Skip to content
Continued

Era 06 · 2023–present

Large Model Era

Natural language became a new way to interact and to program

Start
In 2023, GPT-4 and Ernie Bot were released; the generative AI measures took effect, and the first batch of filed large models went live
End
to the present

Background and drivers

  1. Hardware: GPU compute became a strategic resource; export controls on high-end chips pushed domestic compute and more efficient training methods.
  2. Networks and infrastructure: the construction of intelligent computing centers, under "Eastern Data, Western Computing".
  3. Capital: capital concentrated on foundation models and AI applications; in 2024-2025, open-source models (Llama, Qwen, DeepSeek) sharply drove down prices, and after the "war of a hundred models" the industry entered a price war.
  4. Users: natural language became a new way to interact, and ordinary users used conversation directly to write, search and program.
  5. Policy: China adopted a filing system for generative AI and required content labeling; the EU AI Act took effect; the 2025 amendment of the Cybersecurity Law (in effect 2026-01-01) wrote in provisions supporting and regulating artificial intelligence.

Pain points

  1. Large models spout nonsense with a straight face (hallucinations), and they do not know a company's internal or latest information.
  2. Training and inference are costly and high-latency, making it hard to launch at scale.
  3. A model can only "talk", not "do": it cannot connect to business systems or complete multi-step tasks.
  4. New attack surfaces: prompt injection, jailbreaks, and models being used to generate scams and fake content.
  5. The copyright and personal information of training data, and the difficulty of telling real generated content from fake, mean both regulators and users demand traceability.
  6. The way software roles work is being rewritten: code can be generated, so how do you define quality, review and responsibility?
  7. Dependencies grow ever more concentrated: a problem at a single model vendor, in one open-source component, or in a single security-software update can cause widespread cascading failures (the 2024 xz backdoor, the CrowdStrike outage).

New technology and practices

  1. Retrieval-augmented generation (RAG), vector databases and knowledge bases -> pain point 1: retrieve trusted material first, then have the model answer based on that material and cite its sources.
  2. Open-source models, quantization, LoRA/QLoRA fine-tuning, inference frameworks (such as vLLM's PagedAttention) and mixture-of-experts (MoE) architectures -> pain point 2: prices and barriers dropped dramatically.
  3. Tool calling, AI agents and the Model Context Protocol (MCP, released in 2024) -> pain point 3: models can call APIs, operate browsers, and write and run code.
  4. Alignment training (RLHF), red-teaming, guardrails (input/output filtering) and least-privilege tool authorization -> pain point 4.
  5. A filing system, explicit and implicit labeling of generated content (in effect 2025-09-01), watermarking, and compliance review of training data -> pain point 5.
  6. AI coding assistants and code agents, together with benchmark sets, automated testing and code review standards -> pain point 6.
  7. Software supply chain security (SBOM, dependency signing), multi-model redundancy and canary releases -> pain point 7.

Roles of this era

  • FrontendFrontend Engineer

    Using AI to generate and rewrite code, building conversational interfaces (streaming output, generative UI), and building operating interfaces for agents.

  • BackendBackend Engineer

    Wrapping business capabilities into tools the model can call (API/MCP), and handling asynchronous long-running tasks, cost and rate limiting.

  • ProductProduct Manager

    AI product managers who define "the boundary of model capability plus a fallback experience", using evaluations to replace part of the requirements document.

  • DesignInteraction Designer

    AI-native interaction (conversation, multimodality, presenting uncertain results), with AI-generated assets entering the design workflow.

  • OpsDevOps Engineer (coming soon)

    LLMOps and GPU cluster operations, responsible for model versions and for the stability and cost of inference services.

  • SecuritySecurity Engineer

    AI security (prompt injection, jailbreaks, the model supply chain), content security and filing compliance, and labeling of generated content.

Security & law thread

  • ChinaIssued 2022-11-25, in effect 2023-01-10: the Provisions on the Administration of Deep Synthesis of Internet Information Services (Order No. 12 of the Cyberspace Administration of China and two other departments) (original text (in Chinese)(opens in a new window)).
  • ChinaIssued 2023-07-10, in effect 2023-08-15: the Interim Measures for the Administration of Generative Artificial Intelligence Services (Order No. 15 of the Cyberspace Administration of China and six other departments) (original text (in Chinese)(opens in a new window)).
  • China2023-08-31: the first batch of large models passed filing and went live.
  • China2023-10-25: the National Data Administration was unveiled.
  • ChinaIssued 2024-09-24, in effect 2025-01-01: the Regulations on the Administration of Network Data Security (State Council Order No. 790) (original text (in Chinese)(opens in a new window)).
  • ChinaRevised 2024-12-06, in effect 2025-01-20: the second revision of the Measures for the Administration of Internet Information Services (State Council Order No. 797, the State Council's Decision on Amending and Repealing Certain Administrative Regulations), which is the current version of these measures (current version (in Chinese)(opens in a new window)).
  • ChinaIssued 2025-03-07, in effect 2025-09-01: the Measures for Labeling AI-Generated and Synthetic Content (Guo Xin Ban Tong Zi [2025] No. 2), with the supporting mandatory national standard GB 45438-2025 (original text (in Chinese)(opens in a new window)).
  • ChinaPassed 2025-10-28, in effect 2026-01-01: the amendment to the Cybersecurity Law (Presidential Order No. 61), which added Article 20 to support the development of artificial intelligence and its security regulation (current version (in Chinese)(opens in a new window)).
  • Global2024-08-01: the EU's Artificial Intelligence Act (Regulation (EU) 2024/1689, signed 2024-06-13, published in the Official Journal 2024-07-12) entered into force, applying in phases (official journal text(opens in a new window)).
  • Global2026-07-27: the EU's Regulation (EU) 2026/1744 (the Digital Omnibus on AI, passed 2026-07-08, published in the Official Journal 2026-07-24) entered into force, amending the Artificial Intelligence Act: the obligations for high-risk AI systems in Annex III were postponed to apply from 2027-12-02, and those in Annex I to 2028-08-02 (official journal text(opens in a new window)).
  • Security incidentsFrom 2023 on, prompt injection and jailbreak attacks became a new category of attack.
  • Security incidents2024-03-29: the xz Utils backdoor (CVE-2024-3094) was discovered, a long-dormant open-source maintainer poisoning the supply.
  • Security incidents2024-07-19: a CrowdStrike update knocked a large number of Windows devices offline worldwide (not an attack, but it exposed the single-point risk of security software).

Key events

  1. The Provisions on the Administration of Deep Synthesis of Internet Information Services took effect

    Sources: cac.gov.cn(opens in a new window)

  2. Meta released LLaMA, starting the open-source large model wave

    Sources: arxiv.org(opens in a new window)

  3. OpenAI released GPT-4

    Sources: arxiv.org(opens in a new window)

  4. Baidu released Ernie Bot

    Sources: zh.wikipedia.org(opens in a new window)

  5. The Interim Measures for the Administration of Generative Artificial Intelligence Services were issued (in effect 2023-08-15)

    Sources: cac.gov.cn(opens in a new window)

  6. The first batch of domestic large models passed filing and opened to the public

    Sources: news.cn(opens in a new window)

  7. The National Data Administration was unveiled

    Sources: gov.cn(opens in a new window)

  8. OpenAI released the Sora video generation model

    Sources: openai.com(opens in a new window)

  9. The xz Utils backdoor was discovered

    Sources: en.wikipedia.org(opens in a new window)

  10. A CrowdStrike update triggered a global outage

    Sources: en.wikipedia.org(opens in a new window)

  11. The EU's Artificial Intelligence Act entered into force

    Sources: eur-lex.europa.eu(opens in a new window)

  12. Anthropic released the Model Context Protocol (MCP)

    Sources: anthropic.com(opens in a new window)

  13. The Regulations on the Administration of Network Data Security took effect

    Sources: flk.npc.gov.cn(opens in a new window)

  14. The second 2024 revision of the Measures for the Administration of Internet Information Services took effect

    Sources: flk.npc.gov.cn(opens in a new window)

  15. DeepSeek-R1 was released, an open-source reasoning model that sharply lowered costs

    Sources: arxiv.org(opens in a new window)

  16. The Measures for Labeling AI-Generated and Synthetic Content took effect

    Sources: cac.gov.cn(opens in a new window)

  17. The Standing Committee of the National People's Congress passed the decision to amend the Cybersecurity Law (in effect 2026-01-01)

    Sources: flk.npc.gov.cn(opens in a new window) · npc.gov.cn(opens in a new window)

  18. The EU's Regulation (EU) 2026/1744 entered into force, amending the Artificial Intelligence Act

    Sources: eur-lex.europa.eu(opens in a new window)

On to the next era

This era is still underway. The unsolved pain points that may drive the next segmentation include: whether agents can be reliably authorized to "get things done on a person's behalf", the cost of compute and energy, the trustworthiness and copyright of AI-generated content, and on-device models and embodied intelligence (robots). The next era is not presumed for now.