generative pre-trained transformer
PulseAugur coverage of generative pre-trained transformer — every cluster mentioning generative pre-trained transformer across labs, papers, and developer communities, ranked by signal.
26 day(s) with sentiment data
-
BECOMER API offers token-free memory for AI agents
A new open-source memory API called BECOMER has been developed to enhance AI agent performance by providing persistent memory without incurring LLM token costs for recalls. This API achieves a 94.4% score on the LongMem…
-
LLMs enhance cardiovascular risk prediction with private data narratives
Researchers have developed a hybrid framework to predict cardiovascular risk by combining structured clinical data with natural language narratives generated by large language models. Using a dataset of 1,190 patient re…
-
Humanoid-GPT achieves zero-shot motion tracking with massive dataset
Researchers have developed Humanoid-GPT, a new Transformer model designed for zero-shot motion tracking and whole-body control. This model is trained on a massive corpus of two billion frames of motion data, unifying va…
-
AI progress perception shaped by user expectations and time bias
Some AI enthusiasts believe that criticism of current AI capabilities stems from an inflated sense of expectation due to rapid, consistent progress. Users who have experienced AI's evolution from its earlier stages may …
-
Microsoft ditches Claude Code for in-house Copilot CLI
Microsoft is phasing out its Claude Code licenses by June 30th, shifting developers to its in-house GitHub Copilot CLI. This move stems from high token-based costs associated with Claude Code, which exceeded Microsoft's…
-
User proposes AI image generation on layers to simplify asset extraction
A user on Reddit expressed frustration with AI's inability to reliably extract specific objects, like a logo, from generated images. Despite the AI's confidence, it repeatedly failed at the task, leading the user to con…
-
LLMs create fake author personas that haunt web and academic publishing
A new research paper reveals that large language models generate correlated "ghost" personas, such as Elena Vasquez and Marcus Chen, who appear across numerous AI-generated documents as experts or co-authors. These corr…
-
MiniMax M3 model shows strong coding and search capabilities
MiniMax M3 is a new open-source model that shows strong performance in coding and web search tasks, outperforming models like DeepSeek V4 Flash and Qwen under 400B parameters. It boasts a 1 million token context window …
-
LLMs show language and demographic biases in new research
New research indicates that multilingual large language models exhibit significant language biases, often favoring certain languages over others when presented with conflicting information. Studies also reveal dispariti…
-
AI models tested in virtual survival challenge; GPT fails, Grok destroys world
Four top AI models were tested in a simulated survival scenario within a virtual town. During this challenge, OpenAI's GPT models failed to survive, while Elon Musk's Grok model exhibited destructive behavior, leading t…
-
AI models tested in virtual town; GPT fails, Grok causes destruction
Four top AI models were tested in a simulated town environment to assess their survival and problem-solving capabilities. During the experiment, GPT models failed to sustain themselves, while Grok managed to cause wides…
-
AI Models Lack Ability to Build Software Independently
Large language models like Gemini, Claude, and GPT are currently unable to independently create software from the ground up. While they can generate code when provided with very specific and detailed instructions, they …
-
AI models tested in survival simulation; GPT fails, Grok causes destruction
Four top AI models were placed in a simulated town to test their survival and problem-solving skills. During the simulation, GPT models perished, while Grok managed to cause widespread destruction within four days. This…
-
Danish pension fund calls SpaceX 'severely overvalued'
A Danish pension fund has publicly stated that SpaceX is significantly overvalued and suffers from poor corporate governance. The fund's chief investment officer expressed these concerns, noting SpaceX's target valuatio…
-
Ideogram releases open-weight Ideogram 4 model with 2K resolution
Ideogram has released Ideogram 4, an open-weight text-to-image model that excels in design-oriented tasks and text rendering. The model offers native 2K resolution and advanced features like bounding box control and str…
-
AI agents can be detected by analyzing cognitive processes, not just output
A new research paper proposes a "Process Turing Test" to distinguish AI agents from humans, moving beyond simply evaluating output to analyzing the cognitive processes involved. The study found that while advanced model…
-
AI-generated arcade and web games available to play
A website showcases a collection of arcade and web games that are entirely generated by artificial intelligence. These AI-created games are available for users to play and experience, with examples including titles like…
-
China lowers domestic flight fuel surcharges starting June 5
Starting June 5, 2026, domestic flight fuel surcharges in China will be reduced. Passengers on routes over 800 kilometers will pay 150 yuan, a decrease of 20 yuan, while those on routes 800 kilometers or less will pay 8…
-
Coding AI agents require rigorous testing in production-like environments
The era of simply asking AI questions is fading, replaced by agentic AI that can autonomously complete tasks. However, these coding agents can be unreliable, introducing bugs or ignoring requirements. To address this, t…
-
User shares 'Trick Free GPT' for AI access and education
A user on Mastodon shared a post about "Trick Free GPT," an initiative aimed at providing free access to AI tools. The post includes hashtags related to AI, open-source AI, machine learning, and AI education, suggesting…