The global call for AI red lines is a declaration made on 22 September 2025 calling on governments to define and internationally prohibit unacceptable AI uses and behaviors. The online declaration was announced by Nobel Peace Prize laureate Maria Ressa at the 80th United Nations General Assembly high-level week. The declaration was initially signed by 200 prominent politicians and scientists, including 10 Nobel Prize winners. The call does not specify which red lines to set, but suggests several, such as banning bioweapon design, mass surveillance or AI impersonation. == The declaration == The declaration was published online as an open letter on 22 September 2025. Nobel Peace Prize laureate Maria Ressa announced it in her opening speech at the 80th United Nations General Assembly high-level week in New York, urging governments to "define what AI should never be allowed to do" and "establish clear international boundaries to prevent universally unacceptable risks for A.I." The initiative was organized by three nonprofit organisations: the French Center for AI Safety (CeSIA), The Future Society, and the Center for Human-Compatible Artificial Intelligence (CHAI). The letter argues that humanity faces risks such as engineered pandemics, widespread disinformation, large-scale manipulation, unemployment and loss of control. Proponents argue that national laws are insufficient to address these risks and that "an international agreement on clear and verifiable red lines is necessary". They urge governments to reach an agreement by the end of 2026, and called for robust enforcement mechanisms and the creation of an independent organisation to implement it. The letter does not call for specific red lines, but suggests the possibility of banning lethal autonomous weapons, autonomous replication of AI systems and the use of AI in nuclear warfare. Other examples of possible red lines include social scoring, mass surveillance, bioweapon design, AI-generated child sexual abuse material and AI impersonation. A red line could prohibit either AI behaviors (what AI systems should be guaranteed to never do even if asked to) or AI uses. == Signatories == When published, the online declaration was signed by more than 200 prominent politicians and scientists, including 10 Nobel Prize winners. Signers include former president of Colombia Juan Manuel Santos and researchers Geoffrey Hinton and Yoshua Bengio. It also includes popular authors like Stephen Fry and Yuval Noah Harari. The letter received support from European lawmakers, including former Italian prime minister Enrico Letta, and former president of Ireland Mary Robinson. == Development of red lines == As of 2025, there is no global red line on AI. Some regional red lines exist, such as with the uses deemed "unacceptable" by the AI Act in Europe, and with the US-China agreement not to leave to AI the decision of whether to launch nuclear weapons. At the United Nations Security Council, days after the declaration, Michael Kratsios, Donald Trump's director of the White House Office of Science and Technology Policy, said "We totally reject all efforts by international bodies to assert centralized control and global governance of AI." The topic of AI red lines gained prominence in 2026 with the dispute between Anthropic and the Department of Defense (DoD), which resulted from the DoD requesting Anthropic to remove contractual red lines on fully autonomous weapons and mass domestic surveillance. The event led employees from Google and OpenAI as well as Senate Democrats to further call for red lines on military use of AI. Senator Adam Schiff proposed a bill to "codify" Anthropic's red lines.
Image tracing
In computer graphics, image tracing, raster-to-vector conversion or raster vectorization is the conversion of raster graphics into vector graphics. == Background == An image does not have any structure: it is just a collection of marks on paper, grains in film, or pixels in a bitmap. While such an image is useful, it has some limits. If the image is magnified enough, its artifacts appear. The halftone dots, film grains, and pixels become apparent. Images of sharp edges become fuzzy or jagged. See, for example, pixelation. Ideally, a vector image does not have the same problem. Edges and filled areas are represented as mathematical curves or gradients, and they can be magnified arbitrarily (though of course the final image must also be rasterized in to be rendered, and its quality depends on the quality of the rasterization algorithm for the given inputs). The task in vectorization is to convert a two-dimensional image into a two-dimensional vector representation of the image. It is not examining the image and attempting to recognize or extract a three-dimensional model that may be depicted; i.e. it is not a vision system. For most applications, vectorization also does not involve optical character recognition; characters are treated as lines, curves, or filled objects without attaching any significance to them. In vectorization, the shape of the character is preserved, so artistic embellishments remain. Vectorization is the inverse operation corresponding to rasterization, as integration is to differentiation. And, just as with these other operations, while rasterization is fairly straightforward and algorithmic, vectorization involves the reconstruction of lost information and therefore requires heuristic methods. Synthetic images such as maps, cartoons, logos, clip art, and technical drawings are suitable for vectorization. Those images could have been originally made as vector images because they are based on geometric shapes or drawn with simple curves. Continuous tone photographs (such as live portraits) are not good candidates for vectorization. The input to vectorization is an image, but an image may come in many forms such as a photograph, a drawing on paper, or one of several raster file formats. Programs that do raster-to-vector conversion may accept bitmap formats such as TIFF, BMP and PNG. The output is a vector file format. Common vector formats are SVG, DXF, EPS, EMF and AI. Vectorization can be used to update images or recover work. Personal computers often come with a simple paint program that produces a bitmap output file. These programs allow users to make simple illustrations by adding text, drawing outlines, and filling outlines with a specific color. Only the results of these operations (the pixels) are saved in the resulting bitmap; the drawing and filling operations are discarded. Vectorization can be used to recapture some of the information that was lost. Vectorization is also used to recover information that was originally in a vector format but has been lost or has become unavailable. A company may have commissioned a logo from a graphic arts firm. Although the graphics firm used a vector format, the client company may not have received a copy of that format. The company may then acquire a vector format by scanning and vectorizing a paper copy of the logo. == Process == Vectorization starts with an image. === Manual === The image can be vectorized manually. A person could look at the image, make some measurements, and then write the output file by hand. That was the case for the vectorization of a technical illustration about neutrinos. The illustration has a few geometric shapes and a lot of text; it was relatively easy to convert the shapes, and the SVG vector format allows the text (even subscripts and superscripts) to be entered easily. The original image did not have any curves (except for the text), so the conversion is straightforward. Curves make the conversion more complicated. Manual vectorization of complicated shapes can be facilitated by the tracing function built into some vector graphics editing programs. If the image is not yet in machine readable form, then it has to be scanned into a usable file format. Once there is a machine-readable bitmap, the image can be imported into a graphics editing program (such as Adobe Illustrator, CorelDRAW, or Inkscape). Then a person can manually trace the elements of the image using the program's editing features. Curves in the original image can be approximated with lines, arcs, and Bézier curves. An illustration program allows spline knots to be adjusted for a close fit. Manual vectorization is possible, but it can be tedious. Although graphics drawing programs have been around for a long time, artists may find the freehand drawing facilities awkward even when a drawing tablet is used. Instead of using a program, Pepper recommends making an initial sketch on paper. Instead of scanning the sketch and tracing it freehand in the computer, Pepper states: "Those proficient with a graphic tablet and stylus could make the following changes directly in CorelDRAW by using a scan of the sketch as an underlay and drawing over it. I prefer to use pen and ink, and a light table"; most of the final image was traced by hand in ink. Later the line-drawing image was scanned at 600 dpi, cleaned up in a paint program, and then automatically traced with a program. Once the black and white image was in the graphics program, some other elements were added and the figure was colored. Similarly, Ploch recreated a design from a digital photograph. The JPEG was imported and some "basic shapes" were traced by hand and colored in the graphics drawing program; more complex shapes were handled differently. Ploch used a bitmap editor to remove the background and crop the more complex image components. He then printed the image and traced it by hand onto tracing paper to get a clean black and white line drawing. That drawing was scanned and then vectorized with a program. === Automatic === Some programs automate the vectorization process. Example programs are Adobe Illustrator, Inkscape, Corel's PowerTRACE, and Potrace. Some of these programs have a command line interface while others are interactive that allow the user to adjust the conversion settings and view the result. Adobe Streamline is not only an interactive program, but it also allows a user to manually edit the input bitmap and the output curves. Corel's PowerTRACE is accessed through CorelDRAW; CorelDRAW can be used to modify the input bitmap and edit the output curves. Adobe Illustrator has a facility to trace individual curves. Automated programs can have mixed results. A program (PowerTRACE) was used to convert a PNG map to SVG. The program did a good job on the map boundaries (the most tedious task in the tracing) and the settings dropped out all the text (small objects). The text was manually re-inserted. Other conversions may not go as well. The results depend on having high-quality scans, reasonable settings, and good algorithms. Scanned images often have a lot of noise, which can require additional work to clean up. == Options == There are many different image styles and possibilities, and no single vectorization method works well on all images. Consequently, vectorization programs have many options that influence the result. One issue is what the predominant shapes are. If the image is of a fill-in form, then it will probably have just vertical and horizontal lines of a constant width. The program's vectorization should take that into account. On the other hand, a CAD drawing may have lines at any angle, there may be curved lines, and there may be several line weights (thick for objects and thin for dimension lines). Instead of (or in addition to) curves, the image may contain outlines filled with the same color. Adobe Streamline allows users to select a combination of line recognition (horizontal and vertical lines), centerline recognition, or outline recognition. Streamline also allows small outline shapes to be thrown out; the notion is such small shapes are noise. The user may set the noise level between 0 and 1000; an outline that has fewer pixels than that setting is discarded. Another issue is the number of colors in the image. Even images that were created as black on white drawings may end up with many shades of gray. Some line-drawing routines employ anti-aliasing; a pixel completely covered by the line will be black, but a pixel that is only partially covered will be gray. If the original image is on paper and is scanned, there is a similar result: edge pixels will be gray. Sometimes images are compressed (e.g., JPEG images), and the compression will introduce gray levels. Many of the vectorization programs will group same-color pixels into lines, curves, or outlined shapes. If each possible color is grouped into its object, there can be an enormous number of objects. Instead, the user is asked to s
Visual temporal attention
Visual temporal attention is a special case of visual attention that involves directing attention to specific instant of time. Similar to its spatial counterpart visual spatial attention, these attention modules have been widely implemented in video analytics in computer vision to provide enhanced performance and human interpretable explanation of deep learning models. As visual spatial attention mechanism allows human and/or computer vision systems to focus more on semantically more substantial regions in space, visual temporal attention modules enable machine learning algorithms to emphasize more on critical video frames in video analytics tasks, such as human action recognition. In convolutional neural network-based systems, the prioritization introduced by the attention mechanism is regularly implemented as a linear weighting layer with parameters determined by labeled training data. == Application in Action Recognition == Recent video segmentation algorithms often exploits both spatial and temporal attention mechanisms. Research in human action recognition has accelerated significantly since the introduction of powerful tools such as Convolutional Neural Networks (CNNs). However, effective methods for incorporation of temporal information into CNNs are still being actively explored. Motivated by the popular recurrent attention models in natural language processing, the Attention-aware Temporal Weighted CNN (ATW CNN) is proposed in videos, which embeds a visual attention model into a temporal weighted multi-stream CNN. This attention model is implemented as temporal weighting and it effectively boosts the recognition performance of video representations. Besides, each stream in the proposed ATW CNN framework is capable of end-to-end training, with both network parameters and temporal weights optimized by stochastic gradient descent (SGD) with back-propagation. Experimental results show that the ATW CNN attention mechanism contributes substantially to the performance gains with the more discriminative snippets by focusing on more relevant video segments. == Literature == Seibold VC, Balke J and Rolke B (2023): Temporal attention. Front. Cognit. 2:1168320. doi: 10.3389/fcogn.2023.1168320.
Cleverbot
Cleverbot is a chatterbot web application. It was created by British AI scientist Rollo Carpenter and launched in October 2008. It was preceded by Jabberwacky, a chatbot project that began in 1988 and went online in 1997. In its first decade, Cleverbot held several thousand conversations with Carpenter and his associates. Since launching on the web, the number of conversations held has exceeded 150 million. Besides the web application, Cleverbot is also available as an iOS, Android, and Windows Phone app. == Operation == Cleverbot's responses are not pre-programmed because it learns from human input: Humans type into the box below the Cleverbot logo and the system finds all keywords or an exact phrase matching the input. After searching through its saved conversations, it responds to the input by finding how a human responded to that input when it was asked, in part or in full, by Cleverbot. Cleverbot participated in a formal Turing test at the 2011 Techniche festival at the Indian Institute of Technology Guwahati on 3 September 2011. Out of the 1334 votes cast, Cleverbot was judged to be 59.3% human, compared to the rating of 63.3% human achieved by human participants. A score of 50.05% or higher is often considered to be a passing grade. The software running for the event had to handle just 1 or 2 simultaneous requests, whereas online Cleverbot is usually talking to around 10,000 to 50,000 people at once. == Developments == Cleverbot is constantly growing in data size at the rate of 4 to 7 million interactions per day. Updates to the software have been mostly behind the scenes. In 2014, Cleverbot was upgraded to use GPU serving techniques. Unlike Eliza, the program does not respond in a fixed way, instead choosing its responses heuristically using fuzzy logic, the whole of the conversation being compared to the millions that have taken place before. Cleverbot now uses over 279 million interactions, about 3-4% of the data it has already accumulated. The developers of Cleverbot are attempting to build a new version using machine learning techniques. An app that uses the Cleverscript engine to play a game of 20 Questions has been launched under the name Clevernator. Unlike other such games, the player asks the questions and it is the role of the AI to understand, and answer factually. An app that allows owners to create and talk to their own small Cleverbot-like AI has been launched, called Cleverme! for Apple products. == In popular culture == Cleverbot received media attention after being featured in the popular 2010 creepypasta ARG web serial Ben Drowned by Alexander D. Hall. In early 2017, a Twitch stream of two Google Home devices modified to talk to each other using Cleverbot garnered over 700,000 visitors and over 30,000 peak concurrent viewers.
Just This Once
Just This Once is a 1993 romance novel written in the style of Jacqueline Susann by a Macintosh IIcx computer named "Hal" in collaboration with its programmer, Scott French. French reportedly spent $40,000 and 8 years developing an artificial intelligence program to analyze Susann's works and attempt to create a novel that Susann might have written. A legal dispute between the estate of Jacqueline Susann and the publisher resulted in a settlement to split the profits, and the book was referenced in several legal journal articles about copyright laws. The book had two small print runs totaling 35,000 copies, receiving mixed reviews. == Creation == The novel's creation spanned the fields of artificial intelligence, expert systems, and natural language processing. Scott French first scanned and analyzed portions of two books by Jacqueline Susann, Valley of the Dolls and Once Is Not Enough, to determine constituents of Susann's writing style, which French stated was the most difficult task. This analysis extracted several hundred components including frequency and type of sexual acts and sentence structure. "Once you're there, the writer's style emerges, part of her actual personality comes out, and the computer can be programmed to make a story." French also created several thousand rules to govern tone, plotting, scenes, and characters. The text generated by Hal, the computer, was intended to mimic what Susann might have written, although the output required significant editing. French credits Hal's work with "almost 100% of the plot, 100% of the theme and style." French estimates that he wrote 10% of the prose, the computer Hal wrote about 25% of the prose, and the remaining two-thirds was more of a collaboration between the two. A typical scenario to write a scene would involve Hal asking questions that French would answer (for example, Hal might ask about the "cattiness factor" involved in a meeting between two key female characters, and French would reply with a range of 1 to 10), and the computer would then generate a few sentences to which French would make minor edits. The process would repeat for the next few sentences until the scene was written. == Legal issues == Jacqueline Susann's publisher was skeptical of the legality of Just This Once, although French doubted that an author's thought processes could be copyrighted. Susann's estate reportedly threatened to sue Scott French but the parties settled out of court; the settlement involved splitting profits between the parties but the terms of the settlement were not disclosed. The publication of Just This Once raised questions in the legal profession concerning how copyright law applies to computer-generated works derived from an analysis of other copyrighted works, and whether the generation of such works infringes on copyright. The publications on this topic suggested that the copyright laws of the time were ill-equipped to deal with computer-generated creative works. == Reception == The book's publisher Steven Shragis of Carol Group said of the novel, "I'm not going to say this is a great literary work, but it's every bit as good as anything out in this field, and better than an awful lot." The novel received some positive early reviews. In USA Today, novelist Thomas Gifford compared Just This Once to another novel in the same genre, American Star by Jackie Collins. Gifford concluded: "If you do like this stuff, you'd be much, much better off with the one written by the computer." The Dead Jackie Susann Quarterly declared that Susann "would be proud. Lots of money, sleaze, disease, death, oral sex, tragedy and the good girl gone bad." Other reviews were mixed. Publishers Weekly wrote, "If the books of Jacqueline Susann and Harold Robbins seem formulaic, this debut novel of sin and success in Las Vegas outdoes them all. And that, in a way, is the point.... All novelty rests in the conceit of computer authorship, not in the story itself." Library Journal stated "French invested eight years and $50,000 in a scheme to use artificial intelligence to fulfill his authentic, if dubious, desire to generate a trashy novel a la Jacqueline Susann. Shallow, beautiful-people characters are flatly conceived and randomly accessed in a formulaic plot ... a sexy, boring morality tale. Of possible interest to computer buffs for its use of Expert Systems and the virtual promise of more worthy possibilities; others should read Susann." Kirkus Reviews wrote: "The deal here is that author French is not the author, he's just the midwife, having allegedly programmed his computer to write about our times just the way Susann would... almost perfectly capturing glamorous Jackie's turgid but E-Z reading prose style and ultrareliable mix of sex, glitz, dope 'n' despair.... One wonders, though, if French's tale spinning PC will do as well on the talkshows as Jackie did. The computer weenies have been trying to tell us for years, garbage in-garbage out."
15.ai
15.ai was a free non-commercial web application and research project that uses artificial intelligence to generate text-to-speech voices of fictional characters from popular media. Created by a pseudonymous artificial intelligence researcher known as 15, who began developing the technology as a freshman during their undergraduate research at the Massachusetts Institute of Technology (MIT), the application allows users to make characters from video games, television shows, and movies speak custom text with emotional inflections. The platform is able to generate convincing voice output using minimal training data; the name "15.ai" references the creator's statement that a voice can be cloned with just 15 seconds of audio. It was an early example of an application of generative artificial intelligence during the initial stages of the AI boom. Launched in March 2020, 15.ai became an Internet phenomenon in early 2021 when content utilizing it went viral on social media and quickly gained widespread use among Internet fandoms, such as the My Little Pony: Friendship Is Magic, Team Fortress 2, and SpongeBob SquarePants fandoms. The service featured emotional context through emojis, precise pronunciation control, and multi-speaker capabilities. Critics praised 15.ai's accessibility and emotional control but criticized its technical limitations in prosody options and non-English language support, with mixed results depending on character complexity. 15.ai is credited as the first platform to popularize AI voice cloning in memes and content creation. Voice actors and industry professionals debated 15.ai's implications, raising concerns about employment impacts, voice-related fraud, and potential misuse. In January 2022, it was discovered that a company called Voiceverse had generated voice lines using 15.ai without attribution, promoted them as the byproduct of their own technology, and sold them as non-fungible tokens (NFT) without permission. News publications universally characterized this incident as the company having "stolen" from 15.ai. The service went offline in September 2022 due to legal issues surrounding artificial intelligence and copyright. Its shutdown was followed by the emergence of commercial alternatives whose founders have acknowledged 15.ai's pioneering influence in the field of deep learning speech synthesis. On May 18, 2025, 15 launched 15.dev as the sequel to 15.ai. == History == === Background === The field of speech synthesis underwent a significant transformation with the introduction of deep learning approaches. In 2016, DeepMind's publication of the WaveNet paper marked a shift toward neural network-based speech synthesis, which enabled higher audio quality via causal convolutional neural networks. Previously, concatenative synthesis—which worked by stitching together pre-recorded segments of human speech—was the predominant method for generating artificial speech, but it often produced robotic-sounding results at the boundaries of sentences. In 2018, Google AI's Tacotron 2 showed that neural networks could produce highly natural speech synthesis but required substantial training data (typically tens of hours of audio) to achieve acceptable quality. When trained on two hours of training data, the output quality degraded while still being able to maintain intelligible speech; with 24 minutes of training data, Tacotron 2 failed to produce intelligible speech. The same year saw the emergence of HiFi-GAN, a generative adversarial network (GAN)-based vocoder that improved the efficiency of waveform generation while producing high-fidelity speech, followed by Glow-TTS, which introduced a flow-based approach that allowed for both fast inference and voice style transfer capabilities. Chinese tech companies like Baidu and ByteDance also made contributions to the field by developing breakthroughs that further advanced the technology. === 2016–2020: Conception and development === 15.ai was conceived in 2016 as a research project in deep learning speech synthesis by a developer known as 15 (at the age of 18) during their freshman year at MIT as part of its Undergraduate Research Opportunities Program. 15 was inspired by DeepMind's WaveNet paper, with development continuing through their studies as Google AI released Tacotron 2 the following year. By 2019, they had demonstrated at MIT their ability to replicate WaveNet and Tacotron 2's results using 75% less training data than previously required. The name "15.ai" is a reference to the developer's statement that a voice can be cloned with as little as 15 seconds of data. 15 had originally planned to pursue a PhD based on their undergraduate research, but opted to work in the tech industry instead after their startup was accepted into the Y Combinator accelerator in 2019. After their departure in early 2020, 15 returned to their voice synthesis research and began implementing it as a web application. According to a post on X from 15, instead of using conventional voice datasets like LJSpeech that contained simple, monotone recordings, they sought out more challenging voice samples that could demonstrate the model's ability to handle complex speech patterns and emotional undertones. During this phase, 15 discovered the Pony Preservation Project, a collaborative project started by /mlp/, the My Little Pony board on 4chan. Contributors of the project had manually trimmed, denoised, transcribed, and emotion-tagged thousands of voice lines from My Little Pony: Friendship Is Magic and had compiled them into a dataset that provided ideal training material for 15.ai. === 2020–2022: Release and operation === 15.ai was released on March 2, 2020 as a free and non-commercial web application that did not require user registration to use, but did require the user to accept its terms of service before proceeding. At the time of its launch, the platform had a limited selection of available characters, including those from My Little Pony: Friendship Is Magic and Team Fortress 2. Users were permitted to create any content with the synthesized voices under two conditions: they had to properly credit 15.ai by including "15.ai" in any posts, videos, or projects using the generated audio; and they were prohibited from mixing 15.ai outputs with other text-to-speech outputs in the same work to prevent misrepresentation of the technology's capabilities. On March 8, 2020, Tyler McVicker of Valve News Network uploaded a YouTube video showcasing 15.ai. More voices were added to the website in the following months. In late 2020, 15 implemented a multi-speaker embedding in the deep neural network, which enabled the simultaneous training of multiple voices. Following this, the website's roster expanded from eight to over fifty characters. In addition, this implementation allowed the deep learning model to recognize common emotional patterns across different characters, even when certain emotions were missing from the characters' training data. By May 2020, the site had served over 4.2 million audio files to users. In early 2021, the application gained popularity after skits, memes, and fan content created using 15.ai went viral on Twitter, TikTok, Reddit, Twitch, Facebook, and YouTube. At its peak, the platform incurred operational costs of US$12,000 per month from AWS infrastructure needed to handle millions of daily voice generations; despite receiving offers from companies to acquire 15.ai and its underlying technology, the website remained independent and was funded out of the personal previous startup earnings of the developer. === 2022: Voiceverse NFT controversy === On January 14, 2022, 15 discovered that a blockchain-based company called Voiceverse had generated voice lines using 15.ai, falsely showcased them on Twitter as a demonstration of their own voice technology without permission or attribution, and sold them as NFTs. This came shortly after 15 had stated in December 2021 that they had no interest in incorporating NFTs into their work. A screenshot of the log files posted by 15 showed that Voiceverse had generated audio of characters from My Little Pony: Friendship Is Magic using 15.ai and pitched them up to make them sound unrecognizable, a violation of 15.ai's terms of service, which explicitly prohibited commercial use and required proper attribution. When confronted with evidence, Voiceverse stated that their marketing team had used 15.ai without proper attribution while rushing to create a demo. In response, 15 tweeted "Go fuck yourself," which went viral, amassing hundreds of thousands of retweets and likes on Twitter in support of the developer. The tweets showcasing the stolen voices were subsequently deleted. ==== Aftermath ==== The controversy raised concerns about NFT projects, which, according to critics, were frequently associated with intellectual property theft and questionable business practices. The incident was documented in the AI Incident Database (AIID) and the AI, Alg
1.58-bit large language model
A 1.58-bit large language model (also known as a ternary LLM) is a type of large language model (LLM) designed to be computationally efficient. It achieves this by using weights that are restricted to only three values: -1, 0, and +1. This restriction significantly reduces the model's memory footprint and allows for faster processing, as computationally expensive multiplication operations can be replaced with lower-cost additions. This contrasts with traditional models that use 16-bit floating-point numbers (FP16 or BF16) for their weights. Studies have shown that for models up to several billion parameters, the performance of 1.58-bit LLMs on various tasks is comparable to their full-precision counterparts. This approach could enable powerful AI to run on less specialized and lower-power hardware. The name "1.58-bit" comes from the fact that a system with three states contains log 2 3 ≈ 1.58 {\displaystyle \log _{2}3\approx 1.58} bits of information. These models are sometimes also referred to as 1-bit LLMs in research papers, although this term can also refer to true binary models (with weights of -1 and +1). == BitNet == In 2024, Ma et al., researchers at Microsoft, declared that their 1.58-bit model, BitNet b1.58 is comparable in performance to the 16-bit Llama 2 and opens the era of 1-bit LLM. BitNet creators did not use the post-training quantization of weights but instead relied on the new BitLinear transform that replaced the nn.Linear layer of the traditional transformer design. In 2025, Microsoft researchers had released an open-weights and open inference code model BitNet b1.58 2B4T demonstrating performance competitive with the full precision models at 2B parameters and 4T training tokens. == Post-training quantization == BitNet derives its performance from being trained natively in 1.58 bit instead of being quantized from a full-precision model after training. Still, training is an expensive process and it would be desirable to be able to somehow convert an existing model to 1.58 bits. In 2024, HuggingFace reported a way to gradually ramp up the 1.58-bit quantization in fine-tuning an existing model down to 1.58 bits. == Critique == Some researchers point out that the scaling laws of large language models favor the low-bit weights only in case of undertrained models. As the number of training tokens increases, the deficiencies of low-bit quantization surface.