The Archive

Benchmark of tabular foundation models on credit default prediction shows context strategy matters more than model choice.

Position paper proposes treating neural network checkpoints as a first-class generative modality for on-demand weight synthesis.

SCICONVBENCH benchmarks LLMs on multi-turn clarification dialogues for ill-posed scientific task formulation.

Aligned training, a parameter-free SAE reparameterization, eliminates dead features and improves interpretability stability.

Method for learning lifted STRIPS+ action models from traces with minimal state/action information assumptions.

Reddit thread with no content; insufficient information to assess technical or business significance.

CVAE framework enables targeted evasion of ML-based malware detectors via API import injection without retraining.

First Õ(log^1.5 n)-approximation algorithm for graph label selection under budget constraint.

CrossView Suite adds dataset, model, and benchmark for cross-view spatial reasoning in MLLMs with object-level consistency.

SPBM extends penalty-barrier methods to non-convex, non-smooth stochastic optimization for constrained deep learning.

ManiSoft benchmark for vision-language manipulation of soft robotic arms with elastic dynamics simulation and contact-rich task evaluation.

Community speculation about Qwen planning 3.7B parameter model releases; unconfirmed social media discussion.

SAME autoencoder achieves 4096× music compression via transformer backbone and semantic regularization for audio generation.

CATA framework for continual machine unlearning in vision-language models via conflict-averse task arithmetic.

Theoretical analysis of classical momentum acceleration in mini-batch SGD for large-scale model training.

Proxy metrics from token-level statistics predict downstream LLM capabilities faster than cross-entropy loss.

PACE-FNO uses Lie-algebra symmetries to improve neural operator generalization on PDE solution maps.

Anyone else also dealing with random $5.44 charges from Claude???? I never paid for Claude or had my card hacked before. I canceled my card but really strange

Pointwise Riemannian Dimension framework characterizes deep network generalization via learned feature representation geometry.

LAR compresses LLM agent action spaces into latent multi-step behaviors to reduce inference cost and decision horizon.

Typographic attacks via printed text override CLIP-based perception in simulated household robot manipulation pipelines.

AMARIS accumulates evaluation diagnostics across RL steps to adaptively improve rubric-based reward shaping for LLM fine-tuning.

The defense-tech company Anduril has shared new details about the augmented-reality headset for the military it’s prototyping with Meta, including a vision for ordering drone strikes via eye-tracking and voice commands. Quay Barnett, who leads the efforts as a vice president at Anduril following a career in the Army’s Special Operations Command, says his fundamental…

Alexa Plus, Amazon's upgraded AI assistant, can now generate podcasts on "virtually any topic," according to an announcement on Monday. With the update, Amazon says you can give Alexa Plus a topic, and the AI assistant will offer an overview of what its AI hosts plan to talk about, allowing you to steer the conversation and adjust its length before it starts generating the episode. Some "Alexa Podcast" examples shared by Amazon have two AI-generated hosts talking about the history of the Roman Empire, new music, and expectations for the World Cup. Amazon says you can also ask Alexa Plus to ge...

Just wanted to share that I'm pretty happy about Qwen 35b a3b agentic coding performance. I'm running the model in q80 quant, kv cache both q8\_0 as well, with 262144 in 4090 + 5060 ti, via llama.cpp backend with claude code pointing to localhost. For demo/data analytics purposes, it works pretty well. I haven't used it for large codebases, but it definitely is better than gemma4 26b in my use case. One thing that surprises me is that it seems to get better outcome in agentic coding, than chat. When using it with just chat UI, i found the code qwen35b provide a bit too clunky. I wonder o...

Experienced Claude user shares 11 practical tips on Projects, Custom Styles, and Claude Code features after 18 months daily use.

Creator of HeroMachine character generator used Claude Code to complete a long-stalled project over a weekend.

Qwen 3.7 released on Qwen Chat platform, continuing open-weights model availability from Alibaba.