6+ concepts you should know to master AI:. - Test-time compute and test-time scaling.- AI inference.- RLHF and its variations: DPO, RRHF, RLAIF.- Meta-learning.- Causal AI.- Defense AI. All about them in these guides ->

6 AI concepts you should know in 2025. - Test-time compute and how to scale it.- AI inference.- RLHF variations: DPO, RRHF, RLAIF.- Meta-learning.- Causal AI.- Defense AI. Find everything from this list in one place:

【✨テックブログ更新✨】 . In-Context LearningはTest-Time Computeの恩恵を受けられるか？. 推論時間を増やしてIn-Context Learningの事例選択を改善することで、LLMの性能向上を実現できるか検証しました！.

Thinking Less at test-time requires Sampling More at training-time!. GFPO is a new, cool, and simple Policy Opt algorithm is coming to your RL Gym tonite, led by @VaishShrivas and our MSR group:. Group Filtered PO (GFPO) trades off training-time with test-time compute, in order

General Relativity took 8 years of Einstein's brain's test time compute. Once AI reaches into the task durations of years to decades it will begin to invent whole new theories about the physical world. This is the new scaling axis.

Reinforcement Learning of Large Language Models, Spring 2025(UCLA). Great set of new lectures on reinforcement learning of LLMs. Covers a wide range of topics related to RLxLLMs such as basics/foundations, test-time compute, RLHF, and RL with verifiable rewards(RLVR).

New research tackles a core challenge in LLMs. Go beyond memorization to truly.extend multi-step reasoning depth. Leveraging recurrence, memory, and test-time compute scaling is key.

We built 200k-GPU clusters; .We scaled up & curated higher-quality data;.We scaled compute by 100x;.We developed training & test-time recipes;.We made everything RL native;.We stabilized infrastructure and speeded up;. That's how you turn RL into the pre-training scale. Yet I am

Chain-of-Layers (CoLa) is the way to make test-time compute controllable. It treats model layers like building blocks that can be rearranged, so you can build custom versions of the model for each input. CoLa allows to:. - Skip layers for faster, simpler tasks.- Recurrently

Noise Hypernetworks: Amortizing Test-Time Compute in Diffusion Models. "we replace reward guided test-time noise optimization in diffusion models with a Noise Hypernetwork that modulates initial input noise.". "We show that our approach recovers a substantial portion of the

Our work on test-time scaling for robotics has been accepted to @corl_conf! We show that scaling test-time compute via a generate-and-verify paradigm offers a practical and effective path toward building general-purpose robotics foundation models.

Noise Hypernetworks. Amortizing Test-Time Compute in Diffusion Models

Increasing test-time compute can lead to more accurate LLM decisions, but also more unfair. How can we reap the benefits of modern inference techniques while also ensuring unbiased decision making? We explore this question in our new paper! 🧵

Happy to share RoboMonkey, a framework for synthetic data generation + scaling test time compute for VLAs: . Turns out generation (via repeated sampling) and verification (via training a verifier on synthetic data) works well for robotics too!. Training the verifier: we sample N

Too Much Thinking Can Break LLMs: Inverse Scaling in Test-Time Compute. Recent advances in large language models (LLMs) have encouraged the idea that letting models “think longer” during inference usually improves their accuracy and robustness. Practices like chain-of-thought

[LG] Inverse Scaling in Test-Time Compute.A P Gema, A Hägele, R Chen, A Arditi. [University of Edinburgh & EPFL & University of Texas at Austin] (2025).

What is test-time compute and how to scale it?.

The technical report for Goedel-Prover-V2 is out!. 📌 SOTA among all open-source theorem provers.⚡ Among the best overall—including closed-source—under small test-time compute. Read it here:

Parallel agents are emerging as an important new direction for scaling up AI. AI capabilities have scaled with more training data, training-time compute, and test-time compute. Having multiple agents run in parallel is growing as a technique to further scale and improve.