Google Gemini: not even as good as GPT-3.5 Turbo; OpenAI dealing with dangers of AI

🔍 Google Gemini is not even as good as GPT-3.5 Turbo, researchers find. Link

• Google's Gemini Pro, designed to compete with ChatGPT, performs worse on many tasks compared to OpenAI's older model, GPT-3.5 Turbo, according to new research.

• Despite Google claiming superior performance in its own research, an independent study showcases Gemini Pro falling behind GPT models in areas like reasoning, mathematics, and programming.

• However, Google's Gemini Pro excels in language translation across several languages, despite its generally lower performance in other AI benchmarks.

🚗 TomTom and Microsoft are launching an AI driving assistant. Link

• TomTom and Microsoft are creating an AI voice assistant for cars, allowing drivers to control navigation and car functions using voice commands.

• The assistant leverages Microsoft technologies like Azure AI and can be integrated into TomTom's infotainment system or other car platforms.

• This collaboration enhances TomTom's offerings, including a ChatGPT plugin for developers, and continues its competition with Google Maps.

🧠 AI beats humans for the first time in physical skill game. Link

• ETH Zurich researchers created an AI robot that beat the human world record in the labyrinth game, demonstrating advanced motor skills and learning abilities.

• The robot, CyberRunner, used model-based reinforcement learning to improve and completed the labyrinth in 14.48 seconds after 6.06 hours of practice.

• The project will be made open source, allowing anyone to conduct AI research with a modest budget, and potentially enabling large-scale, global experiments.

🤖 OpenAI lays out plan for dealing with dangers of AI. Link

• OpenAI introduced a "Safety Advisory Group" with veto power to oversee AI risk management and maintain high safety standards.

• The company's updated "Preparedness Framework" includes detailed risk categories and prohibits development of AI models with high or critical risks.

• The final decision-making authority rests with OpenAI's CEO and CTO, despite the advisory group's recommendations and veto power.

📄 Research finds people struggle to identify AI from human art, but prefer human-made works. Link

• Participants struggled to differentiate between art created by AI and humans, with only slightly better than chance accuracy.

• Despite uncertainty in identification, participants consistently preferred human-generated art, expressing positive emotions towards it.

• The study suggested the "uncanny valley" effect might contribute to the preference for human art due to subtle differences in AI-created art.

• The research highlighted the advancing capabilities of generative AI models in creating art indistinguishable from human creations.

• As AI continues to evolve, understanding the psychological impact and preferences associated with AI-generated content becomes crucial.

Thank you for subscribing to DataPragmatist. We have something brewing up and I am excited about it. Stay tuned, more information in coming weeks. for the first time