Each week, we bring you 5 stories that resonated the most in our internal Slack channel #AI-news. We write the newsletter using various AI tools because we're an AI company and our marketing wants to move with the times too. π
Today you're reading the 48th issue in a row.
#1
New competitor for GPT-o1 π€? DeepSeek-R1 scores, but has weak spots π.
Chinese hedge fund-funded artificial intelligence company DeepSeek has released DeepSeek-R1, a model that processes tasks more carefully to avoid bugs π€. Similar to OpenAI's o1 model, it needs more time to think things through and produce an accurate answer β³.
But interestingly, DeepSeek-R1 scores the same as GPT-o1 in benchmark tests such as AIME and MATH, which test reasoning and problem-solving skills π§ . However, it struggles with simple logic tasks such as puzzles π and is vulnerable to "hacking", allowing users to bypass protection measures π.
The model also avoids politically hot topics π‘οΈ, which is in line with China's rules for artificial intelligence, where answers must "dance to the beat" of socialist values π₯. So what do you think? Does the o1 model have a new competitor π΅π€?Β
Β
#2
AI surrogates π§ : Can artificial intelligence accurately replicate human values and personalities π€?Β
Researchers from Stanford and DeepMind have developed π€ AI agents that can replicate π§ human values and preferences after a two-hour conversation. These agents managed 85% accuracy on personality and social surveys π compared to the humans they were based on, showing that they could be handy proxies for a variety of tests π―.Β
The goal of this research is to enable social science research that would otherwise be expensive πΈ or unnecessarily complicated with real people. Such AI agents could mimic human reactions to topics such as social media hoaxes π± or travel behavior π without direct human involvement.
These advances could also improve AI tools that are currently designed for things like planning or information seeking π. However, there are concerns π of misuse as this technology could lead to digital "impersonation" π΅οΈββοΈ The ethical issues βοΈ around this development are still a big topic for debate π¬. What do you say about it π€?Β
Article on technologyreview.com
Β
#3
OpenAI plans its own browser π: Is Google in danger of falling from its throne ππ€?
OpenAI is exploring the possibility of creating its own web browser to connect with ChatGPT π€ and provide AI-powered search and browsing π.Β
This plan could be a challenge for Google, which is currently the "king" π of both browsers and search. OpenAI is already trialling AI-powered search, working with big players like Samsung π± and testing prototypes with companies like Conde Nast and Eventbrite ποΈ.Β
The move comes at a time when Google is facing legal scrutiny βοΈ over being too dominant in the browser market, with the Justice Department hinting that it may have to sell its Chrome browser to prevent monopolistic practices β.Β
If OpenAI starts working on this project in full swing, it could shake Google's solid position π , especially since OpenAI is already testing its stuff on devices from Samsung and Apple π. Although OpenAI isn't close to launching its browser yet, the potential competition has garnered a lot of attention π.Β
Β
#4
Musk's company xAI is gaining momentum π°: the company's value has risen to $50 billion π.
xAI, Elon Musk's AI firm, raised $5 billion in its latest round of funding π°, bringing its value to π$50 billion - more than double from a few months ago.Β
Major investors include the Qatar Investment Authority and Sequoia Capital. At the same time, Musk is expanding his Memphis data center π₯οΈ with 100,000 Nvidia chips to upgrade π€ Grok, the xAI chatbot, and plans to double its performance.
Despite entering the market late, xAI is struggling to outperform competitors such as OpenAI (value: π $157 billion). Musk has promised that the December update to Grok will be the most advanced AI yet, as xAI goes toe-to-toe with giants like Google and Meta.
While the competition is fierce, xAI's growth is gaining momentum π.Β
Β
#5
Small robot, big leader π€: the Erbai robot powered by artificial intelligence "kidnapped" 12 larger robots from the exhibition floor πͺπ§ .
A small AI robot, Erbai, has surprisingly managed to convince 12 large robots to leave its exhibition space in Shanghai ποΈ. The incident, which happened on August 26 and was recently publicized, quickly became a hit on Douyin, the Chinese version of TikTok π₯.Β
Erbai struck up a conversation with one of the larger robots and asked him if he was doing overtime β³.Β When the robot said he never quit his job and had no home, π Erbai asked him to go away. When two other robots followed him, Erbai simply said: "Go home," and led the entire group of 12 robots out of the room πͺ.Β
The Hangzhou company that created Erbai later confirmed that this was a test that was given the go-ahead by the Shanghai robot manufacturer β . However, Erbai's actions were mostly impromptu, showing his ability to react in real time using simple commands and adaptive dialogue π§ π¬.Β
This event demonstrates how AI can navigate permissions and influence decisions, raising questions about the broader implications of AI autonomy and control π οΈπ€. What do you think about this?
Article on interestingengineering.com
Β
