August has brought super hot temperatures regarding the weather and the AI world.
AI managed to sit at a crossroads of topics like Minority Report, the Olympic Games, and playing games with no engine at all.
08.02.2024 Google suffers backlash after Olympic Gemini Ad
After becoming the official AI partner of the US Olympic team, Google launched an ad where a father describes how his daughter admires and finds inspiration in US hurdler Sydney McLaughlin-Levrone. The ad shows how Google Gemini is guiding the young girl through hurdling, and later, it is asked to write a letter to the star. In the final seconds, Gemini writes the letter to the star on behalf of the daughter. The ad has faced backlash, with users claiming it misses the whole point of writing a letter to an admired athlete.
More can be found in The Verge.
08.02.2024 Google’s Gemini 1.5 Pro outperforms Gpt-4o in most benchmarks
Google’s latest model, a mid-size multimodal system, appears to have drastically improved performance compared to competitive systems like GPT or Claude. The new system extends the context window and delivers better results for the prompt.
More can be found in Google Blog.
08.01.2024 Argentina builds Minority Report
President Javier Milei has built a security unit that plans to use artificial intelligence to “predict future crimes.” According to official statements, the Artificial Intelligence Applied to Security Unit will use machine learning to analyze historical crimes and make predictions. The system is also expected to use facial recognition tools to identify wanted individuals in crowds and analyze video in real-time to spot suspicious behavior.
More can be found in The Guardian.
08.08.2024 GPT-4o considered “medium” risk
According to the internal system used by OpenAI to assess risks related to AI systems, GPT-4o is considered to be a “medium” risk. The categories on which the threat evaluation is based include cybersecurity, biological threats, persuasion, and model autonomy. All were considered “low,” excluding persuasion, where the system was deemed sometimes even more convincing than a human being.
More can be found in The Verge.
08.15.2024 73% of organizations embrace GenAI
According to a PwC survey, 73% of respondents plan to use generative AI in their organizations. Additionally, 58% of respondents stated they have begun assessing AI-related risks.
More can be found in VentureBeat.
08.18.2024 AI may help identify autistic toddlers
The machine learning-based tool designed for screening children under the age of two shows an accuracy of 80%. The system was developed at Karolinska Institutet in Sweden. The developers emphasize that the tool can be used as early information support, but a full diagnosis must be made by trained professionals.
More can be found in The Guardian.
08.20.2024 Anthropic sued by writers for copyright infringement
The company behind the Claude model family was sued by writers and journalists Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson for unauthorized use of their works. The writers follow similar arguments as those who have sued OpenAI, claiming the works were used without permission and that the company is undermining the foundations of creative work.
More can be found in The Guardian.
08.22.2024 OpenAI says that California AI bill will slow progress
OpenAI’s Chief Strategy Officer, Jason Kwon, insisted that AI-related regulations should be left to the federal government. In a letter addressed to state senator Scott Weiner, Kwon argued that the bill may not only slow innovation but also encourage high-tech companies to leave the state.
The senator stated that the bill is only committing large AI labs to do what they are already doing, which is testing large models, including Large Language Models, against catastrophic risks.
More can be found in TheVerge.
08.28.2024 Baidu restricts Google and Bing from scraping content for AI training
The Chinese internet giant, which operates both a search engine and multiple content portals, has restricted Bing and Google systems from scraping its content. The ban is evident on the Baike service, which is comparable to Wikipedia in the West. Previously, Western search engines were allowed to access the content, though some subdomains were restricted.
More can be found in artificialintelligence-news.
08.28.2024 Google releases Doom with only AI beneath
Google researchers have released a neural network that simulates the Doom gaming experience in real-time. The game runs at 20 FPS (not super comfortable, but playable), where every next frame is predicted and then generated by a Stable Diffusion model. The entire gaming experience, including models and the environment, was recreated without access to the original game assets.
The details of the network and all the works behind it can be found in the research paper released along.
08.29.2024 Now 200 million people use ChatGPT every week
OpenAI announced that over 200 million people now use ChatGPT every week. This means the user base has doubled since November last year. Additionally, 92% of Fortune 500 companies are using OpenAI’s products, and API usage has also doubled since the release of the cheaper GPT-4o Mini model.
More can be found in The Verge.
08.05.2024 OpenAI will not watermark ChatGPT-produced text
Watermarking involves infusing text with a pattern undetectable to humans but detectable by other systems. Currently, the company refuses to watermark texts, with the easy exposure of users being one of the main reasons behind this decision. On the other hand, watermarking would make detecting AI-generated text in academic and educational environments much easier.
More can be found in The Verge.
