APREX · EST. 2026
Research
I FOKUS
The Brain in the Machine: Anthropic Finds Consciousness-Like Core in LLMs
New Anthropic research suggests that language models have a kind of internal 'workspace' for thinking — and it could fundamentally change how we understand AI memory.
Flere saker.
JUN 20ResearchStudy: Young workers are getting dumber from overusing AIJUN 16Undergroundgzip beats GPT: Old zip algorithm holds its own as a language modelJUN 14ModelsAnthropic shut down Fable 5 for the entire world – warns of historic precedentJUN 7ResearchAI Designed This Drug in 26 Days. Now It Goes Into Human Trials.JUN 6ResearchBjerkvik Takes the World by Storm: Norway's Arctic AI Giant Lands on the Cover of TimeJUN 1ResearchChina approved the world's first commercial brain implant — paralyzed man wrote againMAY 29ResearchWaterloo students build AI prototypes for learning and workMAY 29ToolsNotebookLM becomes the I/O compass for Google's AI floodMAY 29ResearchAnthropic measures the code-agent gap in social science researchMAY 29ModelsDeepSeek-R1-0528 keeps the reasoning pressure onMAY 29ResearchMagma Wants to Give Multimodal Agents a Sense of SpaceMAY 29ResearchAnthropic trains AI with a hidden goal in order to learn how to expose itMAY 29ResearchMETR gives AI agents a new clock: How long can they work alone?MAY 29ResearchPaperBench tests whether AI can replicate real researchMAY 29Open SourcePhi-4-reasoning shows that small open models can think deeplyMAY 29Open SourceDeepSeek-R1 was the shock that made open reasoning credibleMAY 29Open SourceOlmo 3.1 shows what 'genuinely open AI' actually requiresMAY 24ResearchOpenAI Solved an 80-Year-Old Math Problem — and Mathematicians Confirm ItMAY 22UndergroundStanford DSL beats Triton 14x — and nobody is talking about itMAY 21UndergroundGoogle crushes GPU memory with TurboQuant — 8x performance on 4-bitMAY 20ModelsGemini Omni transforms images, audio, and text into video in one conversationAPR 28ModelsNVIDIA's new AI model beats competitors with 3.3x higher performanceAPR 28ResearchDeepMind Legend Raises $1.1 Billion: Aims to Train AI Without Human DataAPR 27UndergroundGoogle's TurboQuant Crushes the KV-Cache Bottleneck — 6x Memory Reduction Without Quality LossAPR 14UndergroundNow you can query neural network weights like a databaseAPR 12Research9 out of 10 AI agent retry attempts are wasted — here's the solutionAPR 10ResearchAI reads your thoughts with 82% accuracy — Norwegian researcher warns