12-31-Daily AI News Daily

2025/12/31 23:34:00

It’s the last day of 2025.

Feeling a bit dazed. This year felt like a rollercoaster, or like I was constantly being pushed forward. Not really aiming for a serious year-end summary, just wanna ramble a bit and sort out this mess in my head.

Back in January, I was really hoping to use AI to slack off.

I had just installed Cline then, envisioning AI writing all my code while I just sipped coffee. But when I actually tried it? Nah. Big nope. Simple logic was okay, but try to say a bit more, and the context just exploded, rendering it useless. I was using DeepSeek Coder back then, later switched to the Chat version, same deal – completely unusable.

I was thinking back then, this thing is light-years away from ’not even needing programmers,’ right? 🤔

Since coding was a bust, I dove into image generation. I was cycling through models like SD1.5, SDXL, PONY, and FLUX. Back then, my PC case would get hot enough to fry an egg just running those models. 🥵

The cruel twist? My hard drive crashed, and the CPU burned out. All my data? Gone. 💀

Staring at that black screen, I was completely numb. 🤯

No choice but to buy a new hard drive, re-download models, and set up the environment all over again. 🙄

Later, I felt like just having images wasn’t cutting it, so I moved on to video and audio. I used Index-TTS for voiceovers and Wan2.0 to create dancing girl videos. 💃🎶

I took all this stuff to Douyin (TikTok) and posted videos, hoping to go viral. After a few months, not much traction. I just don’t have that knack for operations, so that whole venture fizzled out. 📉

Then April rolled around, and GPT-4o launched its native image generation feature.

I saw that and completely lost my cool. 🤯 The Ghibli-style images I’d painstakingly tweaked parameters for ages to get? It whipped them up with a single prompt! Aside from NSFW stuff (they said the adult mode for year-end was also pushed to Q1 next year), it was totally on par with FLUX.

This hit me pretty hard. 🥊

It suddenly dawned on me: the tech I’d busted my butt learning at the start of the year was obsolete in just three months. 😱

That speed is terrifying. If I don’t keep up with the latest news, I might still be happily patting myself on the back tomorrow, using tools that are already obsolete. 😬

To cure this anxiety, I needed a remedy. 💊

I ended up writing CloudFlare-AI-Insight-Daily. The idea was simple: use CloudFlare to grab FOLO’s subscription feeds daily, then toss them to AI to compile into a daily report.

This time, I played it smart. I didn’t use those early-year models; instead, I leveraged Kilo plus Gemini-2.5-flash to write the code. 💡

1M context? That was pure bliss! ✨

The headache of insufficient context, which plagued me at the beginning of the year, was completely gone. The AI basically wrote all the code; I just had to glance over it and patch things up.

Once the first version was done, I casually posted it on L-site and Ruan Yi-feng’s Weekly. 🚀

I hadn’t really expected much, but somehow, it just blew up! 🔥 It got over 100 Stars initially, and now it’s shockingly over 1000.

The most absurd thing? The subscriptions on FOLO. My Douyin videos got no views, but this auto-generated daily report somehow snagged 18k subscribers and has been topping the charts ever since! 🤯

Sometimes, that’s just how it goes: plant a willow without intent, and it grows into shade. 🌳

This project gave me a bit of confidence and brought in some traffic. 💪

Then Google released Gemini CLI, which caused the Flash model’s quota to plummet, making it unstable. 📉

I didn’t want to pay, nor did I want to put up with the hassle. 🙅‍♂️

I remembered Kilo also supported Gemini CLI, so I just decided to reverse-engineer the Gemini CLI altogether. 💡

Thus, AIClient-2-API was born.

This project reverse-engineered the CLI into an API interface, supporting both OpenAI and Gemini protocols. After I released it, it again landed on Ruan Yi-feng’s Weekly! 🥳

Later, I casually reverse-engineered Kiro’s Claude, Qwen Code, and Antigravity, and even added multi-protocol conversion support. Now, I have so many tokens I can’t even use them all! 🤑

With all these tokens, I got ambitious and wanted to build some cool stuff. ✨

I whipped up several sites in one go: podcast.hubtoday.app , prompt.hubtoday.app , plus AI Portrait Master, and a Legacy Management System. 🛠️

While building prompt.hubtoday.app , I launched a ‘Build in Public’ challenge: ‘Venmo me 50 and watch the entire AI development process from scratch.’ 💸

The plan was to finish the web, backend, and desktop versions in a month. But in less than 3 weeks, all the first-phase features were done! 🚀

Phase two hasn’t even started yet, and just the other day, folks in the group were bugging me for updates… Guess I’ll just pretend I didn’t see that. 🙈

Lately, I’ve been working on an overseas application template based on Next16. The basic features are pretty much written, and the online testing environment is all set up. 🌍

To make overseas payments easier in the future, I even made a special trip to Hong Kong recently and opened four bank accounts. 🇭🇰💰

Looking back at this year, I was truly propelled by technology. ⚡

At the start of the year, I was struggling with insufficient context, but by year-end, the current setups are like gods battling it out! 🤯

For coding, we’ve got Claude Code, Codex, Trae. 💻

For text, there’s Gemini 3.0, and a solid alternative, Doubao. ✍️

For image generation, there’s the ridiculously powerful Nano Banana Pro, and for open-source, Flux. 🎨

For video, we have Veo3.1, Sora2, and open-source Wan2.6. 🎬

For audio, there’s MiniMax, and open-source Index-TTS2. 🎧

All amazing stuff. The slightly less impressive ones? Nobody even mentions them anymore. ✨

But now, I’m actually less anxious than I was at the start of the year. 😌

Sure, there are always industry ’toxic elements’ hyping things up, peddling anxiety, and claiming humanity is doomed. 🙄

My current take? AI is strong, for sure, but it’s still a ways off from completely replacing humans. It can help me write code, but it doesn’t know what code I need to write, nor why I need to write it.

AI is a great helper, but it’s no excuse for me to stop working. 💪

This year, I got ’tech-stabbed’ multiple times – learning fast, forgetting fast, and seeing things become obsolete even faster. 🔪

But this sense of urgency forces me to constantly learn and experiment. 🔥

It’s been quite fulfilling. 😊

Alright, that’s it for now. 2025, Happy New Year, 2026! 🎉

Last updated on