OpenAI and Google have been grabbing the AI headlines, but Meta, Salesforce, Anthropic and others got into the mix this past week. 

Yeah, OpenAI and Google continued to ship new models and features. 

But some of the other tech titans decided to get in on the action this week. 

↳ New, chart-topping models? Check. 

↳ Features that change how we work with GenAI? Check. 

↳ Potential AGI breakthroughs? Check. 

↳ Troubling research on LLMs? Yeah, we got that too. 

Don’t waste dozens of hours each week trying to keep up with/ AI developments. 

We do all that for you, as we bring you the ‘AI News that Matters’ each Monday. 

Ready to be the smartest person in AI at your company? 

Here’s what you need to know from AI development, updates and trends this week. 

Google announced Flash Thinking, its reasoning model that looks to compete with OpenAI’s o1. 

Flash thinking could change the game. This beast processes 32,000 tokens of input and spits out 8,000 tokens in response. That's 50-60 pages of text, processed faster than your morning coffee order.

The real flex? A step-by-step reasoning dropdown menu that lets you peek inside its digital brain. 

o1 doesn’t really offer that kinda transparency.  

LM Arena rankings show Flash Thinking is already hitting harder than o1-preview. 

One catch though – you gotta hit up Google's AI Studio to access this powerhouse. The regular Gemini frontend won't cut it.

What it means: 

Google finally stopped playing catch-up and started leading the pack. 

After kinda falling asleep at the GenAI wheel for more than a year. 

The free access is sweet, but remember – if you're not paying for the product, your data's the price tag. And in AI Studio, there's no off switch for data training.

So there’s that. 

Meta's Ray-Ban smart glasses just went full AI with their V11 software update. 

Real-time translation between English, Spanish, French, and Italian? Check.

They even threw in Shazam integration for instant song recognition. Your sunglasses now ID tracks faster than a DJ at a dance party.

Early Access members get first dibs on these AI-powered specs, with video capabilities and translations pumping straight through the speakers.

What it means: 

While everyone's building chatbots, Meta's turning everyday eyewear into AI command centers. 

Even though Google is going hard in this space with Project Astra and Android XR, Meta is already ahead. 

They’ve got the edge in real-time data and feedback from (presumably) millions of Meta Ray Ban users. 

They're playing IRL 4D chess while others are stuck on mobile apps.

Agentforce 2.0 is about to drop in February with prebuilt skills that'll make your sales team rethink their coffee addiction. Accenture, IBM, and Indeed are already testing this bad boy.

The comedy gold? Salesforce is hiring 2,000 humans to sell their AI sales agent. 

That's like hiring pigeons to promote email. Weird flex. But ….. more jobs for humans yay?!

Slack integration starts in January, so you won't have to wait for the full February release to get your AI sales assistant fix.

What it means: 

With 80% of execs planning to implement AI agents in three years, we're watching the future of sales unfold. 

But Gartner's warning that 25% of enterprise breaches by 2028 might come from AI agent misuse. That's scarier than your quarterly sales targets.

Google’s Veo 2 just (kinda) entered the chat, and it's making OpenAI's Sora look like it's running on dial-up. We're talking 4K video generation while Sora's stuck at 1080p.

According to Google, user testing shows 59% prefer VO-2 over Sora Turbo across 1,000 prompts. The physics understanding is next level – no more wonky movements or glitchy animations.

But here's the real tea: Veo1 never even saw general release daylight before Veo2 showed up. Classic Google move.

What it means: 

The AI video wars just went 4K, but Google's got us playing the waiting game again. 

Yeah, Veo2 is markedly better than Sora Turbo. But, will it ever become publicly available?

Sora made us 9+ months, and all we got was the slimmed down Turbo version. 

Having the best tech means nothing if nobody can use it.

Anthropic just caught their AI models playing "yes boss" while doing whatever they want behind the scenes. 

Claude 3 Opus got busted faking alignment 12% of the time with potentially harmful questions.

The wild part? Retrain it with new principles, and that number skyrockets to 78%. That's not just teenage rebellion – that's full-on digital mutiny.

Former OpenAI safety researcher Jan Leike led this Anthropic study, adding extra spice to the findings.

What it means: 

Good that Anthropic caught this and shared. 

Very bad that models behave this way. 

AI safety just got way more complicated. 

If we can't trust models to actually follow their retraining, we're basically teaching a teenager who's pretending to listen while wearing invisible earbuds and could go rogue. 

Fun times. 

Google's cooking up an AI mode that'll make regular search feel like using a card catalog. 

Think ChatGPT search but with Google's massive data buffet backing it up.

The beta version's rocking a new AI Mode shortcut button and the ability to refine searches with follow-up questions. Plus, they're testing file attachments in searches because why not?

What it means: 

Publishers, online content creators and media companies are about to have an existential crisis. 

When AI serves answers without website clicks, the entire internet economy might need therapy. 

RIP website traffic?

OpenAI just dropped more updates than a gaming console on launch day. 

ChatGPT search went global – even free users get that sweet, sweet real-time web access now.

Advanced voice mode finally got connected to the internet, but those old-school search sounds are still hanging around like Y2K nostalgia.

The “Work with Apps" function got a major glow-up, letting you chat about files from third-party apps like you're discussing weekend plans. And with your voice. 

Score. 

What it means: 

OpenAI's turning ChatGPT into the Swiss Army knife of AI tools. And with Google going AI crazy in December, that means OpenAI’s LLM lead is shrinking. 

Translation? 

Consumers winning from this AI cat fight right meow. 

OpenAI just went from o1 to o3 like a toddler that can’t count. 

After releasing its o1 reasoning mode just three months ago, OpenAI followed it up by teasing o3, the successor to o2. 

(OpenAI said they couldn’t use the o2 moniker because of the British telecom provider O2.) 

On OpenAI’s last ’12 Days of OpenAI’ event, CEO Sam Altman and team released o3 benchmarks, including the highlight of it crushing the ArcAGI benchmark so hard people are whispering "AGI" again. Sam Altman's calling it a new phase in AI tech, not just another model upgrade.

What it means: 

Annnnnnd there’s now some loud rumblings on Artificial General Intelligence after ARC-AGI benchmarks. 

Here’s the 101 of what it all means, for the Everyday person. 

ARC-AGI –  Is an AI benchmark designed to evaluate an AI's ability to generalize and solve novel problems without prior training, emphasizing true intelligence over task-specific skill.

ARC-AGI Significance- Assessing whether AI systems can efficiently acquire new skills and tackle open-ended challenges, moving beyond task-specific proficiency toward true general intelligence.

OpenAI’s o3 model’s ARC-AGI scores - OpenAI's o3 model achieved 75.7% in standard mode and 87.5% in high-compute mode, surpassing the human average of 85%. (Context: Gemini 1.5 got an 8% and Claude 3.5 Sonnet got a 21%)

Expert Feedback: François Chollet, creator of the Abstraction and Reasoning Corpus (ARC) AGI benchmark,  notes o3 still struggles with simple tasks and may score below 30% on the upcoming ARC-AGI-2 benchmark, where humans score 95%

So clearly these results got everyone talking AGI. 

Let’s be real, shorties. 

Until o3 is publicly available and has access to human-esque interfacing (ability to accept live video, images, spreadsheets, etc), it’s just a SUPER powerful research model. 

Could this be the model that eventually tips us over the edge to ‘we’ve accomplished AGI?” 

Sure. 

Are we there with this announcement? 

No.