
OpenAI Lets Developers Customize GPT-3, AI Brings Children’s Drawings to Life, and Other AI News
We aim to inform readers not only about developments in the bitcoin industry, but also about what is happening in related technology spheres—cybersecurity, and now the world of artificial intelligence (AI).
ForkLog AI has gathered the most important AI news of the past week.
- The US Treasury added eight Chinese AI companies to the ‘blacklist’.
- “MTS AI” will buy VisionLabs, a facial-recognition developer, for 7 billion rubles.
- Russian researchers have developed a device that allows controlling computer programs by thought.
- OpenAI has allowed users to modify the GPT-3 language model.
- Apple released iOS 15.2 with a feature to recognise nude photos for teen protection.
- Honda will deploy connected cars with AI to assess road markings.
- Meta taught AI to ‘animate’ children’s drawings.
- The week’s most important AI deals.
US Imposes Sanctions on Chinese AI Companies
The US Treasury added eight Chinese technology companies to the economic ‘blacklist’, blocking the sale of shares to American investors. The firms are suspected of monitoring Uyghur Muslims in Xinjiang.
Sanctions will include developers of AI algorithms and systems in cybersecurity, cloud technologies and video surveillance. Among them are drone maker DJI, as well as AI startups in biometric identification CloudWalk Technology and Megvii.
The listed developers are already on the sanctions list, prohibiting US companies and agencies from engaging in commercial cooperation with them.
SenseTime Halts IPO Due to US Sanctions
The Chinese AI startup SenseTime halted its initial public offering on the Hong Kong exchange after being added to the US economic blacklist, which blocks the sale of its shares to Americans.
The company will defer the IPO and publish an additional prospectus with amendments and an updated listing timetable. It will also reimburse all investor applications submitted, without accrued interest.
SenseTime planned to place 1.5 million shares and raise at least $767 million.
«МТС AI» купит разработчика систем распознавания лиц VisionLabs
A subsidiary Intema of "MTS AI" will buy VisionLabs, a developer of facial-recognition systems, for 7 billion rubles.
The Russian operator will own a controlling stake in VisionLabs’ founders (51.77%), as well as structures of Sber (25.07%) and venture fund AFK Sistema’s Sistema_VC (23.16%).
Under the terms of the deal, the company will continue to operate autonomously and maintain its existing business strategy and team. The board will be chaired by Alexander Khanin, who, together with co-founder Ivan Laptev, plans to convert their shares into ownership stakes in the MTS AI structures.
According to MTS AI CEO Alexander Khanin, the acquisition will strengthen the company’s product portfolio in video analytics. It can be used to unlock the potential of the MTS digital services ecosystem and for corporate clients in Russia and internationally, he added.
In Russia, a device for interacting with computers by thought has been developed
Russian researchers have created the ‘Balalaika’ device for controlling computer applications using gestures and thoughts.
The device may help post-stroke patients in rehabilitation to recover motor and cognitive functions. It could also be useful in esports to improve players’ performance by reading their mental intent to act.
According to the developers, intuitive and reliable interaction with programs is ensured by AI integrated into the device.
OpenAI Lets Developers Customize GPT-3
The non-profit OpenAI has allowed clients to modify the GPT-3 language model.
Developers can train the algorithm on their own datasets, using existing sets of ‘almost any size and type’, or gradually adding information based on user feedback. This will increase the model’s accuracy and create new versions tailored to specific tasks or applications.
According to the developers, thanks to fine-tuning, one client managed to increase the AI’s correct responses from 83% to 95%, while another reduced errors by 50%.
Apple Releases iPhone Update with New Safety Features
Apple released the iOS 15.2 update. It adds an algorithm to detect intimate photos in messages involving minors and a feature to identify non-original parts.
The company also released the App Privacy Report tool. It lets users see how often apps access their personal data, camera and microphone over the past seven days.
Also introduced is Digital Legacy, which transfers information from the user’s iCloud account to a trusted person in the user’s Apple ID after death.
Microsoft Develops AI Tool to Detect Bugs in Code
Microsoft developed the BugLab deep-learning algorithm for detecting and fixing bugs in software code.
The system is based on two competing neural networks that during training “play hide-and-seek”. The first analyzes code and injects a bug, while the second determines whether there is a bug, where it is, and how to fix it.
Researchers tested the tool on a labeled test dataset and real open-source GitHub projects. It automatically identified 26% of errors and fixed them, and also uncovered 19 previously unknown typos.
According to the developers, BugLab performs about 30% better than existing alternatives, but it requires refinement due to numerous false positives.
Twitter Adds Automatic Video Captions
Twitter has launched a feature that automatically generates captions for videos.
Where are video captions when you need them? They’re here now automatically on videos uploaded starting today.
Android & iOS: auto-captions will show on muted Tweet videos; keep them on when unmuted via your device’s accessibility settings
Web: use the “CC” button to turn on/off pic.twitter.com/IHJAI31IvX— Twitter Support (@TwitterSupport) December 14, 2021
The captioning system will support most languages in the near future and will be available to all app users on iOS, Android and in browsers.
According to The Verge, captions are generated only for newly uploaded videos on the platform. Users will not be able to report errors or low-quality generation; however, developers “are always looking for ways to improve accessibility”, the source notes.
Honda Cars to Help Assess Road Marking Quality
The Honda Research Institute is using connected cars to monitor the condition of road surfaces.
Vehicles equipped with GPS sensors and on-board cameras will detect substandard markings in real time.
The AI-based system will analyse lane markings on both sides of the vehicle and classify them into four colour zones. Green and yellow indicate respectively ‘ideal’ or ‘good’ road markings, red — requires repair, and grey indicates no markings.
The collected data will be sent to competent authorities and services to fix potentially dangerous sections of the road.
AI Learns to ‘Animate’ Characters from Children’s Drawings
Meta has developed an AI that ‘animates’ children’s drawings.
The algorithm recognises the character in a sketch and separates it from other objects. It then builds its skeleton, places key joints that make limb movements more realistic, and animates the figure.
Meta stated that the project could help create a conscious AI that understands the world from a human perspective.
A demo version of the tool is available at the link.
The Week’s Most Significant AI Deals
From 11 to 18 December 2021, AI startups raised more than $726 million. Here are the most interesting deals.
- Rubbles raised $6 million to expand its product line for optimising business processes.
- Phytoform raised $5.7 million to scale genome-editing technology for agricultural crops.
- The creators of the unified business communication platform Dialpad raised $170 million.
- Cylus raised $30 million to protect railways and metros worldwide from cyber threats.
- Papaya raised $50 million to expand the team developing an app to facilitate bill payments.
- Signal AI raised $50 million to support its platform for gathering data from public sources.
- Wowzi raised $3.2 million to connect users of social networks with advertisers.
- Enquire AI raised $5.5 million to improve expert-search tools in the field.
- Guardio raised $47 million to support its web extension for protection against cyberattacks.
- ForwardX Robotics raised $31 million.
Also on ForkLog:
- In South Korea, facial recognition to combat COVID-19.
- The UN chief called to take measures against ‘killer robots’.
- China spoke out against the use of AI in military applications.
- The French regulator ordered Clearview AI to delete citizens’ data.
- Media: Huawei tracked users for the Chinese government.
Subscribe to ForkLog AI news on Telegram: ForkLog AI — all the AI news!
Рассылки ForkLog: держите руку на пульсе биткоин-индустрии!