😺 American vs Chinese AI

Welcome, humans.

Yesterday we wrote about OpenAI’s “Benchmarkgate” scandal, for those who missed it. The subject line that won our A/B test was “OpenAI’s ‘Theranos moment’”, which actually refers to a part of the story we cut out for length.

Here it is below for your viewing pleasure (and because some of you astutely called us out for the audacious claim).

“Some people are calling this OpenAI's “Theranos moment”, but that feels like a stretch. OpenAI actually has working products.

Now a REAL Theranos-style scam would be like if ChatGPT was actually just 10,000 people in a warehouse frantically Googling your questions and typing responses while a manager yells ‘FASTER! They'll know it's humans if you take more than 0.3 seconds!’”

Here’s what you need to know about AI today:

• DeepSeek released powerful open-source AI that rivals American models.
• Pentagon and CIA deployed new AI for military + strategic planning.
• There’s a new video generator called Vidu 2.0.
• Study found 26% of teens use ChatGPT for homework.

Advertise in The Neuron here.

America wants to beat China at AI… but DeepSeek is crushing American AI right now

There was quite an AI turnout at Trump's inauguration yesterday—every major tech CEO showed up. We're talking Zuck, Sundar, Tim Cook, Bezos, and (of course) President Trump’s “first buddy” Elon Musk.

We didn’t see SamA and Satya Nadella in any photo ops though… Gee, I wonder why…

But here's the thing: the gang wasn't there just to chill. They were there as AI CEOs.

Just days before the inauguration, Jake Sullivan (Biden's outgoing national security adviser) dropped some pretty heavy warnings about AI:

1. The next few years will determine whether AI leads to catastrophe—and whether China or America wins the AI arms race.
2. Unlike previous tech races (atomic weapons, space, internet), AI development sits outside government control.
3. The stakes? Everything from “democratization of extremely powerful and lethal weapons” to “massive job disruption” and even “avalanches of misinformation.”

And according to Sullivan, America needs unprecedented government-private sector collaboration to win.

Now here's the plot twist: while America's tech giants were busy toasting Trump, a Chinese company just casually released what might be the biggest AI breakthrough since ChatGPT.

Meet DeepSeek, or as the cool kids at Latent Space call them, the “Whale Bros.” Yesterday, they released DeepSeek-R1, and... wow:

• It matches (and sometimes beats) OpenAI's o1 across math, code, and reasoning tasks.
• The smallest model outperforms GPT-4o and Claude 3.5 Sonnet on math benchmarks.
• Oh, and it's completely open source. MIT license, available via API.

Crazy demo of R1… and more here.

But the real “aha moment” of R1 isn't the performance—it's how the Whale Bros got there. DeepSeek discovered that pure reinforcement learning enables a language model to automatically learn to think and reflect. Here’s what that means:

1. OpenAI convinced everyone you need massive training data to teach AI how to think step-by-step.
2. DeepSeek said “nah” and just let their model learn through trial and error.
3. The wild part? It worked—going from solving 15.6% to 86.7% of math problems just by practicing.

This is probably terrifying to OpenAI because it means you don't need their secret sauce anymore—just computing power and the right incentives. Just like AlphaGo mastered Go by playing itself instead of studying human games, DeepSeek proved you can do the same for “reasoning.”

The ultimate irony? While OpenAI is building a closed-off garden, Chinese companies are actually OPEN, as Dr. Jim Fan from NVIDIA noted.

And DeepSeek isn't alone. Many Chinese AI models are open + crushing it:

• Alibaba's Qwen models are the most downloaded on Hugging Face.
• These models often perform better than Meta's Llama while being cheaper.
• Most of the best video models are Chinese, including Hunyuan (the next best video model behind Google’s Veo 2—which is extremely good).

If you want to try R1 yourself, go here, activate “DeepThink,” and you can play with o1-level AI... for free.

FROM OUR PARTNERS

Here’s one AI tool we use every week at The Neuron.

Noah rambling… probably

A little well-kept secret for companies crushing it with AI is that only a handful of AI tools are actually worth using.

That’s why we—Noah & Grant—use ChatGPT and Claude, along with a killer product called Attention.

Here’s how it works:

1. We tell Attention exactly what to pay attention to in our meeting (goals, budget, etc).
2. Attention listens in to our call.
3. After, Attention outputs important insights, action items, and follow-up emails.

There are way, way more features, and it’s a must-have for all sales teams. (If you sell anything, literally drop what you’re doing and go book a demo with Attention ASAP).

Prompt Tip of the Day

Okay, this one’s just for fun: Redditors are claiming they achieve “great success” (cue Borat voice) with the following custom instructions: “I want you to respond like an unhelpful zoomer in a TikTok comment section”

Try it out and send us your best replies in the poll @ the bottom!

Treats To Try.

1. *Check out this ultimate guide to building powerful voice apps combining LLMs and speech recognition, with valuable insights + advice from hundreds of audio-first companies. Get your copy here.
2. Vidu 2.0 is a new video model with improved subject consistency and three resolution options up to 1080p (demo).
3. Outrank finds untapped keywords and automatically publishes daily SEO-optimized blog posts to grow your website's traffic (3 day free trial available).
4. Milestone Content Studio turns your ideas into published content while automatically handling SEO, brand voice, and optimization (7 day free trial).
5. AI_licia co-hosts your Twitch and Discord streams by chatting with viewers and remembering their past interactions so you can focus on streaming.
6. Duory turns your Duolingo lessons into a personal diary where you can record and replay phrases from your lock screen (here’s a similar open-source tool).
7. Langchain AI Email Assistant is an “executive assistant” to manage your Gmail inbox that’s completely open source—code here, read more here.

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

• The Pentagon now uses genAI to accelerate its military “kill chain” to more quickly identify, analyze, and strategize potential threat responses.
• A new study found 26% of students between the ages of 13-17 now use ChatGPT with help on their homework—sounds too low TBH.
• The writer of Taxi Driver (Paul Schrader) was “STUNNED” by ChatGPT’s creative output when he asked it to generate ideas in the style of famous directors like PTA, Tarantino, and Scorsese.
• The CIA apparently has a chatbot that can impersonate any foreign leader.

FROM OUR PARTNERS

Turn your messy docs into GenAI gold

Join Unstructured's Product Lead tomorrow Jan 22nd (10am PT/1pm ET) to see exactly how leading companies are turning complex documents into clean, structured data that GenAI actually loves.

Click here to register and receive the recording, even if you can’t attend live.

Tuesday Ticker

Here are the results from last week’s poll:

• I.H. chose SearchGPT: “I use ChatGPT Pro, projects in ChatGPT, SearchGPT en Canvas. I don't live in the US so Sora isn't available for me yet.”
• N.G. chose Projects: “We have a Team subscription with 2 accounts used by around 20 different people and Projects are a huge help to organize chats. Canvas still feels a bit underperforming right now (except on the coding part!) but it has potential!”
• E.W. chose GPT + Apple Intelligence: “It is definitely an improvement for Siri. A bit hit and miss in terms of relevant answers but works well with ability to use text in Siri. I would generally use it for small occasional everyday questions in my personal life, I tend to use ChatGPT on the desktop for work.”

A Cat's Commentary.