Lattepunk R1
Lattepunk
Did a Chinese AI ruin your retirement?
Last week, the part of the internet that obsesses over technology collectively lost its shit. All because of a new AI model by DeepSeek came out. If you don’t care about AI, like…at all…this news doesn’t matter. New AI models drop all the time. But this one was different. This one could have indirectly affected you.
The S&P 500 is made up of roughly 35% of tech stocks, so if your retirement fund is allocated to that, you may have noticed a dip. Nvidia made history by losing $600 billion dollars in value. Again, all this cause of a new AI model. I’m definitely not qualified to explain this, let alone in a simple way. We nerding out a bit today reader, I’ll try to keep simple.
The whole thing about AI is to make a computer that is smarter than humans at all things. That’s called Artificial General Intelligence (or AGI for short). According to those making it, it’s going to set the world on a new trajectory. Nilay Patel of The Verge jokingly refers to it as these companies trying to make God. Which is simple way of thinking about it. They want to build a computer version of God that will fix every problem ever and will make humanity better. Don’t focus on the ridiculousness of it, just know that to make God it’s very expensive. You know ChatGPT right? They announced a new project to spend $500 BILLION over four years towards this purpose. That doesn’t account for the money that was already spent.
Let’s be reasonable!
One day before that announcement, DeepSeek dropped its DeepSeek R1 model. It’s a reasoning model. When you use ChatGPT, it defaults to its GPT4o model. They have another model called o1, which is a reasoning model. I’ve said that twice in quick succession, what the fuck is a reasoning model? Normally, you type something into the chat box and it finds the right words to reply to whatever you wrote. Reasoning models “think” before they answer back, essentially making them “smarter” in their replies. That’s the model that R1 competes with. o1 is ChatGPT’s reasoning model, R1 is DeepSeek’s reasoning model. Cool? Let’s keep going.
All these models are compared on specific scale. That really important to us, just know all this stuff is subjective. But reasoning models are expensive (remember, $500 BILLION). So when DeepSeek dropped R1, a model that was just as good or possibly better than o1, the world noticed. When it was found out that it cost only $6 million dollars to make it, everyone freaked out. NY Times had a great piece, as did WIRED, on who the hell is DeepSeek. As the information came out, that’s when the stock panic happened.
A new king of AI?
The DeepSeek app became the number one app in the both the Apple App Store and the Google Play Store, unseating…drumroll please…ChatGPT. AI was having another moment. 404 Media covered it well. So the cost is what caused that, but why was it actually popular? Few reasons, but I want to cover one specific. It was open sourced. What the fuck is open sourced you may be asking? That means you could look at all the code, reuse the code however you see fit, all for free. Seriously, go look at it yourself. This is huge because even though the makers of ChatGPT are called OpenAI, how their AI works is actually not known. This is able to enable ANYONE to implement AI into their business/job/career for free as long as you have the technical knowhow (more on that a bit later). But there is one little problem: DeepSeek is from China.
Here’s three headlines from WIRED (seriously a great website, visit it regularly):
DeepSeek’s Popular AI App Is Explicitly Sending US Data to China
Exposed DeepSeek Database Revealed Chat Prompts and Internal Data
DeepSeek’s Safety Guardrails Failed Every Test Researchers Threw at Its AI Chatbot
The West has a simple philosophy: America = good, China = bad. It’s harsh to generalize like that but it did seem to be the biggest criticism (among many, all legit). You remember about TikTok being banned, then unbanned, and now waiting for possibly another ban? All that was because China = bad. Xenophobia is real here in the states. I don’t know who General Tso was, but the man is a legend in my stomach. All that DeepSeek hyped led to a lot of US concern. One more WIRED headline:
Here’s where another stand out feature about R1: since it’s open source, I can just use it on my personal computer! When you install the ChatGPT app (or any other AI app) onto your device, it just connects you to the server that the AI model lives on. None of that stuff is actually running on your device. DeepSeek can run on your device. You don’t need to worry about sending your data to China if it all happens on your computer.
So can you run this DeepSeek model?
Do you own a roughly $6000 computer? Then probably not. The model is pretty intensive to run. But they took other open source models that and applied their R1 reasoning to them. Those models are smaller and definitely run on your computer (or phone if you don’t love your battery life). Here’s a quick rundown on how I used DeepSeek (or any other model) locally:
I downloaded this program called Ollama.
Then, open your Terminal and type this:
ollama run deepseek-r1:7bThat will download the model to your computer and start a conversation. It''ll look something like this:
YOUR_COMPUTER_NAME_HERE ~ % ollama run deepseek-r1:7b
>>> Send a message (/? for help)(If you have questions about using Ollama on your computer, then it sucks to suck cause I’m not helping you more than that.)
But yeah, DeepSeek! Now let’s hope we don’t go to jail for using it. Oh! If you were wondering “How is OpenAI going to respond to DeepSeek?”, that’s a great question! Besides anger, fear not. They got a new reasoning model for you. o3-mini is ready for you. What happened to o2? Name was taken by a UK telecom. Can’t win ‘em all.
some quick hits
Remember the Pebble smartwatch? Google open sourced the software and they’re making a comeback!
A porn app is finally going to be allowed on iOS. It’s called Hot Tub and you can only install it if you’re in Europe. Apple isn’t happy about it!
Bookshop.org is online book store that supports local bookstores in your community. It’s already a cool service in itself. Now you can buy ebooks. Go support local!
things i’ve read
Twelve Dudes and a Hype Tunnel: Scenes from the ‘Super Bowl for Excel Nerds’ | Yan Zhuang for NY Times
Subaru Security Flaws Exposed Its System for Tracking Millions of Cars | Andy Greenberg for WIRED
Why Amazon is struggling to crack Argentina | David Feliba for Rest of World
How shutdown Bay Area tech companies ditch their fancy gear fast | Stephen Council for SFGATE
Her dad, the 10,000 records he left behind and a viral lesson in grief | Janay Kingsberry for Washington Post
Emergency Braking Will Save Lives. Automakers Want to Charge Extra for It | Carlton Reid for WIRED
AI haters build tarpits to trap and trick AI scrapers that ignore robots.txt | Ashley Belanger for ArsTechnica
Scans for the memories: why old games magazines are a vital source of cultural history – and nostalgia | Keith Stuart for The Guardian
How Richard Mille Takes Quartz Watches to a Surprising Level | Tim Barber for WIRED
Faking It: Deepfake Porn Site’s Link to Tech Companies | bellingcat
Elon Musk rose to the top of video game charts. Now he has confessed to cheating. | Drew Harwell for The Washington Post
I tried the tech that makes hands-free smart locks actually work | Jennifer Pattison Tuohy for The Verge
He Went to Jail for Stealing Someone’s Identity. But It Was His All Along. | Mitch Smith, Erin Schaff for NY Times
A man stalked a professor for six years. Then he used AI chatbots to lure strangers to her home | Katie McQue for The Guardian
If you enjoyed this, share it with someone.