Google's Gemini Gambit Reshapes the AI Race

Google's Gemini Gambit Reshapes the AI Race

From The Bit Baker Quarterly Roundup — Q1 2024

PLUS: OpenAI's video bombshell, Anthropic's triple threat, and the EU draws a line in the sand

Good morning, Dave. Three months ago, ChatGPT was still the name everyone reached for when talking about AI. By the end of March, that reflexive association started cracking. Google killed off Bard, rebranded everything under Gemini, and shipped a model that matched GPT-4 on benchmarks — then Anthropic walked in and arguably topped both.

What made Q1 2024 distinct wasn't any single launch. It was the shift from a one-horse race to a genuine three-way fight. And while the labs battled over model rankings, Brussels quietly passed the most consequential piece of AI legislation the world has seen.

In this quarter's Bit Baker:

• Google's Gemini gamble — killing Bard and betting the brand
• OpenAI's Sora preview stuns the creative world
• Anthropic's Claude 3 family arrives and challenges GPT-4
• The EU AI Act clears its final parliamentary vote

Google Kills Bard, Goes All-In on Gemini

The Bit Baker: Google made its boldest AI move yet on February 8 — retiring the Bard brand entirely and unifying everything under the Gemini name, with Ultra 1.0 as the flagship model behind a new $20/month tier.

Unpacked:

• Gemini Ultra is Google's first model to match or beat GPT-4 across major benchmarks in reasoning, coding, and multimodal understanding — a milestone after months of playing catch-up.
• The rebrand goes deeper than a name swap. Google folded Duet AI into Gemini too, making it the single brand for consumer, developer, and enterprise AI products across the Google ecosystem.
• A new Google One AI Premium plan at $20/month bundles Gemini Advanced with 2TB of storage — directly mirroring ChatGPT Plus pricing and positioning it as a head-to-head competitor.

Bottom line: Google essentially admitted Bard was a branding liability and chose to start fresh. Whether Gemini can dislodge ChatGPT from the cultural default slot depends less on benchmarks and more on whether Google can make it feel indispensable inside the apps people already use.

OpenAI Previews Sora and Breaks the Internet

The Bit Baker: On February 15, OpenAI unveiled Sora, a text-to-video model capable of generating photorealistic, minute-long videos from text prompts — and the demo clips left the internet stunned.

Unpacked:

• The sample videos showed a level of coherence nobody expected yet: a woman walking through Tokyo in snow, an SUV on a mountain road, historical footage that looked archival. Object permanence, lighting, and physics all held together in ways previous video AI couldn't manage.
• Sora wasn't released publicly — OpenAI gave access only to red teamers and select artists for safety testing, making this a research preview rather than a product launch.
• The announcement landed weeks before the primary election cycle intensified, triggering immediate debates about deepfakes, misinformation, and whether regulatory frameworks could keep pace with generation quality this high.

Bottom line: Sora didn't ship as a product in Q1, but it didn't need to. The preview alone forced every media company, platform, and regulator to recalibrate their assumptions about synthetic video. OpenAI effectively set the agenda for every AI-and-media conversation that followed.

Anthropic Ships Claude 3 — and It's Not Messing Around

The Bit Baker: Anthropic launched the Claude 3 family on March 4 with three models — Opus, Sonnet, and Haiku — and Opus posted benchmark scores that matched or exceeded GPT-4 across vision, math, and coding tasks.

Unpacked:

• Claude 3 Opus topped GPT-4 on multiple benchmarks while adding genuine multimodal capabilities — it can analyze photos, charts, and technical diagrams, a first for Anthropic's models.
• All three models ship with a 200,000-token context window, roughly 150,000 words. That's enough to drop an entire codebase or a novel-length document into a single prompt.
• The tiered approach — Opus for power, Sonnet for balance, Haiku for speed — gave developers pricing flexibility that OpenAI's lineup lacked at the time.

Bottom line: Claude 3 turned Anthropic from a "safety-focused alternative" into a legitimate frontrunner. For the first time, the AI race had three credible contenders at the frontier, not two.

EU Parliament Passes the World's First Major AI Law

The Bit Baker: The European Parliament voted 523-46 to adopt the EU AI Act on March 13, creating the world's first comprehensive legal framework for artificial intelligence regulation.

Unpacked:

• The Act sorts AI systems into risk tiers — from banned (social scoring, real-time biometric surveillance in most cases) to high-risk (hiring tools, credit scoring) to limited and minimal risk — each with escalating compliance requirements.
• Companies face staggered deadlines: prohibited practices must stop within 6 months, high-risk rules kick in at 24 months, and general-purpose AI models like GPT-4 and Gemini fall under transparency obligations within 12 months.
• The legislation's reach extends beyond EU borders. Any company serving EU users must comply, mirroring GDPR's "Brussels effect" — which means OpenAI, Google, and Anthropic are all in scope regardless of where they're headquartered.

Bottom line: Brussels drew first. The EU AI Act won't stop anyone from building models, but it will shape how every major AI company designs products, documents training data, and communicates capabilities. The rest of the world is now either copying, adapting, or deliberately choosing a different path.

The Shortlist

NVIDIA reported record Q4 fiscal 2024 revenue of $22.1 billion — a 265% year-over-year jump — with data center revenue alone hitting $18.4 billion as AI chip demand showed no signs of cooling.

Google released Gemma, a family of lightweight open-source models (2B and 7B parameters) built on the same research as Gemini, giving developers a free on-ramp to Google's AI stack.

Apple canceled its decade-long Project Titan car effort in late February, reassigning roughly 2,000 employees to generative AI projects in a dramatic strategic pivot.

Microsoft hired Inflection AI co-founder Mustafa Suleyman to run a new consumer AI division, a $650 million acqui-hire that folded Copilot, Bing, and Edge under one AI-focused leader.

Stability AI lost founder and CEO Emad Mostaque, who resigned citing concerns about power concentration in AI — leaving the Stable Diffusion maker searching for new leadership amid financial pressure.

References

1. Google goes all-in on Gemini and launches $20 paid tier
2. Google One AI Premium plan announcement
3. OpenAI unveils Sora text-to-video model
4. Sora first impressions from red teamers and artists
5. Anthropic launches the Claude 3 family
6. Anthropic's Claude 3 models on Google Cloud Vertex AI
7. EU Parliament adopts the AI Act
8. NVIDIA Q4 fiscal 2024 financial results
9. Google releases Gemma open-source models
10. Apple cancels Project Titan car effort
11. Mustafa Suleyman joins Microsoft to lead Copilot
12. Stability AI CEO Emad Mostaque resigns