AI research

DeepMind’s AI-powered soccer bots run circles around their manually programmed rivals

Summary Researchers at Google DeepMind have used reinforcement learning to teach small humanoid robots soccer skills like turning, kicking, and chasing the ball. The soccer robots were first trained in a simulation using the MuJoCo physics engine, and then transferred their skills to small Robotis OP3 humanoid robots with 20 articulated joints. Training took place …

DeepMind’s AI-powered soccer bots run circles around their manually programmed rivals Read More »

Tailored AI counter-evidence can reduce belief in conspiracy theories, study finds

Summary A new study by researchers at MIT and Cornell University suggests that conversations with language models like GPT-4 can significantly weaken belief in conspiracy theories. In the study, 2,190 conspiracy believers first described in detail a conspiracy theory they believed in and the evidence they thought supported it. They then engaged in a three-stage …

Tailored AI counter-evidence can reduce belief in conspiracy theories, study finds Read More »

OpenAI’s Sora is the “GPT-1 of video” with plans to scale and unlock emergent AI capabilities

Summary OpenAI’s Sora AI model is capable of generating minute-long videos of impressive quality. In a presentation, the developers compare it to GPT-1, the precursor to modern language models. OpenAI sees Sora as the foundation for better understanding and simulating the real world – a crucial step on the path to AGI. In a presentation …

OpenAI’s Sora is the “GPT-1 of video” with plans to scale and unlock emergent AI capabilities Read More »

Google’s Mixture-of-Depths uses computing power more efficiently by prioritizing key tokens

Summary Google Deepmind researchers have introduced “Mixture-of-Depths”, a method to use the computing power of transformer models more efficiently. Traditional transformer models use the same amount of computing power for each token in a sequence. In contrast, Google Deepmind’s “Mixture-of-Depths” (MoD) allows the model to flexibly and selectively distribute this computing power to the tokens …

Google’s Mixture-of-Depths uses computing power more efficiently by prioritizing key tokens Read More »

Archetype AI’s “Newton” model learns from sensor data to answer complex real-world questions

Summary Startup Archetype AI has introduced its first “physical AI” model called Newton, designed to interpret sensor data and combine it with natural language to answer questions about the real world. Unlike large language models or other multimodal models trained only on text and images, Archetype AI develops AI models that better understand the physical …

Archetype AI’s “Newton” model learns from sensor data to answer complex real-world questions Read More »

StableDrag’s simple point-and-click image editing makes turning Mona Lisa’s head easy

Summary Many AI image generators already provide a powerful tool for modifying image content with text, called inpainting. Point-based editing makes adjustments even easier. Researchers from Nanjing University and Tencent have developed a new AI-based image editing method called StableDrag that allows elements to be easily moved to new positions while maintaining the correct perspective, …

StableDrag’s simple point-and-click image editing makes turning Mona Lisa’s head easy Read More »

Anthropic’s Claude 3 beats OpenAI’s GPT-4 at text summarization

Summary The context window of large language models, represented in tokens, indicates how much information an AI model can process simultaneously. Today, this window is large enough for models to summarize entire books. A new study assesses the quality of these summaries across multiple dimensions. The context windows of large language models have been growing …

Anthropic’s Claude 3 beats OpenAI’s GPT-4 at text summarization Read More »

Apple sets its sights on personal robots as next frontier after Vision Pro debut

Summary According to a report by Bloomberg, Apple is still trying its hand at robots after the failure of its car project. Apple is looking for new growth markets and has set its sights on developing personal robots, according to informed sources. The people, who spoke on condition of anonymity, told Bloomberg that the project …

Apple sets its sights on personal robots as next frontier after Vision Pro debut Read More »

Open-source AI agent SWE-agent nips at the heels of Cognition AI’s $21 million Devin

Summary AI researchers unveil an AI agent that can program better than GPT-4, putting the open-source project in competition with offerings like the recently launched Devin AI programmer. The Princeton NLP team has developed SWE-agent, an open-source system that converts language models like GPT-4 into software engineering agents. These agents can fix bugs and problems …

Open-source AI agent SWE-agent nips at the heels of Cognition AI’s $21 million Devin Read More »

Researchers find LLMs struggle with exploration, a key capability for useful AI agents

Summary Researchers investigate whether large language models can effectively exhibit exploratory behavior, considered a key element for useful AI agents. A team of researchers from Microsoft Research and Carnegie Mellon University investigated the ability of large language models to perform exploration, a key aspect of reinforcement learning and decision-making. They found that common models such …

Researchers find LLMs struggle with exploration, a key capability for useful AI agents Read More »

Scroll to Top