Social recruiting (social hiring or social media recruitment) is recruiting candidates by using social platforms as talent databases or for advertising. Social recruiting uses social media profiles, blogs, and other Internet sites to find information on candidates. It also uses social media to advertise jobs either through HR vendors or through crowdsourcing where job seekers and others share job openings within their online social networks. Social recruiting's effectiveness and return on investment have been difficult to determine, since applicants do not usually apply through the social channels which first attracted them. In May 2013, Maximum Employment Marketing Group released the Social Recruitment Monitor, which ranks the reach, engagement, and interactivity of employers' social recruiting efforts around the world. == Social recruitment software == The social recruitment software market (a form of e-recruitment) is often included in the wider talent management software sector. Bersin & Associates valued the wider talent management market at over $2bn in 2007. Social recruitment increasingly sits at an intersection of a number of fast-moving areas including social networking, recruitment and now cloud computing. Additionally, mobile recruiting has become another hot topic, especially with the rise in tablet and smartphone usage. In 2012, there was a rise of tech companies using social recruiting applications to find and screen applicants. As more companies saw value in filling jobs by putting them on the social platforms where millions of people spend at least 37 minutes daily, there developed a much larger focus on social recruiting among the talent acquisition community. By mid-2013, many major enterprise companies such as Pepsi, Gap, AIG, and Oracle had begun effectively utilizing social recruiting software, making it clear that large corporations were open to automating or streamlining (and ultimately investing in) their social recruiting processes.
Inception (deep learning architecture)
Inception is a family of convolutional neural network (CNN) for computer vision, introduced by researchers at Google in 2014 as GoogLeNet (later renamed Inception v1). The series was historically important as an early CNN that separates the stem (data ingest), body (data processing), and head (prediction), an architectural design that persists in all modern CNN. == Version history == === Inception v1 === In 2014, a team at Google developed the GoogLeNet architecture, an instance of which won the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC14). The name came from the LeNet of 1998, since both LeNet and GoogLeNet are CNNs. They also called it "Inception" after a "we need to go deeper" internet meme, a phrase from Inception (2010) the film. Because later, more versions were released, the original Inception architecture was renamed again as "Inception v1". The models and the code were released under Apache 2.0 license on GitHub. The Inception v1 architecture is a deep CNN composed of 22 layers. Most of these layers were "Inception modules". The original paper stated that Inception modules are a "logical culmination" of Network in Network and (Arora et al, 2014). Since Inception v1 is deep, it suffered from the vanishing gradient problem. The team solved it by using two "auxiliary classifiers", which are linear-softmax classifiers inserted at 1/3-deep and 2/3-deep within the network, and the loss function is a weighted sum of all three: L = 0.3 L a u x , 1 + 0.3 L a u x , 2 + L r e a l {\displaystyle L=0.3L_{aux,1}+0.3L_{aux,2}+L_{real}} These were removed after training was complete. This was later solved by the ResNet architecture. The architecture consists of three parts stacked on top of one another: The stem (data ingestion): The first few convolutional layers perform data preprocessing to downscale images to a smaller size. The body (data processing): The next many Inception modules perform the bulk of data processing. The head (prediction): The final fully-connected layer and softmax produces a probability distribution for image classification. This structure is used in most modern CNN architectures. === Inception v2 === Inception v2 was released in 2015, in a paper that is more famous for proposing batch normalization. It had 13.6 million parameters. It improves on Inception v1 by adding batch normalization, and removing dropout and local response normalization which they found became unnecessary when batch normalization is used. === Inception v3 === Inception v3 was released in 2016. It improves on Inception v2 by using factorized convolutions. As an example, a single 5×5 convolution can be factored into 3×3 stacked on top of another 3×3. Both has a receptive field of size 5×5. The 5×5 convolution kernel has 25 parameters, compared to just 18 in the factorized version. Thus, the 5×5 convolution is strictly more powerful than the factorized version. However, this power is not necessarily needed. Empirically, the research team found that factorized convolutions help. It also uses a form of dimension-reduction by concatenating the output from a convolutional layer and a pooling layer. As an example, a tensor of size 35 × 35 × 320 {\displaystyle 35\times 35\times 320} can be downscaled by a convolution with stride 2 to 17 × 17 × 320 {\displaystyle 17\times 17\times 320} , and by maxpooling with pool size 2 × 2 {\displaystyle 2\times 2} to 17 × 17 × 320 {\displaystyle 17\times 17\times 320} . These are then concatenated to 17 × 17 × 640 {\displaystyle 17\times 17\times 640} . Other than this, it also removed the lowest auxiliary classifier during training. They found that the auxiliary head worked as a form of regularization. They also proposed label-smoothing regularization in classification. For an image with label c {\displaystyle c} , instead of making the model to predict the probability distribution δ c = ( 0 , 0 , … , 0 , 1 ⏟ c -th entry , 0 , … , 0 ) {\displaystyle \delta _{c}=(0,0,\dots ,0,\underbrace {1} _{c{\text{-th entry}}},0,\dots ,0)} , they made the model predict the smoothed distribution ( 1 − ϵ ) δ c + ϵ / K {\displaystyle (1-\epsilon )\delta _{c}+\epsilon /K} where K {\displaystyle K} is the total number of classes. === Inception v4 === In 2017, the team released Inception v4, Inception ResNet v1, and Inception ResNet v2. Inception v4 is an incremental update with even more factorized convolutions, and other complications that were empirically found to improve benchmarks. Inception ResNet v1 and v2 are both modifications of Inception v4, where residual connections are added to each Inception module, inspired by the ResNet architecture. === Xception === Xception ("Extreme Inception") was published in 2017. It is a linear stack of depthwise separable convolution layers with residual connections. The design was proposed on the hypothesis that in a CNN, the cross-channels correlations and spatial correlations in the feature maps can be entirely decoupled. Training each network took 3 days on 60 K80 GPUs, or approximately 0.5 petaFLOP-days.
Sunspring
Sunspring is a 2016 experimental science fiction short film entirely written by an artificial intelligence bot using neural networks. It was conceived by BAFTA-nominated filmmaker Oscar Sharp and NYU AI researcher Ross Goodwin and produced by film production company, End Cue along with Allison Friedman and Andrew Swett. It stars Thomas Middleditch, Elisabeth Grey, and Humphrey Ker as three people, namely H, H2, and C, living in a future world and eventually connecting with each other through a love triangle. The script of the film was authored by a recurrent neural network called long short-term memory (LSTM) by an AI bot named Benjamin. Originally made for the Sci-Fi-London film festival's 48hr Challenge, it was released online by technology news website Ars Technica on 9 June 2016. == Premise == Sunspring narrates the story of three people - H (Middleditch), H2 (Grey), and C (Ker) - set in a futuristic world and entangled with murder and love. == Cast == Thomas Middleditch as H Elisabeth Grey as H2 Humphrey Ker as C == Production == Oscar Sharp originally created the film for the 48hr Film Challenge contest of Sci-Fi-London, a film festival which focuses on science fiction. For the challenge, contestants are given a set of prompts (mostly props and lines) that have to appear in a movie they make over the next two days. It eventually contested in the festival and was nominated among the final top ten films Sharp collaborated with his longtime associate Ross Goodwin, an AI researcher in New York University to create the AI bot, which was initially called Jetson. The bot, which later came to call itself Benjamin, wrote the screenplay including stage directions and dialog. The garbled script was then interpreted by Sharp who directed the actors to construe the plot points themselves and enact the play. According to Ars Technica, the final plot turned out to be a tale of romance and murder, set in a dark future world. === Benjamin, the automatic screenwriter === Called the world's first automatic screenwriter, Benjamin is a self-improving LSTM RNN machine intelligence trained on human screenplays conceived by Goodwin and Sharp. It was trained to write the screenplay by feeding it with a corpus of dozens of sci-fi screenplays found online—mostly movies from the 1980s and 90s. == Music == The film contains a song from Brooklyn-based electro-acoustic duo Tiger and Man, with lyrics written by Benjamin using a database of 30,000 folk songs. As well as a score written by composer Andrew Orkin. == Reception == CNet called it "a beautiful, bizarre sci-fi novelty." Critic Amanda Kooser said, "...probably won't start a rush for replacing human screenwriters with machines. Some day, neural networks may get better at imitating the art of coherent storytelling, but we're not there yet. That doesn't mean "Sunspring" isn't entertaining or worthy of viewing. It is. It's a thought experiment come to life, a novelty." As of April 2019, it has surpassed 1 million views on YouTube.
Whisper (speech recognition system)
Whisper is a machine learning model for speech recognition and transcription, created by OpenAI and first released as open-source software in September 2022. It is capable of transcribing speech in English and multiple other languages, and can translate several non-English languages into English. Whisper is a weakly-supervised deep learning acoustic model, made using an encoder-decoder transformer architecture. OpenAI claims that the combination of different training data and post-training filtering used in its development has led to improved recognition of accents, background noise, and jargon compared to previous approaches. While the model does not outperform larger, more specialized models and still experiences AI hallucination, it has been showed to be useful for general sound recognition and has many applications across different industries. == Background == Speech recognition has had a long history in research; the first approaches made use of statistical methods, such as dynamic time warping, and later hidden Markov models. At around the 2010s, deep neural network approaches became more common for speech recognition models, which were enabled by the availability of large datasets ("big data") and increased computational performance. Early approaches to deep learning in speech recognition included convolutional neural networks, which were limited due to their inability to capture sequential data, which later led to developments of Seq2seq approaches, which include recurrent neural networks, which made use of long short-term memory. Transformers, introduced in 2017 by Google, displaced many prior state-of-the-art approaches across a wide range in machine learning, and started becoming the core neural architecture in fields such as language modeling and computer vision. Weakly-supervised approaches to training acoustic models were recognized in the early 2020s as promising for speech recognition approaches using deep neural networks. According to a NYT report, in 2021 OpenAI believed they exhausted sources of higher-quality data to train their large language models and decided to complement scraped web text with transcriptions of YouTube videos and podcasts, and developed Whisper to solve this task. Whisper Large V2 was released on December 8, 2022, followed by Whisper Large V3 being released in November 2023, during the OpenAI Dev Day. In March 2025, OpenAI released new transcription models based on GPT-4o and GPT-4o mini, both of which have lower error rates than Whisper. == Architecture == The Whisper architecture is based on an encoder-decoder transformer. Input audio is resampled to 16,000 Hertz (Hz) and converted to an 80-channel Log-magnitude Mel spectrogram using 25 ms windows with a 10 ms stride. The spectrogram is then normalized to a [-1, 1] range with near-zero mean. The encoder takes this Mel spectrogram as input and processes it. It first passes through two convolutional layers. Sinusoidal positional embeddings are added. It is then processed by a series of Transformer encoder blocks (with pre-activation residual connections). The encoder's output is layer normalized. The decoder is a standard transformer decoder. It has the same width and Transformer blocks as the encoder. It uses learned positional embeddings and tied input-output token representations (using the same weight matrix for both the input and output embeddings). It uses a byte-pair encoding tokenizer, of the same kind as used in GPT-2. English-only models use the GPT-2 vocabulary, while multilingual models employ a re-trained multilingual vocabulary with the same number of words. Special tokens are used to allow the decoder to perform multiple tasks: Tokens that denote language (one unique token per language). Tokens that specify task (<|transcribe|> or <|translate|>). Tokens that specify if no timestamps are present (<|notimestamps|>). If the token is not present, then the decoder predicts timestamps relative to the segment, and quantized to 20 ms intervals. <|nospeech|> for voice activity detection. <|startoftranscript|>, and <|endoftranscript|> . Any text that appears before <|startoftranscript|> is not generated by the decoder, but given to the decoder as context. Loss is only computed over non-contextual parts of the sequence, i.e. tokens between these two special tokens. == Training data == The training dataset consists of 680,000 hours of labeled audio-transcript pairs sourced from the internet using semi-supervised learning. This includes 117,000 hours in 96 non-English languages and 125,000 hours of X→English translation data, where X stands for any non-English language. Preprocessing involved standardization of transcripts, filtering to remove machine-generated transcripts using heuristics (e.g., punctuation, capitalization), language identification and matching with transcripts, fuzzy deduplication, and deduplication with evaluation datasets to avoid data contamination. Speechless segments were also included to allow voice activity detection training. For the files still remaining after the filtering process, audio files were then broken into 30-second segments paired with the subset of the transcript that occurs within that time. If this predicted spoken language differed from the language of the text transcript associated with the audio, that audio-transcript pair was not used for training the speech recognition models, but instead for training translation. The model was trained using the AdamW optimizer with gradient norm clipping and a linear learning rate decay with warmup, with batch size 256 segments. Training proceeded for 1 million updates (approximately 2-3 epochs). No data augmentation or regularization, except for the Large V2 model, which used SpecAugment, Stochastic Depth, and BPE Dropout. The training used data parallelism with float16, dynamic loss scaling, and activation checkpointing. === Post-training filtering === After training the first model, researchers ran it on different subsets of the training data, each representing a distinct source. Data sources were ranked by a combination of their error rate and size. Manual inspection of the top-ranked sources (high error, large size) helped determine if the source was low quality (e.g., partial transcriptions, inaccurate alignment). After training, it was fine-tuned to suppress the prediction of speaker names and low-quality sources were then removed. == Capacity == While Whisper does not outperform models which specialize in the LibriSpeech dataset, when tested across many datasets, it is more robust and makes 55.2% fewer errors than other models. Whisper has a differing error rate with respect to transcribing different languages, with a higher word error rate in languages not well-represented in the training data. The authors found that multi-task learning improved overall performance compared to models specialized to one task. They conjectured that the best Whisper model trained is still underfitting the dataset, and larger models and longer training can result in better models. Third-party evaluations have found varying levels of AI hallucination. A study of transcripts of public meetings found hallucinations in eight out of every 10 transcripts, while an engineer discovered hallucinations in "about half" of 100 hours of transcriptions and a developer identified them in "nearly every one" of 26,000 transcripts. A study of 13,140 short audio segments (averaging 10 seconds) found 187 hallucinations (1.4%), 38% of which generated text that could be harmful because it inserted false references to things like race, non-existent medications, or violent events that were not in the audio. == Applications == The model has been used as the base for many applications, such as a unified model for speech recognition and more general sound recognition. Whisper has also been integrated into the workflow of biomedical research. In 2025, a study on Alzheimer's disease detection used the model to transcribe spontaneous speech recordings. The transcripts that were generated by the model were combined with LLM vector embeddings and traditional classifiers to help classify the patients' health. Another application is when OVALYTICS incorporated Whisper to transcribe YouTube videos and automate content moderation systems, which improved its detection of offensive content. The model has also been used in academic libraries and cultral heritage institutions to generate transcripts and captions for their digitized audiovisual collections. In a 2025 case study, Emory University Libraries found that Whisper reduced the labor used in transcription by around 30-35%, shifting work from text creation to text correction. However, human review is still necessary to make sure accuracy, formatting, and accessibility are all standard.
DEAP (software)
Distributed Evolutionary Algorithms in Python (DEAP) is an evolutionary computation framework for rapid prototyping and testing of ideas. It incorporates the data structures and tools required to implement most common evolutionary computation techniques such as genetic algorithm, genetic programming, evolution strategies, particle swarm optimization, differential evolution, traffic flow and estimation of distribution algorithm. It is developed at Université Laval since 2009. == Example == The following code gives a quick overview how the Onemax problem optimization with genetic algorithm can be implemented with DEAP.
TeaOnHer
TeaOnHer is a male-oriented dating surveillance mobile app that allows men to anonymously rate and comment on women they are dating. It was set up in response to the existence of Tea, a female-oriented dating app that allowed women to rate and comment on men. In 2025, Cosmopolitian magazine described it as America's second most popular mobile app, with it being the second most popular app in the lifestyle section of Apple's App Store. The TeaOnHer app has fewer features than the rival Tea app, focusing instead on anonymous commenting. It is listed as having been developed by a company called Newville Media Corporation. TechCrunch reported in 2025 that TeaOnHer had leaked credentials of some of its users.
Gundam Build Metaverse
Gundam Build Metaverse (Japanese: ガンダムビルドメタバース, Hepburn: Gandamu Birudo Metabāzu) is a Japanese original net animation anime mini-series produced by Sunrise Beyond, and the fifth series within the Gundam Build Series sub-series. The series celebrates the 10th anniversary of the Gundam Build franchise, including characters from the previous installments. == Plot == The story is set in the same universe of the Gundam Build series in an online metaverse space where users can use avatars to move around and interact with other users, including conducting Gunpla (Gundam plastic model) battles with them. The story centers on Rio Hōjō, a boy who lives in Hawaii, and who learns how to build Gunpla from a local hobbyist named Seria Urutsuki. In the metaverse, a figure known as Mask Lady teaches him the art of Gunpla battling, and he strives to get better at it every day. With his custom Lah Gundam, he seeks out ever stronger opponents. == Characters == === Main characters === Rio Hojo (ホウジョウ・リオ, Hōjō Rio) Voiced by: Chika Anzai A young boy from Hawaii who is an enthusiast of Gunpla Battle and is an apprentice of the mysterious Diver "Mask Lady". Rio's Gunpla is the Lah Gundam, modeled after an entry-grade RX-78-2 Gundam, from the original Mobile Suit Gundam anime series. Seria Urutsuki (ウルツキ・セリア, Urutsuki Seria) / Mask Lady (マスクレディー, Masuku Reidi) Voiced by: Rio Tsuchiya A clerk at a local hobby shop and the instructor at their Gunpla class, Seria becomes Rio's Gunpla mentor using the alias "Mask Lady". Seria's Gunpla is the ZGMF-X20A-PF Gundam Perfect Strike Freedom Rouge, based on both the MBF-02 Strike Rouge and the GAT-X105+AQM/E-YM1 Perfect Strike Gundam from Mobile Suit Gundam Seed and the ZGMF-X20A Strike Freedom Gundam from Mobile Suit Gundam Seed Destiny. === Returning characters === Fumina Hoshino (ホシノ・フミナ, Hoshino Fumina) Voiced by: Yui Makino A veteran Gunpla Battler from the early days of the sport and the Leader of "Team Try Fighters", she works as an advertiser and announcer within the Metaverse realm. Tatsuya Yuuki (ユウキ・タツヤ, Yūki Tatsuya) / Meijin Kawaguchi III (三代目メイジン・カワグチ, Sandaime Meijin Kawaguchi) Voiced by: Takuya Satō A builder and three-times Gunpla Battle world champion who inherited the name of the legendary Meijin Kawaguchi, known as "Meijin Kawaguchi III", and still the current title holder. His newest Gunpla is the Gundam Amazing Barbatos Lupus based on the ASW-G-08 Gundam Barbatos Lupus from Mobile Suit Gundam: Iron-Blooded Orphans. Riku Mikami (ミカミ・リク, Mikami Riku) / Riku (リク) Voiced by: Yūsuke Kobayashi The Founder and former leader of the legendary force, "Build Divers". His Gunpla is the Gundam 00 Diver Arc, the latest version of the original GN-0000DVR Gundam 00 Diver from Gundam Build Divers, incorporating elements from the 00 Gundam from Mobile Suit Gundam 00 and the Gundam AGE-FX from Mobile Suit Gundam AGE. Sarah (サラ, Sara) Voiced by: Haruka Terui An EL-Diver and member of the Build Divers. Momoka Yashiro (ヤシロ・モモカ, Yashiro Momoka) / Momo (モモ) Voiced by: Nene Hieda Member of Build Divers. Her gunpla is the MOMOKAPOOL (R×R), an upgraded version of her PEN-01M Momokapool from Gundam Build Divers Aya Fujisawa (フジサワ・アヤ, Fujisawa Aya) / Ayame (アヤメ) Voiced by: Manami Numakura Member of Build Divers. Her Gunpla is the F-Kunoichi Kai, an SD Gunpla based on the F91 Gundam F91 from Mobile Suit Gundam F91. Sei Iori (イオリ・セイ, Iori Sei) Voiced by: Mikako Komatsu A builder and one time Gunpla Battle World Champion. His current Gunpla is the GAT-X105B/EG Build Strike Exceed Galaxy, the latest version of the original GAT-X105B Build Strike Gundam from Gundam Build Fighters. Aria von Reiji Asuna (アリーア・フォン・レイジ・アスナ, Arīa fon Reiji Asuna) Voiced by: Sachi Kokuryu A prince from the country called Arian that exists within a space colony in another dimension, who became friends with Sei Iori and together won the Gunpla Battle World Championship. He somehow manages to log into the metaverse to reunite with his friend, piloting the SB-011 Star Burning Gundam. Sekai Kamiki (カミキ・セカイ, Kamiki Sekai) Voiced by: Kazumi Togashi A veteran builder and former member of Team Try Fighters. He is currently the Japanese National representative Champion. In the series he develops a rivalry relationship with Hiroto similar to that of Kyoya and Rommel. His current Gunpla is the Shin Burning Gundam, the latest version of the original KMK-B01 Kamiki Burning Gundam from Gundam Build Fighters Try which is based on the Burning Gundam and Master Gundam. Hiroto Kuga (クガ・ヒロト, Kuga Hiroto) / Hiroto (ヒロト, Hiroto) Voiced by: Chiaki Kobayashi A veteran diver, the one responsible for discovering more EL-Divers, and a former member of the legendary force "Avalon", who later joined the unofficial, "BUILD DiVERS" and eventually became the current Force Leader, and as well as the current title holder of "Hero of Gunpla". In the third episode he is the only Build Diver member who participates in the tournament, while his fellow force-mates are in the audience routing for him and Rio. His Gunpla is the Plutine Gundam, which is a combination of his Core Gundam II Plus, upgraded from the Core Gundam II featured in Gundam Build Divers Re:Rise equipped with the Pluto Armor. Magee (マギー, Magī) Voiced by: Taishi Murata A flamboyant veteran Diver who owns a shop in the metaverse and is an acquaintance of Seria's. Freddie (フレディ, Furedi) Voiced by: Ai Kakuma An alien anthropomorphic dog boy from planet Eldora, a support member to both Build Diver teams, who manages to access the metaverse from his home planet along his fellow Eldorans. Ogre (オーガ, Ōga) Voiced by: Wataru Hatano Kyoya Kisugi (キスギ・キョウヤ, Kisugi Kyōya) / Kyoya Kujo (クジョウ・キョウヤ, Kujō Kyōya) Voiced by: Jun Kasama Leader of the legendary force "Avalon" and the reigning and current title holder of "World Champion". He along with Hiroto Kuga, Maria Urutsuki, and Tatsuya Yuuki are currently at the top of the entire gunpla world community. His current gunpla is an recolored version of his AGE-TRYMAG Gundam TRY AGE Magnum from Gundam Build Divers Re:Rise. Susumu Sazaki (サザキ・ススム, Sazaki Susumu) Voiced by: Ryo Hirohashi Kaoruko Sazaki (サザキ・カオルコ, Sazaki Kaoruko) Voiced by: Ryo Hirohashi Mahiru Shigure (シグレ・マヒル, Shigure Mahiru) Voiced by: Rinko Natsuhi Keiko Sano (サノ・ケイコ, Sano Keiko) Voiced by: Ami Naito === Others === Maria Urutsuki (ウルツキ・マリア, Urutsuki Maria) / Mascarilla (マスカリージャ, Masukarīja) Voiced by: Ai Kakuma A mysterious masked woman with a harsh rivalry with Seria and a similar avatar as hers, she is later revealed as Seria's younger sister Maria, who began to loathe her sister after she quit on their dream to fight for the title of Lady Kawaguchi. She later obtains the title, becoming "Lady Kawaguchi VII". Jeff (ジェフさん, Jefu-san) Voiced by: Kenta Miyake A distant relative of Seria and Maria's and owner of the hobby shop where Seria lives. Mellow Neige (メロウ・ネージュ, Merō Nēju) Voiced by: Chikano Ibuki A sentient A.I. who is the current publicity face of the Gunpla Metaverse. == Episodes ==