ENGLISH NEWS International

CHINA’S DEEPSEEK R1 CRUSHES CHATGPT : META PANICS OVER CHINA’S AI BREAKTHROUGH

Mario Nawfal

@MarioNawfal
🚨🇨🇳CHINA’S DEEPSEEK R1 CRUSHES CHATGPT—AI GAME CHANGER OR SHORTCUT SUCCESS?

DeepSeek R1 isn’t just another AI model—it’s a massive disruption. This Chinese AI runs locally, costs 93% less than OpenAI’s o1, and just topped the App Store charts, something Gemini and Claude never pulled off.

But there’s a catch—its $6M training cost is a mirage. DeepSeek likely leveraged years of research, huge GPU clusters, and distillation from GPT-4o and o1, sidestepping U.S. chip bans and proving that China doesn’t need to buy the cow if it can milk the data for free.

If AI inference moves to the edge—Mac Studio today, superphones in 2 years—it could wipe out the need for power-hungry AI data centers and trigger the biggest PC and smartphone upgrade cycle ever. But if AI stops making breakthroughs? The real winners will be Google, Meta, and X—the companies using AI best.

And then there’s Grok-3. If Elon’s model scales better than OpenAI’s, we might see the AI arms race accelerate even faster. The question now: Is ASI coming sooner than we think?

Source:
@GavinSBaker

Mario Nawfal

@MarioNawfal
🚨🇺🇸META PANICS OVER CHINA’S AI BREAKTHROUGH

Meta has set up 4 war rooms to dissect China’s DeepSeek, scrambling to understand how it slashed training costs and what data it used.

DeepSeek, an open-source LLM, claims to rival OpenAI and Meta’s AI models—but on a fraction of the budget. Now, Silicon Valley is in full-blown crisis mode.

2 of Meta’s war rooms are focused on High-Flyer’s cost-cutting methods, another is digging into its training data.

If DeepSeek really cracked AI efficiency, U.S. tech giants have a major problem.

Source: Business Insider

Mario Nawfal

@MarioNawfal
🚨🇨🇳BREAKING: DEEPSEEK DROPS JANUS-PRO 7B, NEW OPEN-SOURCE AI MODEL, SHAKES AI MARKET AGAIN

Fresh off the chaos caused by its R1 release, China’s DeepSeek just launched another open-source AI model—Janus-Pro 7B—on Hugging Face.

This multimodal model claims to match or surpass task-specific AI models and rivals OpenAI’s DALL-E 3 in unified understanding and generation.

After R1’s debut wiped nearly 15% off Nvidia’s valuation, DeepSeek has now restricted registration to Chinese phone numbers as demand surges.

Another shot fired in the AI arms race.

Source: Seeking Alpha