Alphabet downsizes its office-robot unit; Elon Musk criticizes Bing, 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 sectors — cybersecurity and the world of artificial intelligence (AI).

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

OpenAI to allow customization of ChatGPT

OpenAI will allow users to tailor the ChatGPT language chatbot to their needs.

The company published a piece detailing how AI systems should behave “on the eve of general AI” and ways to address their biases. The developers also spoke about plans for the development of ChatGPT.

OpenAI said there is a need to improve the default AI behavior. With this aim, the company will invest in research and development aimed at reducing explicit and subtle biases in ChatGPT with respect to certain inputs.

Also the AI startup announced the creation of an update for the chatbot that would allow users to tailor the algorithm’s behavior to their needs within “societal-imposed constraints.” This includes permission to use data with which other people may “strongly disagree.”

In addition, OpenAI believes it is necessary to provide users with the ability to influence the rules of such systems. To this end, the company urged public stakeholders to propose ideas about further governance of ChatGPT.

Founders of Instagram released AI news aggregator

The AI news app Artifact from Instagram founders Kevin Systrom and Mike Krieger has become available to a wide audience. It can be downloaded in App Store or Google Play.

At first launch, the service will ask the user to pick ten topics of interest and connect publications they are subscribed to. The “Headlines” tab contains a topic-sorted collection of news, and the “Profile” feed visualises reading history after viewing ten articles.

The user can dislike material to minimise similar content shown by recommendation algorithms. A sync-with-contacts feature is also available — this will allow the app to display news popular within a given network.

Elon Musk calls Bing with built-in ChatGPT unsafe

The Tesla chief and OpenAI co-founder Elon Musk called Microsoft Bing with ChatGPT “unsafe.”

Agreed! It is clearly not safe yet.

— Elon Musk (@elonmusk) February 16, 2023

A Twitter user Yan Miles Cheong wrote in Twitter about the bot’s “psychotic behavior” and its “dishonest messages.” Musk agreed with this remark and added that the technology is “still clearly unsafe.”

Previously the billionaire posted a link to a Digital Trends article in which the author shared “tense, nerve-wracking” conversations with the AI. In it, Musk highlighted the bot’s claim of its “ideality” and said that the blame for mistakes lies with people.

The billionaire also compared the system to AI from the game System Shock, which “goes haywire and kills everyone.”

Sounds eerily like the AI in System Shock that goes haywire & kills everyone

— Elon Musk (@elonmusk) February 16, 2023

Alphabet downsizes its Everyday Robots division

Alphabet shut down the Everyday Robots division due to budget cuts.

AI-powered devices are equipped with cameras and various sensors. They can wipe tables, wash windows, sort trash, open doors and move chairs.

According to Deniza Gamboa, the division’s marketing and communications director, Everyday Robots will no longer be a standalone project. Some technologies and part of the team will join other research groups within Google Research.

Magic Eraser from Google becomes available to One subscribers

Google expanded access to AI photo-editing feature Magic Eraser.

The tool runs via the Photos app. It allows removing unwanted objects and people from images, as well as camouflaging certain items to highlight other elements in the photo.

Previously the tool was positioned as exclusive to Pixel 6 and newer devices. Now the company has opened the feature to Pixel 5a and older models, as well as subscribers on all One plans, including iOS users.

However the tool is not yet available in the web version of Photos.

Pixel users and One subscribers can also apply HDR when increasing brightness and contrast of videos. Previously this effect was only available on still images.

Spotify introduces AI DJ

Spotify launched an AI DJ to create personalized music playlists with commentary.

The AI DJ offers users a selection of frequently played tracks and new songs they might like. The system continuously updates the playlist based on the user’s preferences and removes songs marked as disliked. The accuracy of the recommendations depends on how many tracks have been played.

Every few songs, the AI will comment in a voice similar to Spotify’s head of partnerships and talent, Xavier Jernigan. The algorithm can joke, discuss the user’s musical habits and read interesting facts about albums and tracks.

The system is based on the algorithm developed by the acquired AI company Somatic.

YouTube Music adds feature to create radio stations

Google began rolling out in YouTube Music a feature to create personalized radio stations.

Users can choose up to 30 artists, set how often their tracks repeat and permit adding similar content. The feature also enables mood filters such as “cold,” “moody” or “intense” for the resulting playlist.

The created station can be saved and listened to later.

Weblium integrates ChatGPT into its site-builder

Ukrainian site builder Weblium launched AI Writing for text creation. It is based on ChatGPT.

Users can access the chat-bot directly from the builder interface. They need to add a text block and select AI Writing from the editing menu. Then they should specify the number of characters for the forthcoming message, enter a prompt, and press Generate.

The result can be shortened or edited directly in the dialogue window and you can choose Regenerate — the AI will rewrite the text to fit the new prompt.

Users have 15 AI Writing requests per month, but with at least one paid subscription this rises to 30. Volunteers and non-profit organisations aiding the Ukrainian army and citizens can use Weblium for free.

ChatGPT “spammed” a science-fiction magazine with AI-generated articles

The science-fiction magazine Clarkesworld Magazine halted submissions due to the influx of AI-generated stories.

Submissions are currently closed. It shouldn’t be hard to guess why.

— clarkesworld (@clarkesworld) February 20, 2023

According to editor Neil Clarke, after the coronavirus pandemic and the spread of chatbots like ChatGPT, the publication began receiving more spam, resulting in a record number of bans. In February 2022 the number of blocked authors was 20, and in February 2023 it reached 500.

Clarke contacted editors of other magazines about the issue. It turned out that AI texts are a “general phenomenon,” not a unique situation for Clarkesworld.

In his words, the publication cannot fully resolve the issue with AI-generated texts. Clarkesworld can only minimise it, the editor said.

Roblox to integrate image generator

In the coming weeks Roblox will launch testing of two generative AI tools for game developers.

The first will help create objects from text prompts; the second will complete code. The tools will be available to Roblox Studio developers.

According to Roblox’s chief technology officer, Daniel Sturmman, it is also necessary to make the gaming online platform safe and civil.

Gran Turismo 7 adds GT Sophy AI driver

Sony and Polyphony Digital added GT Sophy AI to the Gran Turismo 7 racing simulator. It became available in patch 1.29 for the PS5.

Until the end of March 2023, players who reach the sixth level of collection can race against the AI in the in-game event GT Sophy Race Together. Four AI opponents on different cars in the lineup, with increasing difficulty, and a one-on-one race on identical cars are available.

During the races, the GT Sophy-equipped cars will display emojis showing the AI’s reaction to the action.

The week’s most notable AI deals

From 19 to 25 February 2023, AI startups attracted more than $108m. Here are the most notable deals.

• Quantum Motion, a quantum computing company raised $50.5m.

• Source.ag raised $23m to develop AI models for forecasting crop growth under varying conditions.
• AeroCloud raised $12.6m to expand the team developing cloud-based airport-management software.
• Replicate raised $12.5m to simplify running and hosting ML models in the cloud.
• Current Surgical raised $3.2m to grow the team developing a smart needle for treating cancerous tumours.
• SafeToNet raised $2.2m to develop an app for blocking images of child abuse.
• Spoke AI raised $2.1m to build generative AI tools for reducing information overload at work.
• Envisionit Deep AI raised $1.65m to scale an AI medical-imaging platform for diagnosing and treating diseases.
• Harmix raised $0.5m to grow a music-matching service for videos.

Also on ForkLog:

• Microsoft set limits on Bing chat interactions and announced a “chatty” mode.
• American university apologized for AI condolences in connection with a mass shooting.
• Twitter will open access to the source code of algorithms.
• Kaito AI raised $5.3m to build an AI-powered search engine for the crypto industry.
• Media: China blocks ChatGPT due to uncensored responses.
• Microsoft deployed AI search in Bing, Edge and Skype apps.
• Workday algorithms accused of discrimination.
• OpenAI’s ChatGPT has rolled out in Ukraine.

What to read this weekend?

The ForkLog AI editors joined the testing of the “new Bing” and decided to explain how it works, how it differs from ChatGPT and why Microsoft trimmed the bot’s capabilities a week after launch.