Boston Dynamics has updated robops, Meta has introduced a competitor GPT-3 other news from the world And
We strive to inform readers not only about events in the bitcoin industry, but also to talk about what is happening in related technological fields - cybersecurity and in the world of artificial intelligence (AI).
ForkLog AI has collected the most important news from the AI world over the past week.
Google has allowed users to request the removal of personal information from search results.
Meta has introduced a large open-source language model with 175 billion parameters.
PrivatBank has launched a voice biometrics function for quick authorization when contacting a hotline.
The robomobile broke the world speed record among drones.
Boston Dynamics has updated Spot robops.
The most important deals from the field of AI over the past week.
Google has allowed users to delete personal information from search results
Google has given users the opportunity to remove personal information from search results.
According to the updated privacy policy of the service, at the request of the client, the system will stop showing certain web pages containing his personal data. Google can remove:
confidential government identification data, including social security number;
bank account number;
credit card number;
handwritten signature images;
photos of identity documents;
medical records;
personal contact information, including address, phone number and email;
login details for any account.
Google also promised to delete personal intimate photos, pornographic photos and images with a demonstration of violence.
Meta has introduced an open source language model
Meta has published a language model code with 175 billion OPT-175B parameters. The neural network is available to AI researchers on request for non-commercial use.
The company said that they are sharing an algorithm trained on publicly available data to help the community understand the principles of this technology. Meta did not rule out that scientists will be able to rid the models of toxicity and bias.
The researchers trained the OPT-175B neural network using 992 Nvidia 80GB A100 GPUs, achieving a performance of 147 TFLOPS per chip. According to Meta, researchers do not need to create an algorithm and train it from scratch, since the tech giant provides code for its deployment on 16 Nvidia V100 GPUs.
The company has also opened access to small pre-trained models of up to 66 billion parameters for everyone.
Volkswagen has entered into a partnership with Qualcomm in the field of drones
Volkswagen carmaker has signed a long-term cooperation with chipmaker Qualcomm to develop self-driving technology.
There are no details of the deal with Qualcomm. It is known that the company will supply systems-on-a-chip (SoC) for the Cariad platform, providing autonomous driving of the fourth level. Volkswagen integrates processors into on-board computers of all brands of cars belonging to the concern.
The automaker also said that it is already negotiating with the Israeli supplier of specialized chips Mobileye.
PrivatBank has launched voice biometrics technology
Ukrainian bank PrivatBank has launched a voice authentication system for customers who have contacted the support service by phone.
To use the function, the user needs to create a "cast" of the voice through the Privat24 application. After that, with each subsequent call to the hotline, the algorithm will be able to "quickly" recognize the client.
After enabling voice biometrics, users will not have to answer additional security questions, the bank said.
Study: YouTube Kids Algorithms Promote Content about Drugs and Weapons
The Tech Transparency Project has discovered that YouTube Kids algorithms offer children videos about weight loss, skin whitening, drug subculture and firearms.
The researchers registered three accounts for one of the age groups. Among the videos for "younger children" was a cooking show on the theme "Breaking Bad" and the process of creating a motorhome in Minecraft, "where crystal methamphetamine is brewed."
The platform also showed "younger children" a rating of shock absorbers that protect shooters from the recoil of weapons, and "older children" - instructions for assembling a shelf with a hidden compartment for a pistol.
The company said that thanks to automatic filters, moderation of people and parental reviews, YouTube Kids videos are suitable for family viewing, but "no system is perfect." After verification, the developers removed or set age limits for tagged videos in the Kids app.
Google fired an AI researcher because of criticism of the company
Google has fired another artificial intelligence researcher because of criticism of the tech giant's work.
According to The New York Times, in March 2022, the company suspended scientist Satrajit Chatterjee for challenging an article that algorithms can design chips more efficiently than humans.
Google representatives declined to comment on the dismissal, but indicated that it happened "for a good reason." The original article was "thoroughly checked" and reviewed, and the work challenging it "did not meet the company's standards," they added.
Earlier, Google suspended two leading researchers of the ethics of artificial intelligence, Timnit Gebra and Margaret Mitchell, which led to a series of other dismissals of their own volition.
Robocar has set a new speed record among drones
The drone of the team of the Milan Polytechnic Institute PoliMOVE broke the world speed record among robocars, accelerating to 309.3 km/h.
The car was traveling on a straight runway at the Kennedy Space Center. The robocar reached a speed of 309.3 km/h in 1 km with two consecutive attempts in opposite directions to eliminate the influence of wind.
Meta used AI to design "green" concrete
Meta company will build new data centers for the metaverse from "green" concrete created with the help of AI. The developed formula of the material allowed to reduce carbon emissions by 40% and exceed the strength requirements.
The tech giant, together with scientists from the University of Illinois at Urbana-Champaign, MIT and IBM, trained a model on a dataset that contains 1000 concrete formulas and structural characteristics, including data on compressive strength for 7 and 28 days. The team also determined the carbon footprint of the samples of the resulting concrete mixture.
The researchers then selected five "most promising" options and iteratively refined them. As a result, they derived a formula in which 50% of cement was replaced by fly ash and slag.
Meta has teamed up with concrete supplier Ozinga to improve the mix and test it in real conditions.
Boston Dynamics has updated Spot Robops
The company Boston Dynamics showed a Spot robot at the factory in the new video No Time To Dance.
In the video, the four-legged device performs a number of tasks in industrial conditions. The robot takes sensor readings, scans equipment using an infrared camera, generates work assignments for employees and dances.
Boston Dynamics also demonstrated some updates to the robops.
Among the innovations:
five stereo cameras capturing full-color images and transmitting enhanced depth of scene data;
tablet controller with bigger screen and longer battery life;
a new system on a chip to accelerate AI processing on the device and compatibility with 5G.
Scientists have developed an algorithm to identify actions in a video
Engineers from the Massachusetts Institute of Technology have developed an AI algorithm for identifying actions in videos and marking them without human help.
The model learns to represent data by capturing concepts common to visual and audio modalities. The system will determine that the crying child in the video is associated with the spoken word "crying" in the audio clip.
The scientists used trait training. The algorithm receives raw information, including videos and their corresponding text signatures, and encodes it, extracting signs of objects and actions from the video.
The AI then displays the data in the representation space. The model groups similar information together as separate points, expressing each in a separate word. The researchers limited the model to 1,000 words for labeling vectors.
The method also uses a common representation space. This allows the algorithm to study the relationship between a video with a juggling person and an audio recording with the words "juggling".
Demonstration of the algorithm. Data: MIT.
The most important AI deals of the last week
From May 1 to May 7, 2022, startups in the field of artificial intelligence raised over $720 million. Here are the most interesting deals.
Anthropic has raised $580 million for research to make AI more reliable and explicable.
Traceable AI received $60 million to provide services to protect APIs from cyber attacks.
The developer of credit card programs for Concerto brands raised $21.2 million.
Accern received $20 million to develop an AI application for analyzing financial documents on the Internet.
The creators of spatial intelligence algorithms for SLAMcore robots have raised $16 million.
Druid has raised $15 million to scale the platform for constructing chatbots.
Developers of the AI platform for creating 3D animation Kinetix received $ 11 million.
Galileo raised $5.1 million for the development of the service to simplify the development of AI algorithms.
RECON Labs has raised $4.4 million to improve the platform for turning 2D images of a product into its 3D model.