimplified recommendations to reduce screen time to instead focus on distinguished types of screen use.[171]
An experiment suggests people and search engines often fail in online searches for evaluating misinformation.
Researchers reported the development of a fuel cell implant powered by bloodglucose. It can also release insulin at certain levels and have enough energy to allow smartphone implant control.[194][195]
Bioinspired neuromorphic motion-cognition nerve in comparison with an ocular-vestibular cross-modal sensory nerve of macaques[196]
Researchers reported the development of neuromorphicAI hardware using nanowires(see also 2020-04-20) physically mimicking the brain's activity in identifying and remembering an image from memory.[197][198] A university reported on a demonstration of multisensory motion cue integration by a neuromorphic nerve for robots.[196]
DeepMind announced that its AlphaFold program had uncovered the structures of more than 200 million foldedproteins, essentially all of those known to science.[202][203]
AI company DeepMind reported that its AlphaFold program had determined the likely structure of nearly every protein known to science.
News outlets reported artificial intelligence art had won the first place in a digital art competition.[204] Such artistic imagery is generated using input consisting of text and sometimes images, usually including parameters such as artistic style (text-to-image generation). Around the time, an expert concluded that "AI art is everywhere right now", with even experts not knowing what it will mean,[205] a news outlet established that "AI-generated art booms" and reported issues of copyright and automation of professional artists,[206] a news outlet investigated how online communities (e.g. their rules) confronted with many such artworks react,[207] a news outlet raised concerns over deepfakes,[208] a magazine highlighted possibilities of enabling "new forms of artistic expression",[209] and an editorial noted that it may be seen as a welcome "augmentation of human capability".[210][additional citation(s) needed] Moreover, additional functionalities – such as enabling the use of user-provided concepts (like an object or a style) learned from few images for novel personalized art generated from the associated word/s[211] or expanding beyond the borders of artistic images in the same style[212] – were reported. On 22 August,[213][214]Stable Diffusion released as open source software, making the technology more accessible and free to use on personal hardware as well as extendable by third-parties (i.e. other software projects).[214][215]
Researchers reported the development of a deep learning system that learns intuitive physics from visual data (of virtual 3D environments) to some degree "from scratch" based on an unpublished approach inspired by studies of visual cognition in infants.[220][221] Two weeks later, other researchers reported the development of a machine learning algorithm that could discover sets of basic variables of various physical systems and predict the systems' future dynamics from video recordings of their behavior.[222][223]
The first data transmission to exceed 1 petabit per second (Pbit/s) using a single laser and a single optical chip was demonstrated by European researchers.[244][245]
A satellite-free GPS-alternative higher-resolution positioning system using existing telecommunications networks was demonstrated, SuperGPS.[249][250]
Impossible Metals announced its first underwater robotic vehicle, 'Eureka 1', had completed its first trial of selectively harvesting polymetallic nodule rocks from the seabed nearly without harming the environment (as compared to other seabed mining) to help address the rising global need for metals for renewable energy system components – mainly batteries. It operates autonomously and uses advanced computer vision, e.g. using AI to determine which rocks have signs of visible life on them so that they are not harvested.[251]
Software
A report by the Royal Society listed potential or proposed countermeasures against misinformation, mainly online misinformation, such as, broadly described, building resilience to scientific misinformation and a healthy online information environment.[252]
Computational biologists reported the largest detailed human genetic genealogy, unifying human genomes from many sources for insights about human history, ancestry and evolution. It demonstrates a novel computational method for estimating how human DNA is related, in specific as a series of 13 million linked trees along the genome, a tree-sequence, which has also been called "the largest human family tree".[image needed][253][254][255]
Researchers reported the creation of a version control system for cell engineering, suggesting it to be a "significant step toward more open, reproducible, easier to trace and share, more trustworthy engineering biology", and possibly increased safety by enabling faster tracing of organisms lab of origin and design details via barcoding.[256][257]
A preprint demonstrated how backdoors can be placed undetectably into classifying (e.g. posts as "spam" or well-visible "not spam") machine learning models which are often developed and/or trained by third parties. Parties can change the classification of any input, including in cases with types of data/software transparency, possibly including white-box access.[258][259][260]
Researchers reported routes for recycling 200 industrial waste chemicals into important drugs and agrochemicals using a software for computer-aided chemical synthesis design, helping enable "circular chemistry" as a potential area of a circular economy.[261][262]
Measured results of the study about change in intelligence in children 9–12 from screen time watching, screen time socializing and screen time gaming[265]
A study suggested that in children at age 9–12 during two years time, gaming or watching digital videos can be positively correlated with measures of intelligence, albeit correlations with overall screen time (including social media, socializing and TV) were not investigated and 'time gaming' did not differentiate between categories of video games (e.g. shares of games' platform and genre).[266][265]
Computing and the 2022 Russian invasion of Ukraine: An editorial published in a journal noted that remote surgery and types of videoconferencing for sharing expertise (e.g. ad hoc assistance) have been and could be used to support doctors in Ukraine.[267] A forum contribution analyzed Russian users' reactions to the Bucha massacre on social media – on nationalist Telegram channels.[268] Computer science and technology was also used to defend against the 2022 Russian invasion such as with military technology,[269][270] to document and communicate war events including via facial recognition of dead Russian soldiers and Russian war crimes,[271][272] and for aggregated information about support opportunities for Ukrainian scientists.[273][274]
A study explored the efficacy of GitHub Sponsors, launched in 2019, in terms of supporting FOSS-efforts and the sponsors' intentions as well as the motivation of sponsors and various quantitative and qualitative analyses relevant to this approach of FOSS-funding.[277][278]
News outlets reported that in July, for the first time, more people watched streaming TV than cable within the U.S.[globalize].[284][285]
A researcher reported that the social media appTikTok adds a keylogger to its, on iOS essentially unavoidable, in-app browser in iOS, which allows its Chinese company to gather, for example, passwords, credit card details, and everything else that is typed into websites opened from taps on any external links within the app. Shortly after the report, the company claimed such capabilities are only used for debugging-types of purposes.[286][287] To date, it has largely not been investigated which and to which extent (other) apps have capacities for such or similar data-collection.[286][287][additional citation(s) needed]
Results of investigations about the issue of recommendation systems shifting their users' preferences "so they are easier to satisfy" were reported, including actively optimizing such software to avoid problematic shifts/manipulation, suggesting "that recommenders that optimize for staying in the trust region can avoid manipulative behaviors while still generating engagement".[297][298]
After domain seizures of Z-Library by copyright law enforcement and moves toward dark web and IPFS technologies by its content providers, the open sourceshadow libraryUIAnna's Archive – which also provides access to a full copy of Z-Library content and scientific articles – was established by a team of archivists,[301][302] essentially providing the largest human book and literature library.
Around the acquisition of Twitter by Elon Musk (27 October), interest in alternatives to the site – described as "one of the world's most high-profile information ecosystems", a contemporary suboptimal public square, and as heavily used by many journalists and news media – increased substantially. However, no alternative such as Mastodon, Reddit or the Bluesky protocol was found to match its features such as ease of use to date, in terms of being able to substitute the site.[307][308][309]
A university reported on the first study of the new privacy-intrusion Web tracking technique of "UIDsmuggling" by the ad industry, which finds it to be prevalent and largely not mitigated by the latest protection tools – such as Firefox's tracking protection and uBlock Origin – and contributes to countermeasures.[317][318]
A study estimated losses of 61 metals to help the development of circular economy strategies, showing that usespans of, often scarce, tech-critical metals are short.[326][327]
Samsung announced the first mass production of computer chips using a 3 nm process. These feature a gate-all-aroundtransistor architecture that reduces power consumption by up to 45%, improves performance by 23% and reduces area by 16% compared to 5 nm.[336][337]
Scientists reported a so far unique and unknown feature of material VO2 – it can "remember" previous external stimuli (via structural rather than electronic states), with potential e.g. data storage.[342][343]
[Data usage] – In a static proprietary article that appeared in and was co-reviewed by a scientific journal, authenticated scientists analyzed data from multiple public databases to create a regional representation of levels of deforestation induced by nations' recent, largely unmodulated, trade-, production- and consumption-patterns.[352][353]
A study found that carbon emissions from Bitcoin mining in China – where a majority of the proof-of-work algorithm that generates current economic value is computed, largely fueled by nonrenewable sources – have accelerated rapidly and would soon exceed total annual emissions of countries like Italy, interfering with climate change mitigation commitments.
A study found that carbon emissions from Bitcoin mining in China – where a majority of the proof-of-work algorithm that generates current economic value is computed, largely fueled by nonrenewable sources – had accelerated rapidly and would soon exceed total annual emissions of countries like Italy, interfering with climate change mitigation commitments.[354][355]
Neuralink revealed a male macaque with chips embedded on each side of its brain, playing a mind-controlled version of Pong. While similar technology has been demonstrated for decades, and wireless implants have existed for years, some observers noted that the organization increased the number of implanted electrodes that are read wirelessly.[356][357][358]
Scientists reviewed materials strategies for organic neuromorphic devices, suggesting that "their biocompatibility and mechanical conformability give them an advantage for creating adaptive biointerfaces, brain-machine interfaces, and biology-inspired prosthetics".[359][360][relevant?]
Researchers published the first in-depth study of Web browsertabinterfaces. They found that many people struggle with tab overload and conducted surveys and interviews about people's tab use. Thereby they formalized pressures for closing tabs and for keeping tabs open. The authors then developed related UI design considerations which could enable better tools and changes to the code of Web browsers – like Firefox – that allow knowledge workers and other users to better manage their tabs.[361][362]
A new record for the smallest single-chip system was achieved, occupying a total volume of less than 0.1 mm³.[364][365]
Scientists demonstrated the first brain–computer interface that decodes neural signals for handwriting and has a record output speed of up to 90 characters per minute – more than double the previous record.
Scientists demonstrated the first brain–computer interface that decodes neural signals for handwriting. The character output speed of a patient with a paralyzed hand was up to 90 characters per minute – more than double the previous record. Each letter is associated with a highly distinctive pattern of activity in the brain, making it relatively easy for the algorithm to distinguish them.[366][367]
Archivists initiated a rescue mission to secure enduring access to humanity's largest public library of scientific articles, Sci-Hub, due to the site's increased legal troubles, using Web and BitTorrent technologies.[368]
El Salvador passed the Bitcoin Law, making it the first country to give cryptocurrency and bitcoin a status of legal tender.[373] The law was passed by the Legislative Assembly of El Salvador on June 8, 2021, giving the cryptocurrency bitcoin the status of legal tender within El Salvador after September 7, 2021.[374] It was proposed by President Nayib Bukele. The text of the law states that "the purpose of this law is to regulate bitcoin as unrestricted legal tender with liberating power, unlimited in any transaction, and to any title that public or private natural or legal persons require carrying out".[375]
Scientists debated the research cognitive impacts of smartphones and digital technology in general and by prevalent forms of use.
In the debate regarding the cognitive impacts of smartphones and digital technology, a group reported that, contrary to widespread belief, scientific evidence does not show that these technologies harm biological cognitive abilities and that they instead change predominant ways of cognition – such as a reduced need to remember facts or conduct mathematical calculations by pen and paper outside contemporary schools. However, some activities – like reading novels – that require long attention-spans and don't feature ongoing rewarding stimulation may become more challenging in general.[379][380]
Researchers used a brain–computer interface to enable a man who was paralyzed since 2003 to produce comprehensible words and sentences by decoding signals from electrodes in the speech areas of his brain.[382][383]
Japan achieved a new world record Internet speed: 319 Tbit/s over ~3000 km which, albeit not being the fastest speed overall, beats the previous record of 178 Tbit/s.[384][385]
Scientists reported that worldwide adolescentloneliness and depression increased substantially after 2012 and that loneliness in contemporary schools appears to be associated with smartphone access and Internet use.[386][387]
DeepMind's AlphaFold AI predicted the structures of over 350,000 proteins, including 98.5% of the ~20,000 proteins in the human body, along with degrees of confidence for accuracy.
DeepMind announced that its AlphaFold AI had predicted the structures of over 350,000 proteins, including 98.5% of the ~20,000 proteins in the human body. The 3D data along with their degrees of confidence for accuracy was made freely available with a database, doubling the previous number of protein structures in the public domain.[388]
Researchers developed machine learning models for genome-based early detection and prioritization of high-risk potential zoonotic viruses.
Scientists concluded that personal carbon allowances (PCAs) could be a component of climate change mitigation. They found that the economic recovery from COVID-19 and novel digital technology capacities open a window of opportunity for first implementations. PCAs would consist of – e.g. monetary – credit-feedbacks and decreasing default levels of per capita emissions concessions. The researchers found that recent advances in machine learning technology and "smarter home and transport options make it possible to easily track and manage a large share of individuals' emissions" and that feedback effective in engaging individuals to reduce their energy-related emissions and relevant new personalized apps could be designed.[394][395][396] Issues may include privacy, evaluating emissions from individuals co-running multinational companies and the availability and prices of products and services.
Cerebras announced a new hardware and software platform that can support AI models of 120 trillion parameters, enabling neural networks greater than the equivalent number of human brainsynapses.[397]
Pathogen researchers reported the development of machine learning models for genome-based early detection and prioritization of high-risk potential zoonotic viruses in animals prior to spillover to humans. They concluded that their tool could be used for virus surveillance for pandemic prevention via (i.a.) measures of "early investigation and outbreak preparedness" and would have been capable of predicting SARS-CoV-2 as a high-risk strain.[398][399]
A loss of public IP routes to the Facebook DNS servers due to malfu
.webp)
.png)
_that_captures_and_stores_small_images_into_DNA_via_encoding_with_light.webp)
.jpg)
.png)
![[icon]](File:Wiki_letter_w_cropped.svg){kind=small}