Deepfake Propaganda, China’s ChatGPT Rival and Other AI News

We aim to inform readers not only about events in the bitcoin industry, but also about what is happening in related technological realms — cybersecurity and the world of artificial intelligence (AI).

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

Baidu to develop a ChatGPT analogue

In March 2023, Baidu will launch its own \”Chinese answer to ChatGPT\” — the Ernie Bot (Wenxin Yiyan). Amid this news, the tech giant’s shares rose 15%.

According to Bloomberg, Baidu plans to complete internal testing before making it public. The chatbot will be built on the ERNIE model with 260 billion parameters, capable of processing text in English and Chinese. The neural network outperforms competitors like GPT-3 in many tasks, the company noted.

Initially, the search giant plans to integrate the technology into its products. It will allow users to receive results in a \”conversational\” style.

Apple to hold a closed AI summit

Apple will host an AI Summit for its employees. It will be held in person at the Steve Jobs Theater.

Interestingly, the event will be held in the Steve Jobs Theater at the Apple HQ and the in-person event will be streamed to employees as well. So, essentially how Apple held media events pre-Covid. Have to wonder if Apple will also return to that for public product launches. https://t.co/uhLv4se9oW

— Mark Gurman (@markgurman) February 6, 2023

According to Bloomberg reporter Mark Gurman, Apple will host the AI summit amid the rising popularity of systems like ChatGPT. It is expected that the company will showcase some outcomes of the gathering during WWDC 2023, which is slated for June.

Andrej Karpathy joined the OpenAI team

The former AI director at Tesla and head of the Autopilot program, Andrej Karpathy, has joined OpenAI.

Some personal news: I am joining OpenAI (again :)). Like many others both in/out of AI, I am very inspired by the impact of their work and I have personally benefited greatly from it. The future potential is especially exciting; it is a great pleasure to jump back in and build!🪄

— Andrej Karpathy (@karpathy) February 9, 2023

He said he was \”very inspired by the impact\” of the startup’s developments and that he has personally benefited greatly from them.

‘It\’s a great pleasure to come back and build,’ Karpathy added.

NTSB: Tesla Autopilot not used in the high-profile Texas crash

The U.S. National Transportation Safety Board (NTSB) ruled that the 2021 crash involving a Tesla was not caused by Autopilot but by speeding. Two people died in the accident.

At the time, local police suggested that no one was in control of the car at impact. One passenger was in the front seat beside the driver, and the second in the back.

However, investigators found that one of the crash participants was at the wheel and was driving the Tesla, but lost control and crashed into a tree, causing the electric car to catch fire. After the crash, the driver moved to the back seat.

According to the NTSB, the car accelerated to 108 km/h in two seconds and hit a tree at 92 km/h. Toxicology tests showed the driver’s blood alcohol content was nearly twice the legal limit in Texas.

Microsoft to embed AI in Excel for automatic formula filling

Microsoft plans to add AI to Excel spreadsheets.

Researchers developed the 60-million-parameter language model FLAME. It is based on Google’s T5 and is designed for auto-filling Excel formulas, fixing them and reconstructing syntax.

According to the developers, despite its \”modest size,\” FLAME outperforms larger models for code completion such as CodeT5 (220 million parameters), Codex-Cushman (12 billion parameters) and Codex-Davinci (175 billion parameters).

Getty Images filed a second lawsuit against Stability AI

Getty Images filed a lawsuit in the Delaware federal court against Stability AI over alleged use of images to train Stable Diffusion.

The company accused the AI developer of \”flagrant infringement of intellectual property rights on a staggering scale.\” The suit alleges Stability AI copied more than 12 million images from a stock photo database without permission or compensation in its bid to build a competing business.

According to Getty Images, the creator of Stable Diffusion violated copyrights and infringed the service’s trademark.

In January 2023 the company filed a similar suit against Stability AI in the United Kingdom.

Quora released AI chatbots to the public

The social platform Quora opened Poe, a platform with smart chatbots for the general public.

The service provides access to several AI models capable of answering questions and engaging in a two-way dialogue. In the Poe chat interface, a range of topics and use cases are offered, such as ‘writing help’, ‘cooking’, ‘troubleshooting’ and ‘nature’.

The company said it launched the service as a standalone, independent product. However, if Poe’s content meets high-quality standards, it will be incorporated into Quora.

The service is currently available only to iOS users, but in a few months it will roll out to other platforms. Quora also plans to open Poe to developers.

Runway unveiled an AI video editor from a textual description

Runway unveiled the AI algorithm Gen-1, capable of changing a video’s style based on a text prompt or image.

The system runs in the cloud via the company’s site. It is currently available to select users, but in a few weeks the technology will be opened to all on the waiting list.

Runway will publish technical information about Gen-1 in the coming days.

Military robodogs will hit the streets of Los Angeles

The Los Angeles Police Department plans to acquire military robodogs.

Members of the Public Safety Committee on the city council voted to approve the purchase by law enforcement agencies. This drew a hail of public criticism.

The overwhelming majority who contacted the hotline warned the agency of a serious threat to ethnic minorities and noted the possible outfitting of the devices with deadly ammunition.

They also cited the decision by the San Francisco Board of Supervisors to curb the capabilities of robots and the New York Police Department’s decision not to deploy such systems.

Research: China used deepfakes for propaganda

The Chinese Communist Party used deepfake TV presenters to promote its interests.

According to a report by the American firm Graphika, AI-generated reporters for the fake news agency Wolf News circulated misleading political content on social media as part of a state operation.

For example, the deepfake news anchor Alex criticized U.S. inaction on gun violence that has overwhelmed the country. Another fake reporter highlighted the importance of \”great power cooperation\” — China and the U.S.

Experts say both Wolf News anchors were almost certainly created using AI technologies from London’s Synthesia.

Twitch blocked the ‘infinite’ AI series

The streaming platform Twitch briefly blocked the AI-generated infinite series Nothing, Forever due to transphobic remarks.

The show, inspired by Seinfeld, depicts the daily lives of pixelated analogs of TV characters. In its development the creators used the GPT-3 Davinci language model, Stable Diffusion and DALL-E image generators, as well as Microsoft’s Azure Cognitive Services.

The storytelling continues and evolves based on Twitch viewers’ feedback in real time.

According to Motherboard, during one stand-up the AI-scripted Nothing, Forever character suggested that transgender people are mentally ill. Soon after, Twitch blocked the show for 14 days.

According to the developers, the problem stemmed from replacing Davinci with Curie, which generated the inappropriate text. OpenAI’s content moderation system did not trigger, they explained.

The authors added that they no longer plan to use Curie as a backup option. They also pledged to do everything possible to prevent similar incidents.

The week’s most important AI deals

From February 5 to 11, 2023, AI startups raised more than $190m. Here are the most notable deals.

• Zeitview raised $55m to develop data collection technology from air and ground.
• Ushur received $50m to develop tools to simplify deployment of conversational AI and intelligent document processing systems.
• Acceldata raised $50m.
• Magic raised $23m to develop a system for generating code.
• Macro raised $9.3m to develop an AI tool for processing texts.
• Moonhub raised $4.4m to support a technology for finding workers for fast-growing companies.

Also on ForkLog:

• ChatGPT tricked OpenAI’s detector of generated texts.
• The GitHub chief urged exclusion of open-source software from the EU AI Act.
• In response to ChatGPT: Google introduced the Bard chatbot.
• Experts: AI will destroy jobs and widen inequality.
• Microsoft unveiled an updated Bing based on ChatGPT.
• DALL-E 2 integrated into a program for generating photorealistic composites.
• Alphabet lost $100b in market capitalization due to Bard advertising error.
• The AI-chatbot race between Microsoft and Google drove up AI tokens.
• The Pentagon wrote a press release with ChatGPT.