---
id: HIST-011
title: "The Contemporary AI Era, 2022 onward (living record)"
alternative_names:
  - "Recent AI Developments Log"
  - "Post-ChatGPT Era"
category: "History → Contemporary Era"
volume: II
chapter: 14
entry_type: "Historical Record"
level: 6
status: "Draft"
last_reviewed: 2026-10-15
version: "0.1.0"
prerequisites:
  - "HIST-010"
related_entries:
  - "TIME-001"
  - "MODEL-001"
  - "FACT-001"
  - "FCST-REG-001"
  - "GOV-LAW-001"
  - "CAP-BENCH-001"
---

# The Contemporary AI Era, 2022 onward (living record)

## 1. Concise Definition
This is a **living record**: a dated, sourced log of significant events in AI since the public release of ChatGPT on 2022-11-30, kept separate from stable explanatory articles. Every entry carries a date and a source, and the **evidence status** of each entry states whether it is a documented event, a developer self-report, or a result evaluated by someone independent of the developer.

## 2. Overview & Significance
The period after 2022 is the one most affected by fast-moving claims, marketing and rapidly changing policy. A stable article cannot track it, but readers still need a dependable chronology. This record therefore follows five rules.

1. **Dated entries only.** Each entry states the date of the event (or of the source) and the date the source was accessed.
2. **Each entry is sourced** to a primary document (paper, regulation, official announcement) wherever possible.
3. **No breakthrough claims without independent evaluation.** A developer's statement that a system achieved something is recorded as a *developer claim*. It is described as a breakthrough only if an independent party has evaluated it, and the evaluator is named.
4. **Fast-moving numbers are not tabulated here.** Parameter counts, prices, benchmark scores and market sizes go in the Living Reference entries: [[MODEL-001] Model Registry](../reference/factbook/MODEL-001_model_registry.md), [[FACT-001] Factbook](../reference/factbook/FACT-001_factbook_historical_dates_and_milestones.md), [[TIME-001] AI/SI Timeline](../reference/timeline/TIME-001_ai_si_timeline.md) and [[FCST-REG-001] Forecast Register](../reference/factbook/FCST-REG-001_forecast_register.md).
5. **Undisclosed means undisclosed.** Where a developer has not published a detail, the record says so and does not infer it.

The earlier background is in [[HIST-010] The Transformer and Foundation Model Era, 2017-2022](./HIST-010_the_transformer_and_foundation_model_era_2017_2022.md).

## 3. Historical Development
This section is the dated log. Evidence status uses these labels: **Documented** (the event itself is established by an official or primary record), **Developer claim** (a result reported by the developer, not independently evaluated), **Third-party measurement** (a result produced by a party independent of the developer, with limits stated).

### 3.1 Dated entry log

| Date | Event | Evidence status | Source |
| :--- | :--- | :--- | :--- |
| 2022-11-30 | OpenAI introduces ChatGPT, a conversational model described as a sibling to InstructGPT, as a free research preview. | Documented | OpenAI (2022) |
| 2023-02-27 | Meta researchers release the LLaMA family (7B to 65B parameters), stating that models were trained on publicly available data and released to the research community; the paper states that LLaMA-13B outperforms GPT-3 (175B) on most benchmarks. | Documented (release); Developer claim (benchmark comparison) | Touvron et al. (2023) |
| 2023-03-15 | OpenAI publishes the GPT-4 technical report. The report states it contains no further details about architecture (including model size), hardware, training compute, dataset construction or training method. | Documented (publication and the nondisclosure); Developer claim (performance statements such as exam results) | OpenAI (2023) |
| 2023-11-01 | The Bletchley Declaration is published following the AI Safety Summit (1-2 November 2023), stating signatories' concern about risks at the "frontier" of AI and resolving to cooperate. | Documented | UK Government (2023) |
| 2024-08-01 | The EU Artificial Intelligence Act, Regulation (EU) 2024/1689, enters into force. | Documented | European Commission (2026) |
| 2024-10 | The Nobel Prize in Physics 2024 is awarded to John J. Hopfield and Geoffrey Hinton, and the Nobel Prize in Chemistry 2024 to David Baker (computational protein design) and Demis Hassabis and John Jumper (protein structure prediction). | Documented | Nobel Prize Outreach (2024a, 2024b) |
| 2025-01-20 | US Executive Order 14110 (30 October 2023) is revoked by Executive Order 14148, "Initial Rescissions of Harmful Executive Orders and Actions". | Documented | Federal Register (2025); Federal Register (2023) |
| 2025-01-22 | DeepSeek-AI posts the DeepSeek-R1 preprint, reporting that reasoning abilities can be incentivised through reinforcement learning. A peer-reviewed version appears in *Nature* on 2025-09-17. | Documented (publication); Developer claim (performance) | DeepSeek-AI (2025a, 2025b) |
| 2025-01-29 | The first International AI Safety Report, written by an independent expert panel of 100 contributors under a chair, is published. It was mandated by the countries attending the Bletchley summit. | Documented | Bengio et al. (2025) |
| 2025-02-02 / 2025-08-02 | Under the EU AI Act, the prohibitions and AI-literacy obligations apply from 2 February 2025, and governance rules and obligations for general-purpose AI models from 2 August 2025. | Documented | European Commission (2026) |
| 2025-03-18 | METR researchers propose a "50%-task-completion time horizon" metric and report that, on their task set, a frontier model then current (Claude 3.7 Sonnet) had a time horizon of around 50 minutes, and that the measured horizon had been doubling about every seven months since 2019. The authors discuss limits to external validity. | Third-party measurement (preliminary; extrapolation is not established) | Kwa et al. (2025) |
| 2025-07-21 | Google DeepMind announces that an advanced version of Gemini with Deep Think achieved 35 of 42 points on the 2025 International Mathematical Olympiad (IMO), graded and certified by IMO coordinators. The IMO's review did not extend to validating the AI system or its underlying model. | Developer claim, with grading certified by the IMO | Google DeepMind (2025) |
| 2025-07 | OpenAI separately reports a gold-level score on the same IMO problems. Its grading was not performed under the IMO's official process, and the claim was disputed in press coverage. | Developer claim (not independently certified) | Euronews (2025) |
| 2025-11-19 | The European Commission adopts the Digital Omnibus proposal on AI. | Documented | European Commission (2026) |
| 2026-05-07 | Political agreement is reached on the Digital Omnibus on AI. | Documented | European Commission (2026) |
| 2026-06-16 / 2026-06-29 | The European Parliament adopts the agreement on 16 June 2026, and the Council gives final approval on 29 June 2026. | Documented (secondary legal reporting) | Hunton Andrews Kurth (2026); Sidley Austin (2026) |
| 2026-07-27 | Regulation (EU) 2026/1744 (the Digital Omnibus on AI) enters into force, amending the AI Act. New application dates reported: 2 December 2027 for stand-alone high-risk systems (Annex III), 2 August 2028 for high-risk systems embedded in regulated products (Annex I), and 2 December 2026 for a watermarking change and a new prohibition on AI systems generating non-consensual intimate or child sexual abuse material. | Documented | European Commission (2026); Hunton Andrews Kurth (2026) |

### 3.2 Entries pending independent evaluation
The following developer claims are recorded but **not** treated as established breakthroughs in this record.
- LLaMA benchmark superiority over GPT-3 (Touvron et al., 2023): developer-run comparison.
- GPT-4 exam and benchmark performance (OpenAI, 2023): developer-run; the report withholds methods.
- DeepSeek-R1 benchmark results (DeepSeek-AI, 2025a, 2025b): peer-reviewed by *Nature*, but the performance results were produced by the developer.
- IMO 2025 performance by AI systems: DeepMind's grading was certified by the IMO but not the system; OpenAI's grading was not under the official process.

## 4. Technical / Conceptual Explanation
Interpretive context for the log, not a technical tutorial.

- **Instruction-following and RLHF** (from 2022): the chat interface of ChatGPT built on instruction tuning and reinforcement learning from human feedback (OpenAI, 2022); see [[TECH-ALIGN-001] Pretraining, Fine-Tuning and RLHF](../vol03_technical/TECH-ALIGN-001_pretraining_fine_tuning_and_rlhf.md).
- **Open-weight releases**: LLaMA was released to the research community (Touvron et al., 2023). Later open-weight practice differs by developer and licence and is tracked in the Model Registry.
- **Reinforcement learning for reasoning**: DeepSeek-AI (2025a) reports that reasoning behaviours can be incentivised with reinforcement learning on verifiable tasks; see [[TECH-REASON-001] Reasoning Models and Test-Time Compute](../vol03_technical/TECH-REASON-001_reasoning_models_and_test_time_compute.md).
- **Measurement of autonomy**: the time-horizon metric of Kwa et al. (2025) expresses model capability as the length of human-timed tasks completed at 50% success. It is a measurement on a defined software-task suite; the authors themselves state limits on external validity.

## 5. Types, Components & Classifications
Entries in the log fall into five classes: **model and system releases**, **research publications**, **governance and law**, **recognitions** (for example prizes), and **third-party evaluations**. Governance entries are the most prone to going stale and are checked against official sources at each review.

## 6. Applications & Concrete Examples
Deployment and adoption figures are not recorded here; they change quickly and need sources with dates and units, which belong in [[FACT-001] Factbook](../reference/factbook/FACT-001_factbook_historical_dates_and_milestones.md). Application-area articles include [[CAP-GENAI-001] Generative AI](../vol04_capabilities/CAP-GENAI-001_generative_ai.md) and [[APP-SCI-001] AI in Science](../vol07_applications/APP-SCI-001_ai_in_science.md).

## 7. Advantages, Limitations & Unresolved Questions
**Limitations of this record.**
- Coverage is selective: it records events with primary sources that were retrieved for this draft, not all significant events. Omission is not a judgement of importance.
- Several legal entries rely on law-firm summaries and the European Commission policy page rather than the Official Journal text. [VERIFICATION NEEDED: confirm the Official Journal publication date (reported as 2026-07-24) and the application dates for Regulation (EU) 2026/1744 against the text on EUR-Lex.]
- One law-firm source reported a procedural disagreement in the Council in late June 2026; later reporting and the Commission page confirm adoption and entry into force on 27 July 2026, and the disputed item is not relied on.
- The DeepSeek-R1 *Nature* publication date (2025-09-17) is taken from Crossref metadata; training-cost figures reported in the press were not verified against the paper and are not included.

**Unresolved.** Whether time-horizon trends generalise beyond software tasks, and what the IMO results imply for general mathematical ability, are open and are not settled by the sources above.

## 8. Current State of Knowledge
As of 2026-10-15 the sourced entries above are the record. The status of governance instruments is as follows: EU Regulation 2024/1689 is in force, with the dates amended by Regulation (EU) 2026/1744 (European Commission, 2026; Hunton Andrews Kurth, 2026); US Executive Order 14110 was revoked on 2025-01-20 (Federal Register, 2025). Capability claims should be read using the evidence-status column. For model-by-model detail see [[MODEL-001] Model Registry](../reference/factbook/MODEL-001_model_registry.md).

## 9. Future Trajectory & Strategic Outlook
This record contains no forecasts. Forecasts live in [[FCST-REG-001] Forecast Register](../reference/factbook/FCST-REG-001_forecast_register.md) with the structured record format of the evidence framework. **Update protocol:** entries are appended, never silently rewritten; corrections are recorded in the revision history; legal and policy entries are rechecked against official sources at each review; an entry's evidence status is upgraded only when a named independent evaluation exists.

## 10. Related Entries & Knowledge Graph
- **Foundational Concepts**:
  - [[HIST-010] The Transformer and Foundation Model Era, 2017-2022](./HIST-010_the_transformer_and_foundation_model_era_2017_2022.md): the preceding period.
  - [[TECH-LLM-001] Large Language Models](../vol03_technical/TECH-LLM-001_large_language_models.md): technical background.
- **Related Methods & Technologies**:
  - [[TECH-REASON-001] Reasoning Models and Test-Time Compute](../vol03_technical/TECH-REASON-001_reasoning_models_and_test_time_compute.md): reasoning models.
  - [[CAP-BENCH-001] AI Evaluation](../vol04_capabilities/CAP-BENCH-001_ai_evaluation.md): evaluation methodology.
- **Advanced Implications & Future Trajectories**:
  - [[GOV-LAW-001] AI Governance](../vol08_safety_governance/GOV-LAW-001_ai_governance.md): law and policy.
  - [[SAFE-ALIGN-001] AI Alignment](../vol08_safety_governance/SAFE-ALIGN-001_ai_alignment.md): safety context.
  - [[TIME-001] AI/SI Timeline](../reference/timeline/TIME-001_ai_si_timeline.md): the full timeline.

## 11. References & Foundational Literature
1. OpenAI. (2022, November 30). *ChatGPT: Optimizing language models for dialogue*. https://openai.com/blog/chatgpt (archived copy: https://web.archive.org/web/20221130/https://openai.com/blog/chatgpt). [Announcement of the ChatGPT research preview.]
2. Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.-A., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., Rodriguez, A., Joulin, A., Grave, E., & Lample, G. (2023). LLaMA: Open and efficient foundation language models. arXiv:2302.13971. [LLaMA release; claims about public-data training and benchmark comparison.]
3. OpenAI. (2023). GPT-4 technical report. arXiv:2303.08774. [GPT-4 report; states that architecture, model size, hardware, training compute, dataset construction and training method are not disclosed.]
4. UK Government. (2023, November 1). *The Bletchley Declaration by countries attending the AI Safety Summit, 1-2 November 2023*. https://www.gov.uk/government/publications/ai-safety-summit-2023-the-bletchley-declaration/the-bletchley-declaration-by-countries-attending-the-ai-safety-summit-1-2-november-2023. [Text of the declaration.]
5. European Commission. (2026). *AI Act: Regulatory framework for AI* (policy page). https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai. [Entry into force of Regulation (EU) 2024/1689 on 1 August 2024; application dates; Digital Omnibus milestones. Accessed 2026-10-15.]
6. Nobel Prize Outreach. (2024a). *The Nobel Prize in Physics 2024*. https://www.nobelprize.org/prizes/physics/2024/summary/. [Prize citation to Hopfield and Hinton.]
7. Nobel Prize Outreach. (2024b). *The Nobel Prize in Chemistry 2024*. https://www.nobelprize.org/prizes/chemistry/2024/summary/. [Prize citation to Baker, Hassabis and Jumper.]
8. Federal Register. (2025). *Executive Order 14148 of January 20, 2025: Initial rescissions of harmful executive orders and actions*. 90 FR 8237. https://www.federalregister.gov/. [Revokes Executive Order 14110 among other orders; Federal Register metadata retrieved via API on 2026-10-15; the list of revoked orders was taken from secondary summaries.] [VERIFICATION NEEDED: confirm that EO 14110 appears in the text of EO 14148's list of rescinded orders.]
9. Federal Register. (2023). *Executive Order 14110 of October 30, 2023: Safe, secure, and trustworthy development and use of artificial intelligence*. 88 FR 75191. https://www.federalregister.gov/. [The revoked order.]
10. DeepSeek-AI. (2025a). DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning. arXiv:2501.12948. [The preprint (2025-01-22).]
11. DeepSeek-AI. (2025b). DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning. *Nature*, 645(8081), 633-638. https://doi.org/10.1038/s41586-025-09422-z. [Peer-reviewed version; authors listed as DeepSeek-AI collaboration (Guo, D., et al., per Crossref).]
12. Bengio, Y., Mindermann, S., Privitera, D., et al. (2025). International AI Safety Report. arXiv:2501.17805. [Independent expert synthesis mandated by the Bletchley summit countries.]
13. Kwa, T., West, B., Becker, J., et al. (2025). Measuring AI ability to complete long software tasks. arXiv:2503.14499. [Time-horizon metric and measured trend; the title of the current arXiv version.]
14. Google DeepMind. (2025, July 21). *Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad*. https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/. [Developer announcement; 35/42, graded by IMO coordinators.]
15. Euronews. (2025, July 22). *Did Google DeepMind or OpenAI win gold at the world's most prestigious maths competition?* https://www.euronews.com/next/2025/07/22/did-google-deepmind-or-openai-win-gold-at-the-worlds-most-prestigious-math-competition. [Tier 4 press coverage of the OpenAI result and the dispute over its grading.]
16. Hunton Andrews Kurth. (2026). *EU Digital Omnibus on AI enters into force*. https://www.hunton.com/privacy-and-cybersecurity-law-blog/eu-digital-omnibus-on-ai-enters-into-force. [Law-firm summary: Regulation (EU) 2026/1744, entry into force 27 July 2026, new dates; accessed 2026-10-15.]
17. Sidley Austin. (2026, June 22). *EU lawmakers reach provisional agreement to delay key EU AI Act obligations*. https://datamatters.sidley.com/2026/06/22/eu-lawmakers-reach-provisional-agreement-to-delay-key-eu-ai-act-obligations/. [Law-firm summary of the 7 May 2026 provisional agreement and 16 June 2026 Parliament vote.]

## 12. Evidential Classification Matrix
| Section / Claim | Evidence Category | Supporting Source | Confidence / Status |
| :--- | :--- | :--- | :--- |
| ChatGPT introduced 2022-11-30 as a research preview | Established | OpenAI (2022) | High |
| GPT-4 report withholds architecture and training details | Established | OpenAI (2023) | High |
| LLaMA outperforms GPT-3 on most benchmarks | Preliminary (developer-run) | Touvron et al. (2023) | Moderate |
| Bletchley Declaration published 2023-11-01 | Established | UK Government (2023) | High |
| EU AI Act in force 2024-08-01; Omnibus in force 2026-07-27 | Established (official policy page); details from secondary sources | European Commission (2026); Hunton Andrews Kurth (2026); Sidley Austin (2026) | High for dates on the Commission page; moderate for other details |
| EO 14110 revoked 2025-01-20 | Established | Federal Register (2025); Federal Register (2023) | High |
| 2024 Nobel Prizes in Physics and Chemistry | Established | Nobel Prize Outreach (2024a, 2024b) | High |
| DeepSeek-R1 reasoning performance | Preliminary (developer-run, peer-reviewed publication) | DeepSeek-AI (2025a, 2025b) | Moderate |
| Frontier-model time horizon doubling about every seven months | Preliminary (single third-party study; limited external validity) | Kwa et al. (2025) | Moderate |
| IMO 2025 gold-level performance by AI | Preliminary (developer claims; one grading certified by IMO) | Google DeepMind (2025); Euronews (2025) | Moderate |
| Independent expert synthesis of capabilities and risks | Well Supported (as a synthesis) | Bengio et al. (2025) | Moderate to high |

## 13. Revision History
- **v0.1.0** (2026-10-15): Initial draft. Citations checked against arXiv/Crossref/publisher pages where marked; not independently fact-checked.