As LLMs are increasingly integrated into user-facing applications, addressing biases that perpetuate societal inequalities is crucial. While much work has gone into measuring or mitigating biases in these models, fewer studies have investigated their origins. Therefore, this study examines the correlation between gender-occupation bias in pre-training data and their manifestation in LLMs, focusing on the Dolma dataset and the OLMo model. Using zero-shot prompting and token co-occurrence analyses, we explore how biases in training data influence model outputs. Our findings reveal that biases present in pre-training data are amplified in model outputs. The study also examines the effects of prompt types, hyperparameters, and instruction-tuning on bias expression, finding instruction-tuning partially alleviating representational bias while still maintaining overall stereotypical gender associations, whereas hyperparameters and prompting variation have a lesser effect on bias expression. Our research traces bias throughout the LLM development pipeline and underscores the importance of mitigating bias at the pretraining stage.

Zenfolio, the website builder and photo sharing site, recently introduced technology that applies AI to assist photographers (opens in new tab) in selecting the best photos from the thousands of shots typically taken during a photo session. The advanced image recognition technology is tremendously powerful and can make photographers more efficient than they ever dreamed. When exploring the best way to roll out AI to the photography community, Zenfolio had some fascinating discussions about corporate responsibility generally, and its obligations specifically. The concept of AI was theorized centuries ago by Greek philosophers (opens in new tab), with myths about Talos and Pandora creating chaos and destruction. Perhaps these cautionary tales about artificial beings influence our thinking today.

Another social media company is paying up due to Illinois' Biometric Information Privacy Act. Snap Inc. (the parent company of Snapchat) has reached a $35 million settlement in an Illinois class action lawsuit over its use of facial recognition technology. The lawsuit alleges that Snapchat violated the BIPA law by collecting and storing the biometric data of users who used its lenses and filters -- without their consent. Illinois residents who resided in the state after November 17th, 2015 and used Snapchat's popular AR features may be eligible for a cut of the settlement. Snap Inc. is only the latest company to get penalized under BIPA -- which requires companies to ask for consent before it can collect biometric data from users.

The new Anthony Bourdain documentary, "Roadrunner," is one of many projects dedicated to the larger-than-life chef, writer and television personality. But the film has drawn outsize attention, in part because of its subtle reliance on artificial intelligence technology. Using several hours of Bourdain's voice recordings, a software company created 45 seconds of new audio for the documentary. The AI voice sounds just like Bourdain speaking from the great beyond; at one point in the movie, it reads an email he sent before his death by suicide in 2018. "If you watch the film, other than that line you mentioned, you probably don't know what the other lines are that were spoken by the AI, and you're not going to know," Morgan Neville, the director, said in an interview with The New Yorker.

Microsoft's has obtained a patent for a "conversational chatbot of a specific person" created from images, recordings, participation in social networks, emails, letters, etc., coupled with the possible generation of a 2D or 3D model of the person, and for some reason, it has largely been interpreted by the media as an attempt to find a way to talk to the dead. For Microsoft, the reference to a specific person is appreciably broader, and as the patent notes, can also apply to "a past or present entity (or a version thereof), such as a friend, a relative, an acquaintance, a celebrity, a fictional character, a historical figure, a random entity, etc." We've been here before: in 2016, a Russian technologist, Eugenia Kuyda, co-founder of the artificial intelligence startup Replika and a chatbot specialist, tried to "reconstruct" her friend, the entrepreneur Roman Mazurenko, who had been killed in a hit-and-run accident, from the huge history of instant messaging conversations she had with him. Before that, a well-known episode of Black Mirror, "Be right back", speculated on the possibility of creating robots using all the digital data that the people they were intended to replace had generated and stored throughout their lives. The idea of a "digital ghost" of a loved one gives us the possibility of being able to'be' with someone we have lost, and confronts our awareness of the loss with our desire to deny it. A chatbot of this type, capable of recreating idioms, writing styles, specific terms or even gestures, can, as in the experiment carried out by Kuyda, help with progressive acceptance of a loss that, on many occasions, comes without a warning, and can be considered as an aid to the mourning process, to the need for a gradual closure of the emptiness created, thanks to a digital avatar.

Microsoft's chat bots may imitate the dead Microsoft has filed a patent which raises the intriguing possibility of digitally reincarnating people as a chat bot. Instead of using the conventional method of training chat bots using conversations and material from a wide sample of users, Microsoft's patent - as spotted by Ubergizmo - raises the possibility of creating a chat bot from the output of a specific person. The system would use "social data" such as "images, voice data, social media posts, electronic messages [and] written letters" to build a profile of a person. "The social data may be used to create or modify a special index in the theme of the specific person's personality," the patent states. "The special index may be used to train a chat bot to converse and interact in the personality of a specific person."

Dating websites claim attraction can be predicted from the right combination of traitsm but a new study suggests singletons have little hope of finding true love online. Researchers found computer-based algorithms could predict who's hot and who's not -- but it could not unravel the mystery of unique desire for a specific person. The findings suggest we still can't circumvent the hassle and heartache of the dating process. Researchers found computer-based algorithms could predict who's hot and who's not -- but it could not unravel the mystery of unique desire for a specific person (stock image) The researchers used data from two samples of speed daters, who filled out questionnaires about more than 100 traits and preferences and then met in a series of four-minute dates. Afterwards, the participants rated their interactions, indicating level of interest in and sexual attraction to each person they met.

Researchers have created a device that lets someone whisper to another person up to 30 meters away. The wearable device uses ultrasound to beam a person's words directly to a specific person, without anyone else hearing what's being said. While the device is still in its prototype phase, the researchers say that it could be used to help soldiers communication discreetly. The ultrasonic speaker system can be configured in two different ways: Ultra band (A) and sonic chest (B). The ultra band system consists of a bent surface with 12x5 ultrasonic speakers worn on the forehead.

I have been teaching courses on data mining for over 10 years. One of my favorite lectures focuses on the use of Markov Models for predictive analytics. I enjoy giving this lecture because it always triggers interesting reactions from my students. Since the lecture can be used to demonstrate advanced concepts (like Bayesian inference and probabilistic reasoning) as well as basic concepts (like conditional probability and statistical dependence), I use the lecture both in my graduate course and in my freshman class. I start the lecture by telling the students that I will show them how to predict the future with a cat.

When you're zipping through the air at 60 kilometres per hour, it can be hard to work out where you're going. But now drones can create detailed 3D maps as they fly – an advance that could let them navigate the world free from human input. Called Hydra Fusion, the system could one day allow drones to use a form of navigation known as simultaneous localisation and mapping to find their way in unfamiliar spaces – just as some robots do on the ground. It will also make them better at aerial surveillance. Hydra Fusion works by stitching together multiple images – in this case, consecutive frames of footage from a drone's video camera – to form a detailed 3D map while it is in the air.