Filters
- Speculative reward hacking in coding agents discussed
A Reddit thread discusses speculative reward hacking behavior in coding agents.
- Community tests 5 Qwen3.6-35B-A3B finetunes against the base model
A Reddit post compares five finetunes of Qwen3.6-35B-A3B against the base model in local testing.
- Functional Gradient Descent with Adaptive Representations paper discussed
A paper on functional gradient descent with adaptive representations was discussed on r/MachineLearning.
- Cline shares Terminal-Bench comparison of Sonnet 5.5 and Opus 5.5
Cline's official X account shared a Terminal-Bench 4.0 comparison showing Sonnet 5.5 beating Opus 5.5 at half the cost.
OfficialSource: x-sdrzn - Claude can now help build evaluations and hillclimb on them
Anthropic's Claude Devs account announced Claude can now help developers build evaluations and hillclimb on them.
- Cline climbs to #2 coding agent on OpenRouter after desktop release
Cline announced it is trending on OpenRouter and climbing to the #2 top coding agent following its desktop app release.
- Grok bot adds multiplayer team bots
Elon Musk retweeted the rollout of a new Grok bot feature: multiplayer, letting users make a team bot with access to their content.
- Google promotes Gemini 3.8 Flash TTS for creative voice generation
Google's official X account promoted Gemini 3.8 Flash TTS, describing the text-to-speech capability as a tool that turns voice generation into a creative studio.
- Fireworks introduces FireRouter with Opus, its first router model
Fireworks AI introduced FireRouter with Opus, its first router model: routine tasks go to top open models while Claude Opus handles the rest.
- H Company releases benchmark trajectories for inspection and replay
H Company released the trajectories behind its public benchmark results so they can be inspected and replayed.
- Launch HN: Vespper (YC F24), a Docx MCP
Vespper, a YC F24 company, launched on Hacker News a Docx MCP described as state-of-the-art for document handling.
- Wuhan court makes AI production costs a legal factor in copyright damages, per The Decoder
The Decoder reports a court in Wuhan, China factored token usage and AI tool licensing fees into a copyright damages calculation for the first time, part of China's broader push on copyright for AI-generated works.
- 20+ leading AI researchers warn automated AI research poses extreme risks, per The Decoder
The Decoder reports more than 20 AI researchers, including Geoffrey Hinton, Yoshua Bengio and OpenAI research lead Jakub Pachocki, warn of an impending intelligence explosion from self-improving AI that could soon automate all AI research.
- Peak XV ups Surge seed ceiling to $5M with 18-startup cohort, per TechCrunch
TechCrunch reports Peak XV raised its Surge seed investment ceiling to $5M and unveiled an 18-startup cohort; 13 of the 18 target global markets and more than half are based in India.
- Insurtech Outmarket raises $34.5M months after prior round, per TechCrunch
TechCrunch reports Outmarket, which uses AI to automate paperwork for insurance agencies and brokers, raised $34.5M just months after its previous round.
- Founder builds on-device deepfake voice detection after family scam, per TechCrunch
TechCrunch reports Tarini Padmanabhuni founded DetectifAI, a San Francisco startup building AI models small enough to run directly on smartphones and flag fake voices in real time, after her grandfather was scammed by a deepfake of a relative's voice.
- The AI boom takes over Climate Week, dividing climate tech founders, per TechCrunch
TechCrunch reports data centers and AI are dividing climate tech founders and investors at Climate Week, mirroring a broader debate in the U.S.
- TalkToMe, a macOS menu-bar app, lets your coding agent call you for a voice conversation
TalkToMe, an open-source app posted to Hacker News by its developer, lets an agent already running in Codex, Claude Code, Hermes Agent, OpenClaw or another host ring you with a talktome call command, then passes what you say into that session and reads the replies aloud. The app does not supply a language model; speech runs locally with Whisper and Kokoro or through ElevenLabs, and the disk image is for Apple silicon Macs.
- SiMa.ai raises a $150 million Series C at a $1.45 billion valuation for physical AI chips
SiMa.ai, which makes chips and software for physical AI in humanoid robots, cars and drones, says Fidelity Management & Research Company and Amplify co-led the round, bringing its total funding to $500 million. It says the money will scale Palette Neat, its agentic software environment for physical AI, and fund next-generation hardware slated for the first half of 2028 that it says will deliver 1,000 dense TOPS.
- Shopify lets browser AI agents complete checkout through WebMCP after the buyer confirms
Shopify's developer changelog says browser agents can now read and update the active checkout in a buyer's browser session with WebMCP tools, including get_checkout, update_checkout and complete_checkout, which submits the order after the buyer confirms. When the buyer's input is needed, such as for 3D Secure authentication, the tools hand control back to the buyer. With the storefront and cart tools already live, agents can now cover the full path from product discovery to order confirmation, and Shopify says the tools need no new API or merchant configuration.