All Stories — Page 87

Founder Chaitanya Giri turned Reddit criticism into a Claude Code workflow built from SVG parts, 16 composition archetypes and one media manifest.

Shuo Yang, Xiaoze Fan and collaborators report the decode rate using CPU compute, a 32 GB RTX 5090 and 192 GiB of system memory.

VIDRAFT's new leaderboard uses hidden labels, harder test sets and assay noise floors to make drug-prediction scores less flattering and more useful.

After an earlier video passed 1 million views, the solo developer put a walkable browser prototype on Ko-fi during the week of Aug. 17th-21st.

Co-creator David Soria Parra is steering the protocol toward long-running, delegated work across enterprise systems.

Chaitanya Giri's local-first harness wraps 12 agent CLIs. Its v0.4.5 release notes document repairs to cost reporting, Apple Silicon semantic memory and communication between workers.

Dan Luu argues coding agents have made once-specialized optimization cheap enough to customize software for each workload.

Outer Biosciences, Michael Polansky's 19-person company, uses machine learning and human skin tissue that remains viable for four weeks to find compounds for beauty and pharmaceutical companies.

The 3x YC founder built a staged, Markdown-based knowledge layer for Claude, ChatGPT and Cursor, priced from free to $99 per month.

Product manager Jin-Hee Lee's feature is headed for an upcoming rollout, publishing Search, Agent and Training choices while preserving existing Disallow rules.

Its B:OS demo took Nemotron 3 Nano from unsupported to 1.43x faster decode than MLX on Apple silicon, by Base Compute's tests.

Virgil Dupras's 11-chapter tutorial turns his collapse-minded OS work into a path beneath the modern software toolchain.

ICONIQ led the Series C after Rillet reported that its annualized revenue rate had doubled in one quarter, though the company did not disclose its underlying revenue.

Google's August 21st release notes list the model as a Preview offering, while SpaceXAI quotes $0.30 cached input and a 500,000-token context window.

An undated MiniMax post mentions M3 and SambaNova hardware, while the companies' disclosed demonstration identifies M2.7 and provides no M3 setup or result.

The flagship model now costs $4 per million input tokens and $20 per million output tokens, while paid-plan usage limits stay put.

After showing a 120-source counter, Grok treated a disputed $23.6B data-center tally as evidence of a foreign-directed US campaign. ChatGPT found that the underlying sources did not establish the money, instructions or control required for that conclusion.

Discovery Loop has announced an undisclosed initial round co-led by Khosla Ventures and Radical Ventures as Jeff Dean and three Google veterans build a machine-learning research system first.

Intern-S2-Preview combines visual pretraining, tool use and long-horizon reinforcement learning in an open 397B-parameter scientific model.

Access now includes SuperGrok Plus, Cursor Pro+, and all Cursor Teams plans, while every other user gets a usage-limited trial.

Powered by Claude Opus 5, the coding-agent system cleared 183 levels in 6,624 actions, though NVIDIA did not test the private sets.

Claude Security entered public beta for Claude Enterprise customers in late April 2026; Anthropic keeps the more capable Mythos 5 in a separate trusted-access program.

Shenzhen-based Morpho reported nearly 2,000 backers in 16 hours; Morpho Studio is operational, though custom-jig features and a roll-to-roll UV-DTF attachment remain in development.

Moonshot launched the 2.8-trillion-parameter model on July 16th and released its weights 11 days later, pairing a one-million-token context window with CEO Yang Zhilin's long-running research thesis.