AI and Web 3 News: ChatGPT5 Turmoil

AI News

1. GPT-5 launches—and hits early turbulence: OpenAI’s new flagship rolled out with a smarter router and deeper reasoning, but early users reported glitches, higher perceived costs, and uneven quality. Coverage suggests expectations have been reset: businesses are upgrading for coding and research, while general users are split. DeepLearning.aiThe Washington Post

2. Microsoft wires GPT-5 into Copilot: Microsoft says its entire Copilot stack—Microsoft 365, GitHub Copilot, and Azure AI Foundry—now runs on GPT-5, with “Smart mode” auto-routing tasks and bigger context windows. The move leans into workflow grounding across emails, docs, and code rather than pure chat. The Rundown AI

3. Reinforcement learning quietly boosts ad text at scale: A Meta team reports that applying RL to ad copy generation improved online ad outcomes versus strong supervised baselines, with both offline and live tests showing gains. For growth teams, it’s a signal that bandit/RL tuning on top of LLMs is moving from lab to production. Import AI

4. Brains and LLMs rhyme on meaning: A new Nature Machine Intelligence study finds that brain activity during language tasks aligns with embeddings from state-of-the-art LLMs, strengthening evidence that model representations capture human-like semantics. Useful for neuro-aligned interfaces and evals that go beyond benchmarks. Import AI

5. DeepMind’s Genie 3 turns pixels into playable worlds: The latest Genie 3 trains a generative world model that produces interactive environments and supports multitask RL—pointing to richer sim-to-real training and agent testing without hand-built levels. Early demos show controllable scenes from video frames. Import AI

6. Bug hunters puncture “BigSleep” defenses: Security researchers reported multiple vulnerabilities in Google’s BigSleep AI isolation framework, earning bounties via IssueTracker. If your safety posture mirrors BigSleep, assume adversarial testing will find cracks—threat-model and patch accordingly. Import AI

Web3 News

1. Liquid staking smashes a fresh record: Sector TVL topped $86B as ETH rallied, with Lido near half the pie and weekly spot-ETH ETF inflows setting new highs. Regulatory clarity from an SEC division on what does—and doesn’t—constitute securities activity in liquid staking is fueling institutional participation. The Defiant

2. ETH flirts with its all-time high: The community is bracing as ether approaches prior peaks, with rising on-chain activity and sentiment tailwinds from ETF demand. If momentum holds, rotation into staking and L2 activity could intensify. bankless.com

3. Google clarifies stance on self-custody wallets: After backlash over compliance wording, Google updated guidance to confirm that non-custodial crypto wallets aren’t targeted by the rules—easing fears for devs and app distributors building self-custody experiences. bankless.com

4. On-chain data: broad upswing, but watch risk: Glassnode’s latest Week On-chain highlights a market-wide lift across majors and improving structure, while reminding that liquidity and profit-taking dynamics can swing quickly near highs. Keep an eye on ETF flows and realized caps as stress markers.Glassnode Insights