February may be the shortest month, but it was packed with action. From new models and chatbot apps available for the public, to Google scrapping its “no weapon development” rule – it was a busy month, indeed.
The month has also brought the infusion of Microsoft Paint with Artificial Intelligence and the first battle over AI copyright infringement to be won by a news vendor – Thomson Reuters itself.
1 Feb 2025 Microsoft Paint gets Copilot
Microsoft added a copilot button to Paint, an iconic image manipulation tool added to every Windows iteration. The button allows one to access all AI-powered features in one place instead of searching for them in the toolbar.
More can be found in The Verge.
2 Feb 2025 Google drops promise to not use AI for weapons
Alphabet, the holding behind Google, Deepmind, and several other companies, removed their rule to not develop technologies that may “cause or are likely to cause overall harm” from their guidelines. This includes delivering systems used in weapons or surveillance. According to Google’s AI head, Demis Hassabis, the guidelines were dropped due to the changing world and the increasing need for AI to “protect national security.”
More can be found in The Guardian.
4 Feb 2025 DeepSeek banned from Australian government devices
The decision is the result of advice from intelligence agencies and, according to the official statements, has no connection to DeepSeek’s country of origin. The key aspect was the risk that the system poses to public resources.
More can be found in The Guardian.
6 Feb Gemini 2.0 available for everyone
The 2.0 Flash Thinking Experimental model was made available to every user through the app from a dropdown menu. Users may access the model both on desktop and mobile. Gemini 2.0 was made available for developers during the I/O conference in 2024 and it has been popular among developers since then.
More information can be found on the Google Blog.
6 Feb 2025 Google adds digital watermark to AI-manipulated images
Google Photos is adding digital SynthID to photos changed using their Magic Editor. The new feature intends to make identifying AI-manipulated images more straightforward.
More can be found in The Verge.
6 Feb 2025 Mistral.ai launches LeChat mobile app for iOS and Android
Mistral.ai, a Paris-based AI development lab, has released an iOS and Android app following the release of LeChat, a web app comparable to ChatGPT. The app swiftly stormed the list of most popular apps in both application stores respectively.
More can be found in VentureBeat.
11 Feb 2025 BBC finds that AI assistants tend to confuse facts when asked about current affairs
According to a study conducted by BBC, AI assistants like ChatGPT, Copilot, Gemini, and Perplexity tend to confuse facts when asked about current affairs, and the summaries delivered have shown “significant issues.” Up to 13% of the “facts” cited by AI assistants were either made up or distorted by AI-powered systems.
More can be found in the BBC study.
12 Feb 2025 Thomson Reuters won a court battle against AI on copyright
A global news vendor won its legal battle over copyright infringement against Ross Intelligence, an AI legal startup. The case was about scrapping data without permission – according to the court ruling, this is beyond “fair use.”
More can be found in The Verge.
12 Feb 2025 Adobe’s video generator available for everyone
The Firefly Video Model is now available for all users, allowing anyone to generate clips. The company highlights that the model is compliant and poses no risks. The model is now capable of generating images, videos, sounds and vector graphics.
More can be found on Adobe Blog.
12. Feb 2025 Large Language Models generalize better when unsupervised
A study delivered by Hong Kong University and the University of California, Berkeley, showed that Large Language Models and Vision Language Models tend to generalize better when left unsupervised, as compared to being fed large amounts of labeled data. In fact, providing the model with too much labeled data hampers its ability to interpret new data unseen before.
More can be found in VentureBeat.
13 Feb 2025 iPhone AI features in China will be powered by Alibaba
Apple Intelligence features have not yet been approved in China. To overcome this challenge, the company decided to use models from local providers. After examining the offer from Tencent, ByteDance and others, Apple decided to pick Alibaba. Apple will also continue its cooperation with Baidu regarding web search.
More can be found in The Verge.
17 Feb 2025 New York Times adopts AI tools
The news vendor is encouraging staff to use headline suggesting tools, editing support, and in suggesting questions to be asked during interviews. Yet the company puts a strong emphasis on information verification, which will be done by human journalists.
More can be found in The Verge.
18 Feb 2025 Google Meet can now create action items after the meeting
Google’s AI assistant, Gemini, will now not only take notes and summarize meetings, but also provide users with a list of action items. The feature goes as far as assigning deadlines and picking out the primary stakeholders.
More can be found in The Verge.
19 Feb 2025 Xbox builds generative AI model for gameplay
Muse is a genAI model that helps to build AI-powered gameplay experiences. The model, trained on the Bleeding Edge multiplayer battle arena game, has proven capable of generating the game environment, including physics, the game’s reactions to player actions, and other elements.
More can be found in Venture Beat.
21 Feb 2025 DeepSeek aims to open-source Artificial General Intelligence (AGI)
The Chinese research lab aims to open source and share their progress on building AGI. As a first step, the company released five repositories. According to the company’s statements, secrecy is hampering progress.
More information can be found in AI News.
21 Feb 2025 AI lets us understand animal emotions
Researchers from the University of Copenhagen delivered an AI-based system that decodes the emotions of pigs, cows and wild boars, among others. The system detects positive and negative emotions based on the patterns of their acoustic vocalizations.
More can be found in ScienceDaily.
24 Feb 2025 Anthropic introduces hybrid reasoning in Claude Code
The new Claude 3.7 Sonnet is the first model with “hybrid reasoning” which allows the system to solve more complex problems and to outperform previous iterations in tasks like mathematics and coding. In addition, Anthropic released the Claude Code tool, which is marketed as a coding tool that is complementary to a human coder.
More can be found in The Verge.
27 Feb 2025 OpenAI announced GPT-4.5 release
OpenAI has launched a research preview of GPT-4.5 model, the most powerful one so far. Contrary to the solutions offered by the competition, this model is not Chain-of-Thought, model focused on reasoning, but rather a typical LLM. According to Sam Altman, this is the first model, that delivers a real feeling of talking to a real person.
More can be found in The Verge.
