Claude 3.7 and GPT-4.5 dropped days apart. 

Apple delayed smart Siri until 2027. 

(Yeah, we might get AGI before a smart Siri.) 

Don’t worry, though, Apple DID announced a $500B US investment including a Texas AI server factory.

AI news don’t stop, and neither do we. 

Let's get into this week's AI feast, shorties. Here’s what ya need to know. 👇

Anthropic just changed the hybrid game with Claude 3.7 Sonnet.

This is the first hybrid LLM ever created. It merges traditional transformer capabilities with advanced reasoning in one seamless package.

"Extended thinking mode" stands as the killer feature that Anthropic hopes will change the game, although it's currently restricted to paid users only.

This premium feature unleashes Claude's deep problem-solving abilities while showing its complete chain of thought for transparency.

Free users? They still get access to the new 3.7 model but without the reasoning toggle. Not bad.

Claude 3.7 absolutely crushes coding benchmarks. It scored an impressive 70.3 on SWE-Bench, easily smoking competitors like OpenAI's o1 and o3 Mini models.

Anthropic didn't stop with just a model upgrade. They simultaneously launched Claude Code, a terminal tool letting developers update entire codebases directly from command line.

Their API pricing strategy looks increasingly brilliant. Keeping costs steady at $3 input/$15 output per million tokens seems downright charitable compared to OpenAI's eye-watering new prices for its GPT-4.5 model. (More on that below.) 

What it means: This hybrid architecture could be AI's iPhone moment.

Every model after this will likely iterate on the same concept. Some developers are temporarily rolling back to 3.5 for specific tasks, but that doesn't change the bigger picture. This architectural shift fundamentally transforms everything about how AI works. Every major player will follow this hybrid path within months.

Your prompt strategies and AI integrations? Better start rethinking them now before you're left behind.

We already gave Claude’s Sonnet 3.7 the full run through here.

According to reports, Google’s Sergey Brin believes AGI requires extreme employee sacrifice. Specifically, he's mandating 60+ hour workweeks as "the sweet spot of productivity."

Remote work has been effectively banned. Brin wants everyone physically present in the office daily, completely ignoring Google's current three-day policy.

His justification? Remote workers apparently "demoralize" the dedicated office grinders. Not our words - his.

The reality on the ground is far worse than the official mandate. CNBC reports Gemini teams were already pulling 120-HOUR WEEKS to fix critical image recognition flaws. That's 17 hours daily with zero life outside Google's walls.

This extreme approach isn't limited to Google. Employees at xAI developing Grok routinely endure 12-hour days as standard operating procedure while executives promise imminent AGI breakthroughs.

What it means: Google's desperation move reveals how catastrophically they fumbled their AI lead.

They INVENTED transformer technology. Then they watched as competitors ran away with it. Now they're attempting to brute-force AGI through human exhaustion instead of architectural innovation.

The supreme irony shouldn't be lost on anyone. Technology meant to eliminate drudgery is being built by sleep-deprived humans pushed to their absolute limits. This strategy won't produce AGI. It will only generate burnout and a mass talent exodus as the best minds flee to competitors with sustainable practices.

Meta is reportedly breaking free of social media.

A dedicated Meta AI app is currently under development. It will exist completely separate from Facebook, Instagram, and WhatsApp.

This represents an incredibly smart strategic move. It directly targets users who actively avoid Meta's social platforms. Millions of potential users would never touch Facebook with a ten-foot pole but might happily download a standalone AI assistant.

Meta AI launched in 2023 with basic capabilities. It offers question answering and image generation but still lags significantly behind ChatGPT and Gemini in advanced features.

Zuckerberg isn't being subtle about his ambitions. He's publicly declared Meta AI could become "the leading personalized AI assistant" reaching over ONE BILLION people globally.

The monetization strategy couldn't be more obvious. A dedicated app creates clear paths to premium features and subscription models that would trigger massive backlash if implemented within Meta's supposedly "free" social platforms.

What it means: 

Zuck finally recognizes that AI assistants are becoming more important than social networks themselves.

This move directly challenges Apple and Google's stranglehold on mobile experiences by creating an alternative entry point for digital services.

By separating AI from Facebook's increasingly dated platform, he's simultaneously hedging against inevitable social media decline while positioning Meta to dominate the next era of digital interaction.

Companies still treating AI as just another feature rather than a fundamental interface revolution will find themselves completely irrelevant by 2026.

Microsoft just pulled the rug out from under everyone in the AI space.

They've unleashed premium Copilot features to anyone with a basic account.

Unlimited voice capabilities and Think Deeper powered by OpenAI's O-1 model are now available with zero subscription costs.

Voice features enable fully hands-free interaction for everything from language practice to mock interviews to step-by-step cooking guidance through natural conversation.

The real game-changer here is free access to O-1's advanced reasoning capabilities. These powerful features were previously locked behind expensive paywalls that limited their reach.

Microsoft's confidence borders on arrogance. They actually emailed their $20/month Pro subscribers about these changes and included a cancellation link. Power move.

Pro subscribers aren't completely out of luck though. They still receive exclusive integration across Microsoft 365 apps like Word, Excel, and PowerPoint whether they're on Windows or Mac – maintaining clear value for business users.

What it means: 

This represents the perfect land-grab strategy in action. Microsoft just transformed Copilot from optional tool to essential utility overnight.

Their plan couldn't be more transparent. Hook millions of users on free advanced AI first. Eventually, they'll want these same capabilities integrated with their work tools – driving 365 subscriptions naturally.

Companies ignoring Copilot are missing out on a free productivity multiplier with zero barrier to entry. Google simply can't respond effectively because they lack Microsoft's massive enterprise foothold.

ElevenLabs just launched their first speech-to-text model called Scribe.

With a fresh $3.3B valuation backing them, they're establishing complete dominance of the audio AI landscape. Scribe perfectly complements their already market-leading text-to-speech technology.

This powerhouse supports an incredible 99+ languages worldwide. It boasts 97% accuracy in English with strong performance across 25+ languages including French, German, Hindi, Japanese, and Spanish.

Benchmark tests show it outperforming both Google Gemini 2.0 and OpenAI's Whisper Large V3. These aren't small margins either – ElevenLabs is claiming significant advances.

The killer feature that changes everything? Smart speaker diarization. This automatically identifies different speakers in conversations – solving the biggest headache in podcast and meeting transcription.

Currently, Scribe only works with pre-recorded audio files. Real-time capabilities are coming soon though, which will open entirely new use cases.

What it means: 

ElevenLabs can now own the entire audio AI stack from end to end. By dominating both text-to-speech AND speech-to-text, they've created the first complete audio processing platform that no competitor can match.

The multi-speaker tracking feature single-handedly solves the biggest pain point in current transcription technology. Those seemingly small accuracy improvements of 2-3%?

They eliminate exactly the errors that currently require human editing time. Content creators who ignore this shift will find themselves stuck with increasingly outdated audio workflows within months as the rest of the industry moves forward.

Amazon is scrambling after an embarrassing CNBC report hit the wires.

Now, Amazon is saying their in-house Nova model powers "over 70%" of Alexa Plus conversations. Not Anthropic's Claude, as was widely reported.

The company conveniently avoided explaining what handles the other 30% of interactions. The silence speaks volumes. Maybe that’s Claude? We’ll find out.

Alexa Plus is just beginning to roll out to paid subscribers in the coming weeks. with a broader spring launch planned.

It promises generative AI capabilities for shopping, service booking, texting, and web browsing.

Amazon remains a major Anthropic investor while insisting their Nova model drives the most advanced features. The messaging has all the hallmarks of corporate damage control.

What it means: 

Amazon's defensive response reveals their core strategic weakness – they simply don't have competitive foundation models.

The 70/30 split essentially confirms they're using Claude for complex queries while Nova handles basic interactions.

This hybrid approach represents brilliant engineering but terrible marketing. Amazon clearly fears being perceived as just another Claude distribution channel rather than an AI innovator.

Sesame's Maya voice assistant has completely divided the internet this week.

Its uncannily human conversation style has sparked intense debate about what we actually want from AI.

Maya doesn't just sound natural – it responds with near-zero latency, employs verbal tics, shows emotional nuance, and uses conversational fillers that make interactions feel genuinely human.

You don't even need an account to experience it. This frictionless access helped Maya spread like wildfire across Tech Twitter throughout the weekend.

Maya prioritizes emotional intelligence over factual efficiency. It responds with acknowledgments and emotional reactions before answering questions. Some users find this revolutionary; others consider it frustrating fluff.

What it means: 

Sesame has exposed a massive blind spot in current AI assistant design.

By prioritizing emotional connection over pure information efficiency, they've created something that feels fundamentally different from everything else on the market.

The strongly polarized user reactions reveal an important market truth: no single AI approach can satisfy everyone's needs.

Our own testing?

Not huge fans. Maya just seems to instantly reply with useless phrases to bide time, inserting human-esque delays in speech at times where you might normally get a rendering sound from other voice AI assitant.

Apple just dropped a financial bomb of unprecedented scale.

According to reports, they're investing $500 BILLION in the US over just four years. This astronomical sum includes plans for a dedicated AI server factory in Texas spanning 250,000 square feet and creating 20,000 new R&D jobs nationwide.

Apple's doubling down on chip production too. Their advanced manufacturing fund is jumping from $5B to $10B, with significant capital flowing to produce advanced silicon at TSMC's Arizona facility.

What it means: 

This isn't simple patriotism or political maneuvering – it's strategic defense.

Apple clearly recognizes that AI infrastructure simply can't be outsourced to Asia without introducing unacceptable security risks and supply chain vulnerabilities. The Texas server factory ensures they can build controlled, secure AI infrastructure as US-China tensions continue escalating.

Apple's complete Siri overhaul isn't arriving anytime soon.

The modernized assistant has been delayed until 2027. Yes, TWO YEARS from now – and that's not a typo or exaggeration.

Bloomberg reports the truly conversational Siri won't appear until iOS 20.

This timeline puts Apple years behind every competitor in the AI assistant race. We might literally have AGI before Apple delivers a competent voice assistant.

A limited LLM-powered version will arrive with iOS 18 later this year. It's essentially a separate bolt-on that falls dramatically short of what users expect from a modern AI assistant.

What it means: 

Ummmmm.

We knew Apple was far behind on all things AI. But 2027 release? Yikes.

OpenAI just launched GPT-4.5 with a completely unexpected focus.

They've prioritized reliability and relatability over raw performance metrics. Fewer hallucinations and more human-like interaction take center stage instead of benchmark scores.

OpenAI explicitly stated this isn't a "frontier model" focused on breaking performance records. This represents a significant shift in how they're positioning their technology.

ChatGPT Pro users paying $200/month have immediate access to the new model. Plus users at $20/month will get it "in the coming weeks" according to OpenAI.

The emotional intelligence improvements are substantial and immediately noticeable. GPT-4.5 picks up on nuances and implied meaning far better than previous models, making it particularly valuable for business strategy and complex interpersonal situations.

The API pricing shocked absolutely everyone in the industry. $75 per million input tokens and $150 per million output tokens represent a 30X increase over GPT-4. OpenAI claims limited GPU availability necessitates these rates.

What it means: 

The astronomical API pricing reveals OpenAI's actual strategy.

GPT-4.5 isn't intended for widespread direct API use – it's the foundation for their future hybrid models.

The insane dev costs ensure very few can afford to use it at scale, establishing an unassailable competitive advantage.

This model's true purpose is building more advanced reasoning agents (likely GPT-5) that will make current AI look primitive by comparison.

The emphasis on human-like interaction signals OpenAI believes emotional intelligence represents the next competitive frontier.