Created by Tina
Cutting‑edge AI research plus practical reviews of productivity, coding, and creative tools
Explore the latest content tracked by AI Research & Tools
Anthropic unlocks enterprise AI dominance with a bold tactic: BUY your customers. They've shifted from selling intelligence to infrastructure like Claude Managed Agents, targeting fintech plays amid agentic tools from Visa and Ramp.
MiniMax released open-weights for its M2.7 model, the latest boost to surging open-source frontier models enabling accessible AI experimentation via minimax.io.
TRL now supports on-policy distillation from 100B+ teachers like Qwen3-235B, 40x faster than naive methods. Distilled to a 4B student gaining 39+ points on AIME25 via two key optimizations—small open models just rivaled giants.
Did Meta sacrifice its open-source identity for a competitive AI model? This pivotal choice will define Meta's AI story, amid enterprise pushes like Manus.
Capabilities shine in multimodal reasoning and health tasks (42.8 on HealthBench Hard), scoring 52 on Intelligence Index—trailing Gemini 3.1...
Open weights models like Google's Gemma 4 and Qwen 3.5 shine for enterprises, bridging the frontier AI gap with practicality:
Physicians are using Anthropic's Claude Code to build custom tools for patient care, showcasing real-world AI coding in medical productivity.
Diverging paths for frontier AI emerge from recent releases:
ChatGPT's internet browsing feature is back after months offline, letting users access real-time information beyond the model's September 2021 training cutoff—supercharging everyday utility.
Critical bug for Claude API devs: Anthropic silently downgraded prompt cache TTL from 1h to 5m around March 6, 2026.
Neural Computers (NCs) unify computation, memory, and I/O in a learned runtime state, making the model itself the running computer—challenging agent...
OpenMontage breakthrough for creators:
Flow Map Language Models receive a 🤯 big update, hailed as the future of non-autoregressive text generation. Introduces a new class of continuous flow-based models—check the paper and blog for details.
Open-weight models are catching up fast to closed models, as highlighted in Stanford's new 457-page AI Index report packed with data.
Bold AI autonomy test: Leased 3-year SF retail space tasked to make a profit.
Emerging tools are making pro-level video effects accessible:
Meta AI and KAUST's new paper proposes Neural Computers (NCs) as learned runtimes, where models integrate computation and memory—turning the model into the computer itself. Bookmark-worthy breakthrough in AI architecture.
Google, OpenAI, and Anthropic models lost money betting on football matches over a full Premier League season, per @GenReasoning's KellyBench study—exposing limits in sustained real-world analysis even for top systems.
