Current RL fine-tuning for LLMs suffers from expensive training, critic networks or advantage estimates, and inefficient exploration. Enter A*-PO: A two-stage policy optimization method that flips the standard approach — slashing compute and memory costs compared to
RESEARCH
-

Text-to-LoRA: Instant Transformer Model Adaptation Technique
By
–
Text-to-LoRA: Instant Transformer Adaption Charakorn et al.: https://
arxiv.org/abs/2506.06105 #ArtificialIntelligence #DeepLearning #MachineLearning -

Diffuse and Disperse: Image Generation with Representation Regularization
By
–
Diffuse and Disperse: Image Generation with Representation Regularization
Paper: https://
arxiv.org/pdf/2506.09027
v1.pdf
… -

Diffusion Models: Improving Internal Representations Through Explicit Regularization
By
–
Diffusion models have been wildly successful for image and text generation, but they rely on regression-based objectives and lack explicit representation regularization. As a result, their internal representations are underutilized or poorly structured. This paper introduces
-
Text-to-LoRA: Generating Task-Specific Model Adapters from Text
By
–
Text-to-LoRA: Instant Transformer Adaptionhttps://t.co/DNy6gzwQKN
— hardmaru (@hardmaru) 12 juin 2025
Generative models can produce text, images, video. They should also be able to generate models! Here, we trained a Hypernetwork to generate new task-specific LoRAs by simply describing the task as a text prompt. https://t.co/QkSaBrZU22Text-to-LoRA: Instant Transformer Adaption https://
arxiv.org/abs/2506.06105 Generative models can produce text, images, video. They should also be able to generate models! Here, we trained a Hypernetwork to generate new task-specific LoRAs by simply describing the task as a text prompt. -
Human and LLM reasoning: apples and potatoes analogy
By
–
Human and LLM reasoning are a different as apples and potatoes. That is, one is a sweeter version of the other, with a shinier peel, but both grow on vines and are often interbred.
-
Snorkel launches expert data platform for frontier LLMs
By
–
It’s not about more data, it’s about better data.
— Snorkel AI (@SnorkelAI) 11 juin 2025
That’s why we built Snorkel Expert Data-as-a-Service: real experts + programmatic dev to power frontier LLMs
Today, we're previewing our new Snorkel Leaderboard:
🔗 https://t.co/Kl7hDRpNJC
Interested 👉 expertdata@snorkel.ai https://t.co/JPL5I3zbCQIt’s not about more data, it’s about better data. That’s why we built Snorkel Expert Data-as-a-Service: real experts + programmatic dev to power frontier LLMs Today, we're previewing our new Snorkel Leaderboard: http://
leaderboard.snorkel.ai Interested expertdata@snorkel.ai -
AGI Achievement Requires Product Not Just Model
By
–
AGI is going to be achieved by a product, not necessarily a “model”.
-
V-JEPA 2: Advanced World Model for Visual Understanding and Robot Planning
By
–
Introducing V-JEPA 2, a new world model with state-of-the-art performance in visual understanding and prediction.
— AI at Meta (@AIatMeta) 11 juin 2025
V-JEPA 2 can enable zero-shot planning in robots—allowing them to plan and execute tasks in unfamiliar environments.
Download V-JEPA 2 and read our research paper… pic.twitter.com/Zdibfs8ZMQIntroducing V-JEPA 2, a new world model with state-of-the-art performance in visual understanding and prediction. V-JEPA 2 can enable zero-shot planning in robots—allowing them to plan and execute tasks in unfamiliar environments. Download V-JEPA 2 and read our research paper
-
Automation and Job Loss: MIT Economist Challenges Simple Narrative
By
–
Does automation always result in job loss? MIT economist @davidautor suggests there's more to the story. Learn more about his framework on evaluating the impact of automation on jobs: https://
hai.stanford.edu/news/assessing
-the-real-impact-of-automation-on-jobs
…
