- Everyday AI
- Posts
- DeepSeek Deep Dive: AI’s Sputnik Moment or a National Security Threat?
DeepSeek Deep Dive: AI’s Sputnik Moment or a National Security Threat?
OpenAI cancels o3 release for GPT-5 plans, Apple’s humanoid robotics plans, Adobe releases AI video model and more!
Everyday AI
February 12, 2025
👉 Subscribe Here | 🗣 Hire Us To Speak | 🤝 Partner with Us | 🤖 Grow with GenAI
Outsmart The Future
Today in Everyday AI
6 minute read
🎙 Daily Podcast Episode: DeepSeek went from the top of the AI world to now being restricted by many. What happened? Is it safe to use? We break down DeepSeek and its rise to popularity. Give it a listen.
🕵️♂️ Fresh Finds: Meta to acquire AI chip firm, Adobe unveils Firefly AI subscriptions and Gemini Deep Research added to mobile. Read on for Fresh Finds.
🗞 Byte Sized Daily AI News: OpenAI cancels o3 release for GPT-5 plans, Apple’s humanoid robotics plans and Adobe releases AI video model. For that and more, read on for Byte Sized News.
🚀 AI In 5: Here’s an AI research hack that allows you to use multiple AI chatbots at once. See it here
🧠 Learn & Leveraging AI: With all the commotion surrounding DeepSeek it might be hard to tell what’s actually going on. Don’t worry we got you covered. Keep reading for that!
↩️ Don’t miss out: Did you miss our last newsletter? We talked about OpenAI Operator, US and UK skipping international AI agreement, Apple and Alibaba’s AI plans for China and Workday’s AI agent platform for enterprises. Check it here!
DeepSeek Deep Dive - AI’s Sputnik Moment or a National Security Threat? 🤷
$100 million fine if your company uses DeepSeek?
A few weeks ago, DeepSeek was the internet's darling. After grabbing international headlines and shaking the U.S. stock market to its core, it's been an interesting week or two for the Chinese AI company.
So what's actually happening here?
Should you actually use the model? Is it safe? Is it really a SOTA open source model?
We dive into DeepSeek and break it all down.
Join the conversation and ask Jordan questions on AI here.
Also on the pod today:
• DeepSeek’s capabilities and benchmarks 📊
• Global Reactions and Controversies 🌎
• Analysis and Criticism of DeepSeek 🧐
It’ll be worth your 56 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Here’s our favorite AI finds from across the web:
New AI Tool Spotlight – Ariadna is an AI talent agent, 3D AI Studio generates and textures 3D models and Bild AI creates drawing and blueprints for builders.
Meta – Meta is in talks to acquire AI chip firm FuriosaAI.
Adobe – Adobe has launched subscriptions for its Firefly AI.
Google – Google says it’ll use machine learning to estimate a user’s age.
Gemini’s Deep Research mode is now rolling out to Android and iPhone.
AI in Media – Scarlett Johansson is calling for AI safety laws after fake videos of celebrities have emerged protesting Kanye’s recent controversies.
AI Copyright - Thomas Reuters has won a court battle over AI, copyright and fair use.
Perplexity – Perplexity tried to sponsor the Red Bull Racing F1 team, but it was blocked by Oracle, the team’s largest sponsor.
Read This – MyFitnessPal app is getting AI-generated meal plans.
1. OpenAI Cancels o3 Launch for GPT-5 Plans 🗳️
OpenAI has decided not to release the "o3" as a standalone model, announced CEO Sam Altman at an AI summit in Paris. Instead, the company plans to launch a comprehensive GPT-5 model, integrating the o-series with other technologies for a more cohesive AI system. This follows the recent unveiling of o3 models in December, amidst rising investor concerns over tech spending.
Additionally, the upcoming GPT-4.5, dubbed "Orion," will be OpenAI's last model without advanced reasoning capabilities, signaling a shift towards more sophisticated AI solutions.
2. Apple’s Humanoid Robotics Plans 💼
Apple is diving into robotics, exploring both humanoid and non-humanoid designs, according to Apple analyst Ming-Chi Kuo. This news follows a research paper from Apple examining human interactions with robots like a Pixar-style lamp. While some see this as a futuristic step, Kuo notes it's in the early stages, with mass production optimistically projected for 2028.
As Apple cautiously navigates this field, the potential for a home robot—perhaps more practical than a humanoid—is on the horizon, promising to reshape how we think about smart home ecosystems.
3. Adobe Unleashes New AI Video Tool in Public Beta 🚀
Adobe has rolled out its Generate Video tool in public beta. This tool enables users to create videos from text or images, now boasting 1080p quality and seamless integration with Adobe's Creative Cloud apps. Amid fierce competition from OpenAI's Sora and Google's Veo, Adobe emphasizes the commercial viability of its Firefly AI, trained on public domain and licensed content.
With new subscription plans and additional features like Scene to Image and Translate Audio and Video, Adobe aims to empower creators while upholding copyright safety.
4. OpenAI Unveils Expanded Model Spec with More Freedom and Flexibility 🔐
OpenAI has released a substantially expanded version of its Model Spec, now 63 pages, to guide AI behavior with an emphasis on customizability, transparency, and intellectual freedom. This update encourages thoughtful interaction on controversial topics, introduces a "grown-up mode" for mature content, and tackles "AI sycophancy" to ensure more authentic interactions.
OpenAI is releasing this under a Creative Commons Zero license, inviting public feedback during a time of heated debate on AI ethics and behavior.
5. Microsoft Powers Up with Renewables ⚡
Microsoft is amping up its renewable energy game, adding 389 megawatts from solar projects in Illinois and Texas to fuel its AI-driven ambitions. This move nudges Microsoft’s renewable capacity to nearly 20 gigawatts, as the tech giant races to meet its 2030 carbon-negative pledge. As solar farms can be deployed swiftly compared to traditional gas plants, Microsoft's commitment underscores the urgency in powering its AI operations sustainably.
Meanwhile, Microsoft's investment in carbon removal technologies further highlights its dedication to reducing emissions, potentially setting a benchmark for tech-driven sustainability initiatives.
6. OpenAI’s o3 Model Triumphs at IOI 2024⚡
OpenAI's o3 reasoning model clinched a gold medal at the International Olympiad in Informatics (IOI) 2024, showcasing its prowess without relying on predefined strategies. According to OpenAI, this marks a significant leap forward in AI development, as o3 outperformed its predecessors like AlphaCode2 and o1-ioi, reaching the 99th percentile on CodeForces with pure reinforcement learning.
Unlike o1-ioi, which needed extra computation time, o3 adhered to strict IOI constraints, setting a new standard for AI in competitive programming. This victory underlines the potential of reinforcement learning to revolutionize algorithmic reasoning and hints at transformative applications in data analysis and beyond.
How to Save Hours of Research with Multiple AI Chatbots at Once.
Click Image To Play Video 👆
Are you tired of spending countless hours on research and competitive analysis?
We’re showing you how you can run prompts through multiple AI models like GPT-4o, Claude, and Gemini to save time on research.
Check out today's AI in 5.
🦾How You Can Leverage:
A Chinese hedge fund turned AI company just convinced the entire world they actually built state-of-the-art AI for less than a Manhattan parking spot.
Lolz.
Two trillion dollars in market losses later, it's time to dive a lil deeper on DeepSeek.
We haven't been ignoring the DeepSeek sitch, y'all.
While tech Twitter had a collective meltdown over sub-$6 million AI and VCs were busy calling it "AI's Sputnik moment," we've been in the lab.
Reading terms of service line by line. Reading new American legislation in response to Chinese AI. Tracking regulatory bans country by country.
The Deep Dive on DeepSeek is actually about a hedge fund-turned-AI-company that temporarily wiped $600B from NVIDIA's market cap faster than you can say "definitely not market manipulation."
This is about ONE MILLION chat logs leaked from unsecured databases while their mobile app casually climbed to #1.
This is about terms of service that basically say "everything you upload belongs to China now xoxo."
Forget the benchmarks.
Forget the hype.
This isn't just about AI anymore.
This is geopolitical influence.
World economies.
This is about who controls the next decade of innovation.
Buckle up and let’s dive deep.
1 – The $5.5 Million Model? Not Quite 💸
In late January, DeepSeek released its newest reasoning model, R1.
The results looked impressive.
They said it was Open Source, cost-effective and it benchmarked with the big boys.
On some tests, its outputs were better than OpenAI’s and Google’s world-leading models.
DeepSeek claimed in their R1 model research paper that their training cost was just $5.57 million dollars.
That’s like pennies on the dollar, as most State of the Art LLMs cost $500 million to more than a billion.
That claim — and the ensuing news coverage and social media freakout — played a major role in DeepSeek’s meteoric rise and the nearly $2 trillion in market cap that was wiped out from U.S. Big Tech.
Was that $5.5ish number the truth?
Apparently not.
That viral $5.5M training cost? Just the final training run costs, he said. Not the full development. Not even close. Their ‘final run’ figure was just a fraction of the real costs.
Try This:
It doesn’t stop there.
Let’s keep digging, shall we?
Semi-Analysis, the leading semiconductor industry research firm, dropped the actual receipts: they estimated the actual costs to be about $1.6 BILLION.
Fifty thousand NVIDIA Hopper GPUs. The same GPUs that aren't even supposed to be exported to certain regions in China.
2 – Open Source? Not Really ❌
OK, so we’ve got the viral pricing SNAFU figured.
We hear what you’re thinking next.
You: It’s Open Source, Jordan! This changes everything! It makes it safe and super cheap to put into production.
Jordan: No. Just no.
Everyone's screaming "open source revolution!" but nobody actually read the Open Source Initiative's Open Source AI Definition (OSAID). Shocking.
(Spoiler alert… we obviously did.)
DeepSeek falls short on at least two OSAID requirements:
First, it restricts usage in specific regions and domains. (You know…. The countries it’s currently banned in.)
Second - and this is the big one - it provides zero access to training datasets or sufficient details for independent model replication.
That's like calling yourself vegan while eating a bacon cheeseburger.
Don’t work like that.
Try This:
Sure, DeepSeek can be good for the open source AI movement. But it’s not actually Open Source.
And you can probably guess…. It’s not very transparent either.
According to the AI Data Transparency Index, DeepSeek’s R1 is one of the least transparent models out there.
(GPT-4o, Claude 3.5 Sonnet and Llama 3.2 all get higher transparency scores.)
3 – Privacy and Data Concerns Everywhere 🔓
January 27th: DeepSeek becomes #1 most downloaded free app.
January 30th: Cybersecurity firm Wiz exposes unsecured database with one million+ chat logs and API keys.
February 3rd: Research reveals unencrypted data transfers.
February 7th: Cisco reported a 100% successful jailbreak rate for DeepSeek R1. (HUGE Yikes)
Yikes.
Let’s get usage straight though shorties.
Companies like Microsoft, Perplexity, AWS and others offer modified versions of DeepSeek models that keep your data at home and have fixed much of the EXTREMELY concerning censorship the model carries with it.
Buuuuut so many people/companies/developers are using DeepSeek natively, via their web app, API or iOS app.
And that’s a legit privacy and data security nightmare.
But here's the nuclear scenario: Their updated terms of service in Section 4.3 just removed ALL data protection.
The old version? Data processed only for service improvement.
The new version?
Everything you upload belongs to them. Forever.
No opt-out. No enterprise controls. Just straight to servers where the Chinese government has unlimited access rights if they want to.
Worse yet?
We’ve seen countless Saas companies, devs and AI shops mindlessly swap out their AI API connection from OpenAI or Google and tag in Deep Seek, thinking it will save them money.
Buuuuuuuut, that means that anyone using said services might be unknowingly uploading their company data and sending it straight to China, with or without love.
(Spoiler alert: Google and OpenAI released new models a day later that wiped out any cost-advantage DeepSeek had.)
Try This:
The Pentagon banned it. NASA banned it. The U.S. Navy banned it.
A new U.S. Senate bill proposes $100M fines for companies using it. Prison time for willful violations.(Don’t worry… that thing has a negative shot at becoming law.)
You still think DeepSeek is worth saving a few cents per million tokens?
Now This …
Let us know your thoughts!
Vote to see live results
What are your thoughts? |
Reply