Jason Says
Today's biggest signal: Perplexity running Astra in production and Devin self-testing with Astra means 'AI supervising AI' is no longer a concept—it's shipping, and the engineer's role is shifting faster than anyone expected.
Perplexity now uses GPT-6 Astra across its full system stack—writing comms, modifying software, and monitoring production—with far less human oversight than earlier models. This marks a significant shift: Astra is the first model publicly trusted with end-to-end production responsibility at a major AI company.
Cognition's Devin now uses GPT-6 Astra to automatically test and demonstrate its own software output, reducing the code review burden on engineers. The 'AI testing AI' loop signals a major shift in developer workflows—less reviewing, more shipping.
Anthropic's Dario Amodei has outlined a plan to 'pace the frontier,' and OpenAI's Sam Altman appears to agree. This rare alignment between the two leading AI labs on intentionally slowing down development could reshape regulatory conversations and industry timelines globally.
Despite filing confidentially for an IPO, OpenAI CEO Sam Altman says going public in 2026 would be 'ill-advised.' With GPT-6 Astra demand surging and the business still in hyper-growth, Altman is choosing to stay private—a clear signal on AI unicorn IPO timing expectations.
Mighty Capital's SC Moatti argues AI itself is no longer a durable moat—only counter-positioning (business models incumbents can't copy) and network economies truly hold up. A must-read framework for indie developers and SaaS founders trying to build defensible AI products.
DeepSeek releases v4.1-Flash with a novel 763B causal encoder-decoder architecture (8B active params, 16B experts) and vision capabilities—significant enough that the community calls it 'should have been v5.' A major open-model release that reshapes the competitive landscape against GPT-6 Astra.
Researchers built the first controlled pure-autoregressive testbed to study how image tokenizer choice affects joint text-image modeling. For developers building multimodal systems: tokenizer selection directly impacts generation and understanding quality—this study gives you quantifiable criteria to choose wisely.
Lumae is a native macOS screen recorder with built-in MCP integration, enabling AI agents to directly edit recorded demos. It's a notable example of MCP expanding into video content creation workflows—relevant for indie developers and content creators looking to automate their demo production pipeline.
Investors: Sequoia Capital
机器人训练数据专业提供商，Sequoia 领投新一轮，估值逼近 5 亿美元，受益于具身智能数据需求爆炸式增长。
Subscribe for daily AI updates + free playbook
📘 Subscribe Free