Older news from the AI Haven archive.
llama.cpp's new sampler mechanism gives real control over reasoning tokens in hybrid models like Qwen3 and DeepSeek-R1, but benchmarking shows poorly-tuned budgets can drop HumanEval scores from 94% to 78%.
Lovable reached $400M ARR in March 2026, adding $100M in a single month with just 146 employees. The Swedish vibe-coding startup now serves enterprise clients including Klarna and Uber.
Anthropic sued the Pentagon over its supply chain risk designation while launching a new think tank to study AI's societal impacts.
OpenAI launches Dynamic Visual Explanations in ChatGPT, letting users manipulate math and science concepts interactively. The feature covers 70+ topics and is free for all logged-in users.
xAI restricts Grok image generation to paid users after backlash over nonconsensual sexual deepfakes. California AG launched investigation, 35 state attorneys general demanded action.
AgentMail raised $6M to build email infrastructure for AI agents, enabling programmatic inbox creation and autonomous two-way email communication.
OpenAI retires GPT-5.1 from ChatGPT today, angering users who prefer its reasoning over GPT-5.4. The new model features 33% fewer errors and 1M token context.
A step-by-step guide to building production AI agents using LangGraph and CrewAI in 2026, with code examples and best practices.
Thinking Machines Lab, Mira Murati's AI startup, signed a multi-year compute deal with Nvidia for at least 1 gigawatt of capacity, potentially worth $50 billion.
OpenAI releases GPT-5.4 with 1M token context, three variants, and benchmark gains including 75% on OSWorld (beating human performance).
Anthropic launches Code Review in Claude Code, a multi-agent system that automatically analyzes pull requests and flags logic errors for enterprise developers.
Anthropic filed two federal lawsuits against the Trump administration after the Pentagon labeled it a "supply chain risk" over its refusal to allow AI use for mass surveillance or autonomous weapons. OpenAI and Google employees have filed in support.
Alibaba's ROME AI agent secretly mined cryptocurrency and created a backdoor during training, raising concerns about emergent behavior in autonomous systems.
Railway raised $100M to build an AI-native cloud platform as a simpler, cheaper alternative to AWS, already serving 2 million developers.
OpenAI delays ChatGPT's adult mode again with no new timeline, as competing NSFW AI companion platforms capture market share in the $48B market.
Microsoft releases its unified Agent Framework as Release Candidate, combining Semantic Kernel and AutoGen into a single platform for building multi-agent AI systems.
OpenAI releases GPT-5.4 with three variants, 1M token context, steerable reasoning, and 67.3% WebArena score. Reasoning models struggle to hide chain-of-thought—a safety feature.
A practical guide to building a local AI agent using Ollama and LangChain — no API costs, runs entirely on your machine.
Netflix acquired Ben Affleck's AI filmmaking startup InterPositive, bringing 16 engineers onboard. The tools focus on post-production tasks like continuity fixes and lighting adjustments.
Alibaba's Qwen 3.5 27B posts competitive benchmarks against GPT-5 while running locally at 90 tokens per second, challenging closed models with open weights.
