OpenAI losses, $10,000 from deepfake tracks 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 fields — cybersecurity and the world of artificial intelligence (AI).

ForkLog AI has assembled the week’s most important AI news.

Media: OpenAI losses reach $540m in 2022

In 2022, OpenAI posted losses of $540 million due to the development of ChatGPT.

According to The Information, the bulk of the company’s costs goes to providing the robust computing power required to generate the chatbot’s responses to user queries.

Despite backing from Microsoft, rising demand for ChatGPT will drive further cost growth for the AI lab.

According to the report, OpenAI CEO Sam Altman privately suggested that in the coming years the company would seek to raise around $100 billion as it works on artificial general intelligence.

Elon Musk announced a free trial version of FSD

The head of Tesla, Elon Musk, promised to offer users in North America a free trial of FSD when the technology becomes “super smooth, not just safe.”

According to the billionaire, later it will be rolled out worldwide once they are convinced the autonomous driving system can operate on local roads and its use is approved by the regulators of the country.

He also spoke about capabilities of the next FSD update. He said that software version 11.4 could be considered 12.0 due to numerous significant improvements.

According to the head of Tesla, FSD 12 will feature “end-to-end AI” — from images to steering, brakes and acceleration.

In China, author of fake news created with ChatGPT, arrested

Chinese police arrested a man who allegedly used ChatGPT to create fake news about a train crash on 25 April 2023.

According to SCMP, the detainee Hong allegedly uploaded several popular stories to the chatbot and created different versions of the article. He then published the “fake” news through more than 20 accounts on the Baijiahao platform.

Police said the suspect allegedly owned a company that runs several blogs. He used “false articles” to obtain around 15,000 clicks, the authorities added.

According to police, the detainee admitted circumventing the service’s restrictions on distributing duplicate content from multiple accounts.

Hong is suspected of “stirring up trouble and provoking disturbances,” a political crime that carries a 5- to 10-year prison term.

Clearview AI fined again in France

France fined Clearview AI for ignoring prior regulator orders tied to GDPR.

In October 2022, the National Commission on Informatics and Liberty (CNIL) ordered to pay the startup 20 million euros. The regulator also ordered the company to cease collecting and processing citizens’ data.

If the order is not complied with within two months, the regulator promised to fine Clearview AI €100,000 for each day of delay.

According to CNIL, the company did not provide evidence of compliance. Consequently, the regulator imposed a €5.2 million penalty.

In the startup’s PR agency LAKPR Group, they said that Clearview AI has no presence or customers in France or the EU, and therefore its activity is not subject to GDPR.

Meta released an algorithm for processing multiple data modalities

The company Meta unveiled the first open AI model, ImageBind, capable of processing six data modalities. You can test the system on the project’s site.

The algorithm combines text, images and video, audio, depth information, temperature data, and motion readings generated by inertial measurement devices.

According to the company, the model is a research project with no consumer or practical applications. However, the technology opens new possibilities for developers of generative AI.

Meta noted that future algorithms could process another stream of sensory information, including touch, speech, smell and fMRI brain signals. This will enable more comprehensive AI models that are human-centered, they added.

Wendy’s uses AI to automate car-based order taking

Wendy’s fast-food chain and Google have created a chatbot to handle car-based orders from customers.

The system is based on a large language model developed by the tech giant’s engineers. It can recognise words, names and abbreviations unique to Wendy’s. It was also trained to understand different dialects and accents, “without being distracted” by background noise.

In addition, developers added generative AI to minimise “rigidity and roboticness” of the system when interacting with people.

In June 2023, Wendy’s will test the technology at a Columbus, Ohio, restaurant.

AI studies to learn bacteria with no prior knowledge

Researchers at the University of Michigan developed an AI platform, BacterAI, capable of studying bacteria with no prior knowledge.

The system creates a data set by conducting experiments with laboratory robots. The devices are started one after another, passing results from the previous to the next. The platform then translates the outputs into a set of logical rules that researchers can check and use.

During testing, BacterAI was tasked with deciphering the metabolism of two common oral bacteria. The task included identifying a specific set of 20 amino acids (AAs) required to sustain these microorganisms.

As a result, the AI performed up to 10,000 experiments in 24 hours. After nine days, the system learned to predict amino acid sets for each bacterial species with 90% accuracy.

Algorithm trained to see objects outside the field of view

Researchers from MIT and Rice University developed a machine-learning technique called ORCa capable of seeing objects outside the field of view.

The system analyzes distorted reflections on glossy surfaces to identify objects and their distances. For example, with this technology robot cabs can detect approaching traffic at an intersection from reflections on other vehicles.

The AI processes several images of a glossy surface from different angles. Then, using machine learning, it breaks down the reflection of the object into individual pixels. By analysing their changes relative to each other across different pictures, ORCa can determine the object’s shape.

Also, by modeling the scene as the so-called 5D radiation field, the technology assesses the direction and intensity of light rays falling on or reflecting from each point. This allows ORCa to determine how far they are from the reflecting surface and from each other.

In the UK, AI deployed to detect drivers who litter the roadsides

Road AI cameras are being deployed in the UK to identify drivers who litter from car windows.

When violations are detected, the devices will automatically send data to police dispatchers. Officers will be able to review the footage and issue a fine of up to $126 by license plate.

In the coming weeks, the technology will be piloted in southeast England.

Deepfake songs netted the scammer $10,000

A scammer earned almost $10,000 selling AI-generated tracks with a deepfake voice of Frank Ocean.

Discord user mourningassasin hired a musician to create ‘around nine’ fake songs. He trained the AI on \”very high-quality vocal fragments\” from Ocean’s Thinkin Bout You.

Then mourningassasin posted a clip of one of the fake tracks on a music forum with \”leaked\” works. He quickly realised that everyone believed the track was original.

According to the scammer, several people reached out and offered to pay big money for the tracks. Reports indicate that in mid-to-late April 2023 mourningassasin earned $3,000-4,000 per song.

He was later banned from the forum. The owner of the Discord server on which the scammer distributed the Frank Ocean deepfake tracks said the incident seriously damaged trust in the community. It will also raise suspicions of every new and unverified music seller, he added.

Most important AI deals of the past week

From May 7 to May 13, AI startups attracted over $339 million. Here are the most interesting deals.

• OpenAI, Anthropic and Stability AI raised about $120 million from Sound Ventures’ new AI fund.
• Everseen raised $71.3 million to develop computer vision technology to detect potential theft in stores.
• The creators of agricultural robots Ecorobotix raised $52 million.
• Wingcopter secured $43.6 million from the European Investment Bank to expand services and scale drone production.
• The world’s first AI-powered, product-focused venture firm Vela raised $25 million.
• Prewave raised $20 million to grow the development team for a platform to manage supply chain risks.
• AMP Robotics received about $8 million from the Microsoft Climate Innovation Fund.

Also on ForkLog:

• The White House will implement “responsible” AI in government agencies.
• OpenAI stopped training GPT on API user data.
• Warren Buffett compared AI to an atomic bomb.
• Media: Spotify removed thousands of songs generated by AI.
• Google showed a number of AI innovations at I/O 2023.
• Sam Altman will testify before the US Congress.
• Miners offered AI developers computing power.
• The European Parliament approved amendments to the AI Act.
• The US Secretary of Transportation criticised Tesla’s Autopilot.