01-17-Daily AI News Daily

Today’s Summary

ChatGPT free version is getting ads, Sam Altman personally steps in to put out fires, promising no impact on answers and no selling conversation data.
GPT-5.2-Codex quietly launches, capable of autonomously completing complex software engineering tasks, and can inject new Prompts on the fly to adjust direction.
Developers rush to try the new model in Cursor—the wait-and-see crowd wins again.

⚡ Quick Navigation

💡 Tip: Want to experience the latest AI models mentioned in this article (Claude 4.5, GPT, Gemini 3 Pro) right away? No account? Head over to Aivora and grab one—ready to go in a minute, hassle-free support.

Today’s AI News

👀 One-Liner

OpenAI is about to stuff ads into ChatGPT’s free version, and Sam Altman personally showed up to explain.

🔑 3 Key Takeaways

#Ads Are Coming #Codex on Fire #Xibei Crashes and Burns

🔥 Top 10 Highlights

1. OpenAI Plans to Test Ads in ChatGPT Free Version

Used to be, the worst thing about ChatGPT free was rate limits and queue times. Now? Ads are coming. Sam Altman dropped a late-night tweet explaining: too many free users, gotta find a business model to keep the lights on. But he drew some hard lines—ads won’t mess with the AI’s answers, and your conversations won’t be visible to advertisers. Paid users (Pro, Business, Enterprise) are completely untouched. Honestly, it’s straight out of Instagram’s playbook. If the ads are actually useful, it’s not the worst trade-off. But free lunch always comes with a bill.

Image

2. GPT-5.2-Codex Programming Model API Now Open

OpenAI quietly dropped GPT-5.2-Codex, billed as the “most powerful agentic programming model.” This thing doesn’t just autocomplete code—it can autonomously tackle complex software engineering tasks, from system architecture to implementation, soup to nuts. Even wilder: it can spot security vulnerabilities in your codebase. Already integrated into Cursor, GitHub, and other mainstream dev tools. Someone on the forums updated Codex last night and found a new gpt5.2codex-xhigh option. Developers are losing their minds. The wait-and-see crowd just won big.

Image

3. Codex New Feature: Inject Prompts Without Interrupting the Model

Used to be, if the AI went off track while coding, you’d just stare helplessly or kill it and start over. Now Codex pulled off a slick move—you can inject new Prompts while the model is thinking, and it’ll adjust course in real time. Turn it on in /experimental, and your follow-up messages get injected straight into the model’s reasoning context. Real-world test: the model’s still planning, I throw in a new requirement, and boom—it immediately factors it in. This is what human-AI collaboration should look like.

Image

4. OpenAI Partners with Cerebras for Lightning-Fast Codex

Sam Altman tweeted: “Very fast Codex coming!” with a pic of the OpenAI-Cerebras partnership announcement. Cerebras makes AI chips, and their wafer-scale processors have always been beasts at inference speed. What does this partnership mean? Codex’s response time could get a massive boost. For developers who’ve been twiddling their thumbs waiting for AI to finish coding, this is huge. Details are still under wraps, but “very fast” is all the hype we need.

Image

5. ChatGPT Memory Feature Gets Major Upgrade

Ever had that moment where you chat with ChatGPT for ages, then next time it has zero memory of who you are? Not anymore. ChatGPT’s memory feature just leveled up—it can now more reliably remember details from past conversations, like your favorite recipes or workout plans. The official line: “more reliably finds and remembers details from previous conversations.” Sounds basic, but for long-term users, it means the AI finally acts like a real assistant that actually knows you.

6. Black Forest Labs Open-Sources FLUX.2 [klein] Image Model

Black Forest Labs released FLUX.2 [klein] in 4B and 9B versions. The standout: blazing-fast image generation with solid quality, plus image editing support. The 4B version is commercially usable, 9B isn’t. I tested the online demo—images pop out in seconds, quality rivals the big players. The image editing is slick too, like turning a regular photo into 3D cartoon style with natural-looking results. Only gripe: the official safety filters are paranoid as hell, blocking anything slightly sensitive.

Image

7. Google Veo 3.1 Major Upgrade: Multi-Image Reference Consistency Maxed Out

Google’s AI video generation model Veo 3.1 just dropped with three killer upgrades: multi-image reference consistency optimization, native vertical output, 4K upscaling. In plain English: feed it multiple reference images and it keeps people, objects, and textures consistent; outputs vertical video natively, no cropping needed; spits out pro-grade 4K. For short-form video creators, these features are gold. Google clearly isn’t sleeping on the AI video generation race.

8. Alipay Launches ACT Protocol, Building AI Agent Commercial Collaboration Standard

AI agents are multiplying, but cross-platform task execution is still janky. Alipay just rolled out the ACT protocol, defining four core infrastructure standards to ensure AI operations are safe and controllable. Core principle: any user financial transaction needs explicit authorization—the AI can’t just spend your money on its own. Sounds obvious, but in the Agent era, this kind of “trust infrastructure” actually matters. Alipay’s inviting multiple parties to join the ACT ecosystem, aiming to build a commercial trust network.

9. Tencent Hunyuan 3D Studio 1.2 Enters Public Beta

Tencent’s 3D modeling tool Hunyuan 3D Studio hit version 1.2. New additions: brush interaction and eight-view generation capability. Brush interaction lets you fine-tune 3D component decomposition and details; eight-view input massively improves structural accuracy in generated models. The rendering pipeline got a rebuild too—texture color fidelity and geometric detail both got better. For 3D content creators, this tool is looking more and more like professional software.

Image

10. Cursor’s Lee Robinson Explains Core Concepts of AI Programming Tools

Rules, commands, MCP servers, subagents, modes, hooks, tools—AI programming tools have so many concepts your head spins. Cursor’s Lee Robinson dropped a video breaking it all down into two core ideas: static context (rules) and dynamic context (skills). Rules load every conversation; Skills load on demand. Get these two concepts down, and everything else is just variations. If you want to actually understand AI programming tools, this video is a must-watch.

Image

📌 Worth Watching

😄 AI Fun Stuff

Xibei vs. Luo Yonghao Showdown, Ending Had Everyone Laughing

Today’s wildest story isn’t AI news—it’s Luo Yonghao absolutely demolishing Xibei again. Best comment from the internet: “Like elementary schoolers about to fight but the teacher catches them and makes them stand in the corner.” 😂 People literally set alarms for 22:30 just to catch the latest summary updates. Xibei’s PR manager apparently quit… This drama beats any AI-generated script.

Image

🔮 AI Trend Predictions

ChatGPT Ad Feature Goes Live

  • Predicted Timeline: Q1 2025 (Feb-Mar)
  • Confidence: 75%
  • Reasoning: Today’s news OpenAI Plans to Test Ads in ChatGPT Free Version + Sam Altman explicitly said “testing starts in coming weeks”—given OpenAI’s pace, full rollout isn’t far off

AI Programming Tools Enter “Skills Ecosystem” Era

Codex Inference Speed Gets Major Boost

  • Predicted Timeline: Q1 2025
  • Confidence: 70%
  • Reasoning: Today’s news OpenAI Partners with Cerebras + Cerebras’s wafer-scale chips have clear inference speed advantages, partnership results should ship soon

❓ Related Questions

How do I experience ChatGPT’s latest memory feature and GPT-5.2-Codex?

ChatGPT’s memory upgrade and GPT-5.2-Codex currently require ChatGPT Plus or higher-tier subscriptions for full access. Mainland China users may face payment difficulties or account registration restrictions.

Solution: Visit Aivora to get ready-made accounts—instant delivery, worry-free support.

Last updated on
01-18-Daily01-16-Daily