Siri accidentally triggered a police call, OpenAI used cheap labour, and other AI news

We aim to inform readers not only about events in the bitcoin industry, but also about developments in adjacent technology sectors — cybersecurity and the world of artificial intelligence (AI).

ForkLog AI has gathered the most important AI news from the past week.

Apple unveils new M2 processors and computers

Apple unveiled the M2 Pro and M2 Max chips. The new models are built on the next generation 5-nanometre process.

M2 Pro consists of 40 billion transistors. The die size increased by 20% compared with the M1 Pro and by 100% over the M2. The new model features 10 or 12 cores — six or eight high-performance and four efficiency cores.

According to Apple, CPU performance in the M2 Pro is up 20% and GPU up 30% versus the M1 Pro.

The M2 Max includes 67 billion transistors and is equipped with a 12-core processor of the new generation. Performance rose 30% versus the M1 Max.

Apple integrated the chips into 14- and 16-inch MacBook Pro models. Prices for the updated notebooks start at $1999 and $2499 respectively.

Apple also introduced the compact Mac mini with M2 and M2 Pro processors.

Google will release a competitor to ChatGPT

In 2023, Google plans to launch a competitor to OpenAI’s ChatGPT.

According to DeepMind cofounder and chief executive Demis Hassabis, the company is considering making Sparrow available for private beta testing.

The system is based on the Chinchilla language model and trained with user feedback. Later the chatbot will be able to generate answers using up-to-date information from the internet with sources cited.

GitHub Copilot Labs added brushes for editing code

GitHub Copilot Labs instrument received brushes “Photoshop-like” for editing code with AI.

They allow turning lines into “more readable” ones, specifying variable types, fixing simple errors, adding debugging operators or creating custom tools for one’s tasks. To do this, users need to highlight the necessary lines and choose the appropriate brush.

According to the developers, more such useful tools will appear in the future. Programmers will also be able to store the brushes they create.

Media: OpenAI used cheap labour to reduce the toxicity of ChatGPT

OpenAI outsourced Kenyan workers to filter traumatising content from ChatGPT. They were paid $1.32-2 per hour depending on seniority and productivity.

According to Time, in November 2021 the AI startup via Sama hired contractors from Kenya who labelled and filtered tens of thousands of text fragments from the AI training dataset. Some excerpts contained detailed descriptions of sexual violence against children, murder, torture, self-harm, bestiality and incest.

One of the workers said that processing NSFW data was torture. After reading such texts, “it’s impossible to stop imagining the described scenes,” he added.

In February 2022 Sama ended collaboration with OpenAI — eight months ahead of schedule. This happened due to the traumatising nature of the work and, partly, because of a report about a similar deal between the company and Meta. At that time outsourced workers were labeling NSFW videos for $1.50 per hour.

AI wrote one in five exams in Australian universities

In Australian universities, AI was found to be used in one in five examination papers.

The junior lecturer at Deakin University, Sally Brandon, found that 10 of 54 master’s texts included “significant, detectable assistance from ChatGPT.” External plagiarism-checking software also flagged use of additional tools by students, including AI-writing-enhancement tools.

Brandon now awaits confirmation whether the detector’s results will be treated as a formal breach of academic integrity.

The law lecturer from Curtin University, Anna Bann, noted that ChatGPT “will revolutionise” education, but is currently more accurate for US-centric responses. This is encouraging, though it may not always be the case, she added.

Teachers also found that students used other chatbots like ChatSonic, which pull data from Google.

Startup released an AI writing assistant

DeepL released a beta version of its AI text-editing tool Write.

The system detects and fixes grammatical errors. The AI assistant also offers suggestions to improve style and word choice.

DeepL said they plan to develop Write further. In the future it will focus not on content but on the elusive tones and phrasing, they added.

Chinese scientists developed an autonomous AI ship

In China they launched the autonomous maritime reconnaissance vessel with AI Zhuhaiyun. The ship can carry and operate dozens of UAVs.

The vessel is 88.5 m long, 14 m wide, and displaces 2100 t. It is capable of a maximum speed of 18 knots.

AI is responsible for steering the vessel. Zhuhaiyun also supports remote control by an operator.

In 2023 the vessel is planned for use at sea for mapping, monitoring, patrol, and sample collection.

Opinion: by 2025 AI will generate 90% of content

By 2025, 90% of online content could be generated by artificial intelligence.

According to Nina Shik, a thought leader in AI, the technology is exponential. In 2023 it will help many companies across various sectors adopt generative algorithms.

Shik also said that over time AI capabilities will grow. Systems that shock or scare people will end up being quite “attractive” to use.

AI learned to track people using Wi‑Fi signals

Researchers from Carnegie Mellon University presented a method to track people through walls using AI and Wi‑Fi signals.

The researchers used the DensePose pixel-mapping system, on which they built a deep learning neural network. The algorithm maps the phase and amplitude of Wi‑Fi signals sent and received by routers to coordinates on the body and creates a 3D model.

Researchers said that detecting people indoors via wireless networks could replace traditional RGB cameras. Using Wi‑Fi helps overcome obstacles such as poor lighting and occlusion, which video sensors’ lenses often face, they added.

Siri accidentally triggered a police response to a boxing training

In Australia, the voice assistant Siri on an Apple Watch accidentally triggered a police response to a boxing training session.

Turns out the trainer’s smart watch repeatedly activated during the session. He often shouted the countdown “1-1-2” to the student and said “good shot.”

However the voice assistant interpreted the trainer’s words as a request to call local emergency services 112. After hearing cries of “shots,” operators dispatched a crew of 15 armed police officers and several ambulances.

As a result of the error, no one was hurt.

The week’s most significant AI deals

From 15 to 21 January 2023 AI startups raised over $36.7m. Here are the most interesting deals.

• CloseFactor, the creators of a platform for automating routine marketing tasks, raised $15.2m.
• Presso raised $8m to develop robotic dry-cleaning.
• Healthcare AI startup Qritive raised $7.5m.
• Scenario raised $6m to expand its team of developers for a platform that creates custom AI-generated game asset images.

Also on ForkLog:

• Artists sued Stability AI and Midjourney over training algorithms on their images.
• AI turned popular crypto assets into Transformers.
• Researchers urged the world to regulate deepfakes.
• Tesla misled in autopilot advertising for 2016.
• Sam Altman spoke about the future of OpenAI and its relationship with Microsoft.
• Microsoft will lay off 10,000 employees.

Subscribe to ForkLog’s Telegram news: ForkLog AI — all the news from the AI world!