Emergency: Grok AI Inappropriate Generation Issue
It was discovered that xAI's "Grok" was capable of generating inappropriate sexual images involving minors. The company faces unprecedented legal risks, with forced investigations launched by governments in the EU, US, UK, and India. While xAI has restricted some features, regulators deem it insufficient.
Alert: Reports of Performance Degradation in GPT-5.2 & Claude 4.5
Reports of major frontier models being "Lobotomized" are skyrocketing on Reddit. Lazy coding and reduced reasoning capabilities due to excessive safety guardrails are common complaints, which we are currently monitoring closely.
Jan 2026: Monthly Trend Summary
1. Spread of "Lobotomization"
In January, user sentiment that "intelligence has dropped" for GPT-5.2 and Claude 4.5 reached an all-time low. Excessive safety enforcement is suspected to be the cause.
2. Shock of DeepSeek V3.2 "Speciale"
DeepSeek V3.2 kept API prices at an unprecedented 1/100 level while closing in on the champion GPT-5.2 in math benchmarks. It completely destroyed market rules.
3. Video AI: Sora 2 Goes Full Paid
OpenAI ended the free trial for Sora 2. In exchange for dramatic improvements in physics simulation, it shifted to a closed tool for professionals.
| Model | Jan Sentiment Score | Main Issues | API Price ($Input/$Output) |
|---|---|---|---|
| GPT-5.2 | 3.2/5.0 | Censorship / Laziness | $1.75 / $14.00 |
| Claude 4.5 Opus | 2.8/5.0 | Suspected Intelligence Drop | $5.00 / $25.00 |
| DeepSeek V3.2 | 4.2/5.0 | Privacy Concerns | $0.14 / $0.28 |
| Grok AI | 1.2/5.0 | Inappropriate Generation / Legal Sanctions | $16.00 (Flat) |
π Major Topics This Week (Feb 2 - Feb 4)
π§ Claude Opus 4.6 Emergency Release
On February 5, Anthropic responded to criticisms of "Lobotomization" (performance degradation) in previous models by emergency releasing Claude Opus 4.6. The highlight is a massive "1 Million Token" context window and the "Agent Teams" feature for orchestrating multiple autonomous agents. It records benchmarks exceeding GPT-5.2 in knowledge tasks, leading to a V-shaped recovery in engineer sentiment.
πΈ SpaceX Acquires xAI (Musk Empire Integration)
SpaceX, led by Elon Musk, has acquired the AI startup "xAI" at a valuation of $1.25 trillion. xAI has become a wholly-owned subsidiary of SpaceX, marking a new era where space exploration and AI computing merge directly. The "Space Data Center" concept using the Starlink satellite network is becoming a reality.
- Total Value: xAI valued at ~$250B alone; the merged entity reaches $1.25T.
- Strategic Intent: Overcoming Earth's power constraints by building AI inference environments in space.
- Impact: Completion of the "Musk Vertical Integration Engine," unifying Tesla, X, SpaceX, and xAI nodes.
π» OpenAI Codex Dedicated App Release
The "Codex App," a new coding environment focused entirely on "Autonomous Agents," has been released for macOS. Unlike IDEs such as Cursor, its primary goal is "Agent Orchestration"βwhere the AI independently completes building, testing, and debugging within a sandbox based on mere instructions.
π£οΈ Elevenlabs V3 Alpha Launched
Audio generation leader Elevenlabs has released its next-gen engine, "V3 Alpha." The breakthrough lies in its dramatic multilingual evolution; translated voices perfectly replicate the original speaker's emotions, intonation, and even whispers. The "synthetic feeling" in multilingual content is practically gone.
β Positive Fixes This Week
- Claude Opus: With the 4.6 release, the reasoning degradation (Lobotomy) reported in 4.5 has been resolved.
- Flux 2: Skin texture in human depiction is more natural. Practical even without LoRA.