Today’s lead · TheHackerNews
EU orders Android access for rival AI assistants
The European Commission ordered Google to give rival AI assistants Android access comparable to Gemini, including camera, microphone, screen context, wake-word invocation, background execution, and app-driving capabilities.
Top signals
4 more
Shanghai AI Lab releases Intern-S2-Preview-397B
Shanghai AI Lab released Intern-S2-Preview-397B, described as an Apache-2.0 open-weight scientific multimodal MoE for scientific intelligence and long-horizon agents.
NadMesh botnet targets exposed AI services
A Go botnet called NadMesh is scanning exposed AI and developer services to steal cloud credentials, Kubernetes service account tokens, model inventories, and MCP tool access.
Claude Fable 5 gets paid-plan access
Claude Fable 5 will be included in Max and Team Premium plans from July 20 at 50% of usage limits, while Pro and Team Standard users keep access through usage credits.
Patreon starts blocking AI training crawlers
Patreon is working with Cloudflare to actively block AI training bots from scraping creator content, moving beyond robots.txt requests.
Tools & repos
3 selectedgithub/copilot-sdk
Multi-platform SDK for integrating GitHub Copilot Agent into apps and services.
PostHog/posthog
PostHog is a developer platform for building self-driving products with AI observability, analytics, session replay, flags, experiments, error tracking, logs, Slack, web, desktop, and MCP surfaces.
Unabyss for Claude
Unabyss for Claude gives Claude shared memories from other AI agents and everyday apps, aiming to keep GPT, Cursor, and Claude in sync with the same context.
Blogs worth your time
1 reads
Fine-tune video and image models at scale with NeMo Automodel and Diffusers
NVIDIA and Hugging Face describe an open-source NeMo Automodel integration that brings distributed diffusion fine-tuning to Diffusers-format models on the Hugging Face Hub without checkpoint conversion or model rewrites.
Community discussions
4 threadsDeepSeek V4 Flash at 1M context in llama.cpp
A LocalLLaMA user shared a llama.cpp setup for DeepSeek V4 Flash with a 1,048,576-token context, prompting debate over whether very large MoE models can be practical on high-end local hardware.
LocalLLaMA debates Gemma4-31B vs Qwen3.6-27B for coding agents
A user said switching Gemma4-31B into the main coding-agent role unblocked a prototype after Qwen3.6-27B struggled, while commenters cautioned against drawing broad model conclusions from one project.
Where do multi-agent systems beat a single agent?
An AI_Agents thread asked for concrete examples where multiple agents clearly outperform one strong agent, with replies emphasizing coordination overhead, specialized workflows, and parallelism only for sufficiently complex tasks.
Voice-agent builders focus on TTS latency
A builder working on inbound support and outbound qualification voice agents asked what TTS systems people use in serious testing or production, arguing that time-to-first-byte matters more than award-winning voice quality.
Funding & acquisitions
3 moves
Databricks raises at $188B valuation
Databricks announced a new funding round valuing the company at $188 billion as it doubles down on AI products around governance, agents, and AI-native data infrastructure.
General Compute lands $400M inference-chip loan
General Compute, an AI inference cloud startup, secured a $400 million loan from Upper90 backed by inference-specific chips, signaling investor demand for cheaper open-model inference infrastructure.
H2LooP raises $2M seed for embedded AI coding tools
Bengaluru startup H2LooP raised seed funding to build hardware-aware AI models for firmware, drivers, and embedded systems used in semiconductors, automotive, aerospace, telecom, defence, data centres, drones, and robotics.
Bengaluru radar
0 eventsThere are no relevant Bengaluru events to highlight today.