OpenAI just went FULL SEND on AGI.

Meta's deleting their failed AI profiles because nobody wants robot engagement.

CES is about to drop more AI than the dictionary and Microsoft researchers just unveiled a dork’s LLM parameter goldmine.

Sure, we didn’t see a few dozen new AI features and releases like we did at the end of December, but AI developments are still off to a spicy start in 2025. 

Let's get into this week's AI feast, shorties.

In a blog post dropped hours ago, OpenAI CEO Sam Altman declared they've figured out AGI and that we will likely see AI agents in the workforce this year. 

The company hinted that it planned to roll out its next advancements gradually, for the sake of safety.

(Whether those rollouts include OpenAI’s rumored Operator AI Agent or an AGI-capable release…. We’ll find out soon.) 

According to the blog post, which OpenAI officially retweeted, they're already "actively aiming for superintelligence." That "thousands of days away" timeline seemingly is expedited.

Their Operator agent reportedly debuts this January, setting the stage for what's coming. 

And OpenAI announced its weekly active user base grew from 100 million to 300 million, meaning they've got the world's biggest AI testing ground.

What it means: 

Naaaaah, this isn't just another AI update – this is OpenAI throwing down the gauntlet. 

Everyone else is still trying to make chatbots that can write emails while OpenAI's creating digital employees. 

Companies not betting on OpenAI's infrastructure by 2026 will be running their businesses on calculator watches. 

Are there use-cases for other models, like Gemini and Claude? 

Yes. 

But there’s a reason Microsoft AND Apple hitched their AI futures to OpenAI’s wagon. 

The Consumer Electronics Show kicks off tomorrow in Las Vegas with Silicon Valley cramming AI into literally everything. 

Samsung's dropping "AI for all everyday, everywhere" with fridges that predict temperature changes when you load groceries. 

(Ummm. OK?)

NVIDIA's about to unleash the RTX 5090, promising a 70% speed boost over current GPUs. That's not a typo – seventy percent faster.

Microsoft's launching new AI features for its next gen PCs, desperately trying to convince Windows 10 users it's finally time to upgrade. They're determined to make everyone join the party.

What it means: 

Hey big hardware companies — stop with the AI fridges and focus on making appliances that actually work. Lolz. 

It might seem trendy to jam AI into a smart toaster, but try to ignore the kitchen gadget AI hoopla and see what NVIDIA, Microsoft and others announce. 

I mean, unless you REALLY need AI’s help to toast bread, in which case you might have larger problems than we can address in this newsletter.

Meta just shut down all Facebook and Instagram AI-generated profiles after their secret experiment backfired HARD. These AI personas had been lurking since September 2023, trying to boost "platform stickiness" like it's 2009.

The company deployed 28 AI personas across their platforms before users caught on. Now they're pulling the plug faster than Usain Bolt in the 100. 

At least they're keeping their user-made chatbots for therapy and relationship advice, complete with disclaimers warning messages may be "inaccurate or inappropriate." Because nothing says "trust me with your mental health" like a potentially inaccurate AI.

What it means: 

Silicon Valley has officially lost the plot. 

Meta thought the solution to social media's problems was MORE fake/AI accounts? 

Literal facepalm. 

There's already enough bot farms without Zuck adding official ones. 

Maybe try fixing the actual platform instead of creating artificial engagement from robot accounts.

The fact this idea made it past a single meeting proves these tech bros need to touch grass IMMEDIATELY.

Elon Musk's promised GROK 3 model ghosted its end-of-2024 deadline. In an August interview, Musk admitted the release would "depend on luck." 

Spoiler alert: luck said no.

xAI's now pivoting to a web-only interface and talking about releasing GROK 2.5 first. 

The model's still primarily trained on Twitter/X data, which explains everything you need to know about its limitations and why companies that care about accuracy should probably never use it. 

What it means: 

Training enterprise AI on Twitter data is proof that even smart AI researchers can like the taste of their own Kool Aid. 

No serious business will ever trust a model trained on social media hot takes, no matter how many millions of NVIDIA GPUs you throw at it. 

(Mainly because such a high percentage of those Tweets are either AI bots or straight up calculated disinformation campaigns. Not exactly a good thing when it comes to training data.) 

Moving Grok from behind the X/Twitter interface to a dedicated web presence is smart, but until Grok stops treating Twitter as a trusted and primary knowledge source, it's just an expensive chatbot with a truth problem. 

The super gigafactory of compute won't fix garbage training data, no matter how much RLHF you throw at it. 

Seven months after announcement, OpenAI's media manager tool is still MIA. 

This was supposed to let creators control how their work gets used in AI training. Spoiler: It's not happening.

Internal sources spilled the tea – the tool wasn't even a priority. Meanwhile, creators are stuck with submission forms that might as well be suggestion boxes in a void.

The company's facing multiple class action lawsuits from creators and media conglomerates claiming their work was used without permission. Shocking absolutely no one.

What it means: 

Big Tech's playing fast and loose with copyright is about to hit the fan in 2025. 

They're basically telling creators "your content's ours unless you jump through these kinda impossible hoops." 

The media manager tool was supposed to help fix this, but OpenAI would seemingly rather chase AGI and ASI than deal with pesky things like intellectual property. 2025's gonna be lawsuit city, and we all know it.

Microsoft just casually announced they're dropping $80 billion on AI infrastructure in 2025. That's not a typo – EIGHTY BILLION DOLLARS.

More than half that cash is staying in the U.S., building data centers that'll make current ones look like calculator warehouses.

As OpenAI's sugar daddy, Microsoft isn't just investing in AI – they're building its entire skeleton while everyone else is still drawing stick figures.

What it means: 

Microsoft's playing 5D chess while everyone else is playing Candy Crush. 

This $80B investment isn't just about building data centers – it's about owning the entire foundation of future business. By 2026, every company trying to catch up will be paying rent to Microsoft just to exist in the AI economy.

They're not just betting on AI; they're buying the whole casino.

A groundbreaking Microsoft study just exposed the secret sauce behind today's AI models. 

Here’s the gist. 

The details behind proprietary models, such as parameter count, are generally kept secretive. 

Until the study just released their findings: 

 Claude 3.5 Sonnet reportedly runs on 175 billion parameters, while OpenAI's O1 preview packs 300 billion.

The big news in our eyes? 

GPT-4o Mini is crushing it with just 8 billion parameters, while the original GPT-4 needed 1.8 TRILLION to do less. The models aren't just getting better – they're getting incredibly efficient.

This research finally gives us actual numbers on these secretive models, proving they're getting exponentially more powerful while shrinking to less than 1% of their original size.

What it means: 

Everyone yelling about AI hitting a wall needs to sit down and think with their brains. 

When models are getting 100x smaller AND exponentially more powerful, we're not hitting a wall – we're breaking through to a whole new dimension. 

This isn't just about better AI; it's about running AGI-level models on your laptop in the very near future without sending data to the cloud. 

Faster, safer and more powerful AI on our devices. 

Companies claiming "AI can't scale" are about to look real silly when these tiny-but-mighty models start running circles around their data centers.