LLMs & Models
Large language models: GPT, Claude, Gemini, Mistral and open weights.
97 published articles
AI research
Over half your AI's reasoning is froth, and nobody noticed until now
LLMs often generate reasoning chains that are correct but padded with unnecessary steps. A new diagnostic benchmark shows current evaluators miss this inefficiency entirely, and half of human-written reasoning steps may be compressible.
2026-07-31
AI Research
A 4.5 point score jump on MATH-500 from a monitoring controller that catches wandering models
New research proposes an external monitoring controller for quantized small language models that detects repetitive or degenerating reasoning paths and triggers a rollback and constrained re-decoding. Accuracy improved by 4.5 percentage points on a broad evaluation set, but the authors stress the findings are not confirmatory.
2026-07-30
Game AI
MAGIC solved the three problems that kept AI from building real game worlds
MAGIC is a four-stage pipeline that converts a single prompt into a playable Unity project with multiple connected scenes. It enforces portal reachability with a flood-fill validator and introduces an evaluation agent that actually walks through each transition, decades of metrics that only looked at single interiors.
2026-07-30
MemTensor's Memory Foundation Models
AI agents kept memory outside the model. Metis puts it inside
MemTensor's Metis is the first prototype of memory foundation models: history compressed into the backbone, read back through memory attention, updated in one forward pass. It challenges the vector-store era of agent memory. The limits are acknowledged but not yet quantified.
2026-07-30
Reinforcement learning
Microsoft's experiential learning fix gives AI models a coach, not just a score
Experiential Learning repurposes the LLM-as-a-Judge into an LLM-as-a-Coach that extracts transferable knowledge from each response and internalizes it via on-policy context distillation, beating rubric-based RL on held-out tasks and reducing reward hacking.
2026-07-30
Mechanistic Interpretability
Gemma 4 knows physics well enough to steer it, and that changes everything
New research on gemma-4-E4B-it reveals that the model's internal representations of materials science mechanisms are causally linked to its answers. The study combines multiple probing and intervention techniques to show that physics knowledge is encoded in state transformations, not just static patterns.
2026-07-29
AI Research
Distributional RL heads make risk claims that are mostly false. A new audit proves it.
A large-scale audit shows that distributional RL agents systematically fabricate risk trade-offs at the very states where practitioners would most trust them. Across QR-DQN, C51, and IQN on MinAtar, zero of the strongest claims were confirmable, and acting on the agents' CVaR advice sometimes performed significantly worse than chance.
2026-07-29
Open Source AI
Nvidia's AV-Flamingo solves the one thing every other video AI gets wrong: time
Nvidia releases AV-Flamingo, an open audio-visual large language model designed for long, complex video understanding. It uses a three-stage curriculum and a timestamped chain-of-thought framework to handle temporal and cross-modal reasoning that trips up most models.
2026-07-28
Behavioral science
LLMs just bypassed the hardest part of behavioral economics: expensive surveys
A study using 3,000 GPT-5-based agents in route choice experiments shows LLMs reproduce cumulative prospect theory biases, loss aversion, reference dependence, probability weighting, with high fidelity. The approach bypasses the bottleneck of parameter estimation that limits traditional behavioral modeling.
2026-07-28
AI Models
Macaron-V1-Venti's 748B four-specialist design beats GPT-5.5 and Opus 4.8
Macaron-V1-Venti from MindLab Research leads benchmarks in personal intelligence, coding, terminal use, and Generative UI, using a novel Mixture of LoRA architecture that keeps the model compact while specialized.
2026-07-27
Open Source AI
Kimi K3 is the biggest open model ever. It's still not the best.
Kimi K3 is the largest open model at 2.8T parameters, but it fails to beat the best proprietary models on overall benchmarks. While it excels on specific coding and agentic tasks, the gap to frontier leaders like Claude Fable 5 and GPT 5.6 Sol exposes the limits of scaling without architectural and data breakthroughs.
2026-07-27
Agentic Finance
Why returns alone lie: NextFund opens the black box on AI trading agents
NextFund records every decision an AI trading agent makes in live markets, letting users compare models, inspect rationales, and diagnose failures. An eight-LLM test across US, China, and Hong Kong equities shows that similar intermediate signals can diverge into starkly different portfolio behaviors, and that quarterly rankings flip depending on which metric matters most.
2026-07-27