METiS-Lilly LNP Deal Highlights AI Promise, Not Proven Results
METiS's collaboration with Eli Lilly centers on applying its AiLNP platform and PhatGPT model to design tissue-specific LNPs, with METiS receiving low...
Created by tong wu
Latest AI research and applied breakthroughs from major labs for engineers and product teams
Explore the latest content tracked by AI Lab Breakthroughs
METiS's collaboration with Eli Lilly centers on applying its AiLNP platform and PhatGPT model to design tissue-specific LNPs, with METiS receiving low...
Sage-A1.5 shows that capability composition—orchestration for subagent delegation, verification for strategy revision, and open exploration...
DeepMind's 2004 vision required unavailable pieces: deep learning, sufficient compute, and integrated systems for general intelligence. By 2026 the...
Monthly climate averages mask short heatwaves that strip snow and expose darker ice for faster melting.
SWIFT improves stronger diffusion models during training by extrapolating regression targets away from a frozen weak denoiser (an early checkpoint),...
TRACE organizes multimodal events into day-level trajectories, separating global context from local signals across EHR, imaging, and notes, then...
OpenAI's expansion to Google Cloud places its models on all three major cloud providers, a strategic move to grow enterprise distribution beyond its...
BrainLoRA applies LoRA adaptation to Alibaba's Qwen3-Omni for predicting fMRI responses to video, slashing feature-extraction compute by ~350x while securing top performance on the Algonauts 2025 challenge.
OpenAI released 722 manuscripts across 372 result families on GitHub, complete with Lean formalizations and reasoning summaries, but kept the...
Google is rolling out Personal Intelligence to Gemini in Android Auto and cars with Google built-in, letting the AI pull data from Gmail, Calendar,...
Physicists are moving past generic prompts to build targeted harnesses that match Claude's strengths in coding and breadth.
Prof. Matthew Schwartz...
MedGemma provides an open vision-language foundation that supports diverse medical tasks through data-efficient fine-tuning, delivering 2.6–18.1%...
OpenAI added targeted evaluations for COBOL migration and software engineering tasks, signaling focus on real-world legacy code handling over broad...
CDS researchers are contesting Anthropic's 2025 report that Claude models could detect injected concepts in neural activity. This highlights ongoing scrutiny of how such AI-neural claims are validated.
Four Gemini 4 Argon agents autonomously completed Minecraft by writing 2,783 self-authored programs over 27 hours, revealing how long-horizon...
Anthropic’s leaked draft set a high bar for federal intervention: Commerce could seek a court order only for foreseeable risks causing 50+ deaths,...
HPE's Spaceborne Computer-4 will attempt to run language models with retrieval-augmented generation and LiDAR mapping on the Moon using a fanless,...