One thing’s certain about trends: they never stop evolving. Couple that with the rapidly moving AI and machine learning landscape, and you’ve got something fluid and ever-changing! @techtargetnews
’ piece on 8 trends to keep a close eye on in 2025 is a great read this week:
RESEARCH
-
8 AI and Machine Learning Trends to Watch in 2025
By
–
-

Quantum Technologies Transform Cybersecurity Solutions for Data Protection
By
–
Quantum technologies continue to reshape cybersecurity by introducing solutions that tackle the growing complexity of protecting sensitive data against evolving threats, highlighting the urgency of adapting innovative tools for the future. Microblog @antgrasso #QuantumComputing
-
Diffusion Models, AI Optimizations and Hard Work Behind Innovation
By
–
a mix of diffusion models, a bunch of ai and web and cpu optimizations and also unreasonable amounts of hard work and probably some luck too
-
AI Opportunistic Screening Detects Undiagnosed Diseases Early
By
–
The use of AI for opportunistic screening is just beginning. #AI can potentially catch previously undiagnosed diseases in the early stages, before the onset of symptoms, when they are easier to treat. https://
bbc.in/3PEWtGx #healthcare #TechNews #health #IoT -

Berlin AI Evening: Product Talks and Research with Industry Leaders
By
–
berlin: join us on 20th feb for an evening of ai product and research talks, and good ol fashion hangs with peers 🙂 announcing our first two speakers:
– @b_johannsen of sensmore (off-road ai)
– alex hammer of dunia (materials ai) request invite at airstreet(dot)com/berlinai -

ESM3 Breakthrough: AI Impact in Biological Research Published
By
–
big congrats to team @evoscaleai for esm3 making it into science! a testament to the importance and impact of ai in bio more to come…
-
Transformer²: Self-Adaptive LLMs with Neural Network Weight Adaptation
By
–
Blog post for Transformer²: Self-Adaptive LLMshttps://t.co/AyeFdqEKsd
— hardmaru (@hardmaru) 17 janvier 2025
Eventually, neural network weights should be as adaptive as the Octopus 🐙pic.twitter.com/me7urXJ6BSBlog post for Transformer²: Self-Adaptive LLMs https://
sakana.ai/transformer-sq
uared/
… Eventually, neural network weights should be as adaptive as the Octopus -
Hardcore Mode for Academic Conference Submissions
By
–
Some RPGs have hardcore mode: if your character dies, it gets deleted. Conferences should have a hardcore track: if your paper is rejected, you can't resubmit it, your entire codebase is deleted, etc. How much better would these submissions be?
-
Tencent Hunyuan Scaling Laws for Floating Point Quantization
By
–
Tencent Hunyuan new work: Scaling Laws for Floating Point Quantization Training
-

StarGen: Spatiotemporal Video Generation Framework for Scene Synthesis
By
–
StarGen: A Spatiotemporal Autoregression Framework with Diffusion Model for Scalable and Controllable Scene Generation https://
arxiv.org/pdf/2501.05763 https://
zju3dv.github.io/StarGen/