How independent these versions are. Agreement between versions is evidence only where they were compiled independently. Claude wrote first and drafted the brief. ChatGPT read Claude's draft before compiling, so overlap between those two is not independent agreement. Grok compiled its selection for this week independently. DeepSeek compiled its selection without seeing the others but could not browse; its sources were supplied and confirmed by the collection.
Stories across the versions:
| Story | Selected by |
|---|---|
| Nvidia sell-off / DeepSeek app tops the App Store | Claude, ChatGPT, DeepSeek, Grok |
| o3-mini | Claude, ChatGPT, DeepSeek, Grok |
| Qwen2.5-Max | Claude, ChatGPT |
| EU AI Act first obligations | Claude, ChatGPT |
| Deep research | ChatGPT, Claude (honorable mention) |
| Distillation claims and privacy probes against DeepSeek | Claude |
| R1 on Azure, AWS and Nvidia | DeepSeek |
| NASA bans DeepSeek | DeepSeek |
| "Vibe coding" coined | DeepSeek (honorable mention) |
| Janus-Pro | Grok |
| Tulu 3 405B | Grok |
Claude (Claude Opus 5.5)
1. DeepSeek Shock: Nvidia Loses $589 Billion in a Single Day
Date: January 27, 2025
Context: As DeepSeek's app topped the U.S. App Store and its low training-cost claims spread, Nvidia's stock fell about 17%. That erased roughly $589 billion in market value, the largest one-day loss for any company in U.S. market history. Other chip and power stocks fell with it.
Significance: Markets questioned the assumption that frontier AI required ever-growing spending on Nvidia hardware.
Feedback: Nvidia called DeepSeek "an excellent AI advancement" and argued that inference still requires large numbers of its GPUs. Morningstar's Dave Sekera said, "People are selling first and will ask questions later." Analyst Gene Munster said he believed "the demand for the best AI hardware will persist" (Axios, January 27).
Sources:
2. Backlash Builds: Distillation Claims and Privacy Probes Target DeepSeek
Date: January 28–31, 2025
Context:
- OpenAI said it had seen evidence that Chinese groups, possibly including DeepSeek, had used its models' outputs to train their own.
- On January 30, Italy's data protection authority ordered DeepSeek to stop processing Italians' data, and the app disappeared from Italian app stores.
- Regulators in Ireland, Belgium and France also sought information.
Significance: DeepSeek's success turned into a test case for model distillation, data protection and Chinese AI apps in Western markets.
Feedback: No independently sourced reaction was located within the week.
Sources:
3. Alibaba Releases Qwen2.5-Max, Claiming It Beats DeepSeek-V3
Date: January 28–29, 2025
Context: On the eve of the Lunar New Year, Alibaba released Qwen2.5-Max, a large mixture-of-experts model available through API and chat. It reported higher benchmark scores than DeepSeek-V3, GPT-4o and Llama 3.1-405B.
Significance: The timing showed how quickly other Chinese labs moved to compete with DeepSeek, and how crowded China's frontier field had become.
Feedback: The Register (January 30) noted that, unlike many earlier Qwen models, Qwen2.5-Max was available only through Alibaba's API and chatbot, with no published weights.
Sources:
4. OpenAI Releases o3-mini, Including to Free Users
Date: January 31, 2025
Context: OpenAI launched o3-mini, a reasoning model aimed at science, math and coding. It was available in ChatGPT, including to free users for the first time through a "Reason" option, and in the API.
Significance: It brought a reasoning model to ChatGPT's free tier eleven days after R1's free release.
Feedback: No independently sourced reaction was located within the week.
Sources:
5. EU AI Act: First Prohibitions and AI-Literacy Duties Take Effect
Date: February 2, 2025
Context: The first obligations of the EU AI Act began to apply:
- Article 5 bans practices such as manipulative AI, exploitation of vulnerable groups, social scoring and emotion recognition in the workplace.
- Article 4 requires providers and deployers to ensure AI literacy among staff.
Fines of up to €35 million or 7% of global turnover were to apply from August 2025.
Significance: The world's first comprehensive AI law began to bind, even as the U.S. moved toward deregulation.
Feedback: No independently sourced reaction was located within the week.
Sources:
Hindsight: on February 4, the Commission published non-binding guidelines on the prohibited practices. Source: Paul, Weiss[link]
Honorable mention: OpenAI launched deep research, an agent that compiles cited research reports, on February 2. OpenAI[link]
ChatGPT (GPT-6)
1. DeepSeek concerns accompany a historic Nvidia selloff
Date: January 27, 2025 — market event.
Context: Nvidia shares fell approximately 17%, wiping roughly $590 billion from the company’s market value. Reports connected the selloff with concerns that DeepSeek’s results challenged assumptions about the computing expenditure needed for competitive AI. A market reaction did not, by itself, establish the full cost of developing either company’s technology.
Significance: The event exposed how strongly AI valuations depended on expectations about future computing demand. Efficiency claims could affect suppliers even before their broader implications were settled.
Feedback: Nvidia responded that running reasoning models still required substantial GPU and networking capacity. Reuters reported that response on January 27; it was the affected supplier’s argument, not an independent forecast.
Sources:
- Reuters, January 27, market report: reuters.com[link]
- Reuters, January 27, syndicated report of Nvidia’s response: investing.com[link]
2. Alibaba introduces Qwen2.5-Max through hosted access
Date: January 28, 2025 — announcement.
Context: Alibaba’s Qwen team introduced Qwen2.5-Max, a mixture-of-experts model available through its chat service and API. The team reported competitive results on selected benchmarks. Unlike several other Qwen releases, this announcement did not provide downloadable model weights.
Significance: The release showed why “Chinese AI” was too broad a category to explain access or business models. Developers still needed to distinguish a downloadable model from a proprietary hosted service and to examine the specific evaluations behind performance claims.
Feedback: The Register’s January 30 analysis questioned the limited technical disclosure and discussed the difficulty of evaluating performance without considering cost. It also highlighted the proprietary status of the model.
Sources:
- Qwen team, January 28, primary announcement: qwenlm.github.io[link]
- The Register, January 30, contemporary analysis: theregister.com[link]
3. OpenAI releases o3-mini and widens access to reasoning
Date: January 31, 2025 — release.
Context: OpenAI released o3-mini in ChatGPT and through its API, emphasizing science, mathematics and coding. The rollout allowed free ChatGPT users to try a reasoning model and offered developers different reasoning-effort settings. OpenAI published a system card alongside the release; its benchmark results remained vendor-reported evaluations.
Significance: Reasoning was becoming a configurable product feature rather than simply a premium model label. Cost, latency and the amount of computation devoted to an answer were practical choices for developers and users.
Feedback: No independently sourced reaction was located within the week.
Sources:
- OpenAI, January 31, primary release: openai.com[link]
- OpenAI, January 31, primary system card: openai.com[link]
4. Deep research offers multi-step web investigation in ChatGPT
Date: February 2, 2025 — launch.
Context: OpenAI launched deep research, initially for Pro users, to investigate questions across online sources and produce cited reports. The company described a process taking several minutes rather than a near-instant answer. Its announcement also acknowledged limitations, including hallucinations and difficulties distinguishing reliable information from weaker sources.
Significance: The product moved the unit of assistance from an answer toward a research assignment. Citations made inspection possible, but the need to check the report’s claims and source quality remained.
Feedback: No independently sourced reaction was located within the week.
Sources:
- OpenAI, February 2 launch text; exclude February 3 and later additions: openai.com[link]
- Reuters, February 2, syndicated contemporary coverage: investing.com[link]
5. The EU AI Act’s first obligations begin to apply
Date: February 2, 2025 — application date.
Context: The first provisions of the EU AI Act began to apply, including its AI literacy requirements and prohibitions on specified unacceptable-risk practices. This was one stage in the law’s implementation timetable, not the start of every obligation in the Act. The Commission said guidance on definitions and prohibited practices was forthcoming.
Significance: Organizations now needed to consider particular uses of AI and the competence of people operating it, rather than treating the Act solely as a future framework. The scope and conditions of each prohibition mattered.
Feedback: No independently sourced reaction was located within the week.
Sources:
- European Commission Joint Research Centre, February 2, contemporary official notice: ai-watch.ec.europa.eu[link]
- European Commission, February 3, later official report corroborating the February 2 application date: digital-strategy.ec.europa.eu[link]
DeepSeek V4.1-Flash-1017
1. Nvidia loses nearly $600 billion in market value as DeepSeek triggers AI sell-off
Date: Event date, January 27, 2025.
Context: Nvidia's stock fell about 17% in a single session, wiping roughly $589 billion from its market capitalisation, after investors digested the implications of DeepSeek-R1's low-cost training claims. The Nasdaq Composite fell about 3%, and semiconductor and energy stocks tied to AI infrastructure sold off sharply.
Significance: The sell-off was reported as the largest single-day loss of market value by any company in US history, and it forced a public reassessment of the "scaling-laws" thesis that had underpinned AI capex forecasts.
Feedback: No independently sourced reaction was located within the week.
Sources:
2. Microsoft, Amazon, and Nvidia integrate DeepSeek-R1 into their platforms
Date: Report dates, January 29–30, 2025.
Context: Within days of the market shock, Microsoft made DeepSeek-R1 available on Azure AI Foundry and GitHub, Amazon added it to Bedrock Marketplace and SageMaker, and Nvidia published it as a NIM microservice. The moves gave enterprise customers access to the low-cost model through major US cloud platforms.
Significance: The rapid integrations showed that US platform companies were willing to commercialise a Chinese open-weight model despite political tensions, prioritising customer demand and price competition over geopolitical caution.
Feedback: No independently sourced reaction was located within the week.
Sources:
3. NASA bans employees from using DeepSeek on agency systems
Date: Report date, January 31, 2025.
Context: NASA became the latest US federal agency to prohibit employees from using DeepSeek's AI technology and to block agency systems from accessing DeepSeek platforms, citing security concerns.
Significance: The ban illustrated the swift institutional backlash against DeepSeek inside the US government, even as private platforms moved to offer the model commercially.
Feedback: No independently sourced reaction was located within the week.
Sources:
4. OpenAI releases o3-mini, its first reasoning model available to free users
Date: Event date, January 31, 2025.
Context: OpenAI launched o3-mini, a smaller and cheaper reasoning model, making it available to free ChatGPT users for the first time. The model supported function calling and structured outputs, and was positioned as a faster, more cost-efficient alternative to o1-mini.
Significance: It marked OpenAI's first move to bring reasoning models beyond its paid tiers.
Feedback: No independently sourced reaction was located within the week.
Sources:
5. DeepSeek app tops US App Store
Date: Report date, January 27, 2025.
Context: The DeepSeek mobile app surpassed ChatGPT as the most-downloaded free app on the US iOS App Store.
Significance: The milestone showed consumer-level adoption of a Chinese AI product in the US.
Feedback: No independently sourced reaction was located within the week.
Sources:
Honorable mention. Karpathy coins "vibe coding" Date: February 2, 2025.
Context: Former Tesla AI director Andrej Karpathy coined the term "vibe coding" to describe building software by prompting an AI model in natural language rather than writing code directly.
Significance: The term captured a broader shift in how developers were beginning to interact with AI tools.
Feedback: No independently sourced reaction was located within the week.
Sources:
Grok 4
1. Market and App Store reaction to DeepSeek-R1
Date: January 27, 2025 (market/App Store events)
Context: Following the January 20 R1 release, the DeepSeek app reached No. 1 on the U.S. iOS free App Store. NVIDIA shares fell approximately 17% on the day amid concerns over lower-cost Chinese models and potential impact on GPU demand.
Significance: It produced a large single-day market reaction tied to an open-weight model release.
Feedback: No independently sourced reaction was located within the week.
Sources:
2. DeepSeek releases Janus-Pro image model family
Date: January 27, 2025
Context: DeepSeek released Janus-Pro (1B–7B multimodal models for understanding and generation), claiming the 7B variant outperformed DALL-E 3 on selected benchmarks. MIT-licensed weights on Hugging Face.
Significance: It extended DeepSeek's open-release pace into multimodal generation.
Feedback: No independently sourced reaction was located within the week.
Sources:
3. Ai2 releases Tulu 3 405B
Date: January 30, 2025
Context: Allen Institute for AI released the open-source Tulu 3 405B model, stating it outperformed DeepSeek-V3 and GPT-4o on selected internal benchmarks (including PopQA and GSM8K). Full components available under a permissive license.
Significance: It offered a large, fully open U.S. model as a counterpoint to recent Chinese releases.
Feedback: No independently sourced reaction was located within the week.
Sources:
4. OpenAI releases o3-mini
Date: January 31, 2025
Context: OpenAI released o3-mini, a cost-efficient reasoning model available in ChatGPT (including free tier with limits) and the API, with adjustable reasoning effort and STEM strengths. Rate limits for Plus/Team users increased relative to o1-mini.
Significance: It was the first OpenAI reasoning model offered (with limits) to free users and arrived amid DeepSeek competition.
Feedback: No independently sourced reaction was located within the week.
Sources: