I hope all 9429 of you had a great weekend 🔥
Welcome to the 22 new members of the TOC community.
If you're enjoying my writing, please share Terminally Onchain with your friends in Crypto!
Sections below:
AI's Google S-1 Moment
Opens source for the non-technical
Coinbase listings uproar
Let's dive in.
This past weekend, the main character on tech twitter was obviously the Deepseek R1 model. If you were out touching grass, here's the 30 second rundown:
DeepSeek (Chinese AI startup) launched the model DeepSeek R1 which matches or surpasses OpenAI’s o1 model in performance
Open-source under the MIT License model
The R1 model was developed for under $6 million vs the billions of dollars it cost OpenAI and American tech giants like Meta and Google
Even the API pricing is significantly lower. $0.14 per million tokens compared to OpenAI’s $7.50
Broader implications is that everyone is worried American AI is behind China's AI. And there were a ton of discussions around "it being over" for Nvidia.
Now, I'm sure all of you have seen countless reports, hot takes, and debates on the topic so I won't go too in depth here. Instead, I'll just share a few points from others that helped me form a better opinion.
Overall, even if you're bearish and think DeepSeek fudged the training cost numbers a bit, this is still a phenomenal achievement. I mean, the performance was on par for a fraction of the cost and the paper is out in the public for American companies to copy!
Initially, much of the conversation was centered around the fact that because DeepSeek proved it costs so much less to train a model, then Nvidia will take the biggest hit because there won't be a need for that many GPUs anymore. If you took one look at the market yesterday, it was clear that this opinion was reflected with Nvidia and semiconductors in general being down 15%.
But then Satya tweeted this which changed the narrative. Basically, if the cost of a technology goes down --> then it becomes accessible to more people --> which leads to increased usage of that tech --> and that slowly drives costs back up again over time. jEvOnS pArAdOx
And to be honest, this feels pretty straightforward to me. This isn't unprecedented. It's happened countless times with prior technologies such as computer storage, fuel for transportation, and agriculture.
The two best tweets I saw on the matter were from Yishan and Zhiding Yu.
In one sentence, the takeaway was that Deepseek R1 does not have to be an extreme geopolitical matter.
Yes...obviously there are tensions between which country has more control over AI. But this specific moment of a Chinese company sharing their results and open sourcing the work has been dramatized into a war like narrative. Remember, the Deepseek model weights are all open source! Meaning you can go and use their model locally on your computer without having to use their app that the Chinese government may or may not be monitoring. In fact, Perplexity has already implemented the model into their product so you can use it there as well.
A bit long, but read this from Yishan:
After everyone has been sensitized over the H1BLM uproar, we are conditioned to think of OMG Immigrants China as some kind of Alien Other. As though the Alien-Other Chinese Researchers are doing something special that's out of reach and now China The Empire is somehow uniquely in possession of Super Efficient AI Power and the US companies can't compete. The subtext of "A New Fearsome Power Now Under The Command of the CCP" is what's driving the current sentiment, and it's not really valid.
Like, no. These are guys basically working on the same problems we are in the US, and not only that, they wrote a paper about it and open-sourced their model! It is not actually some sort of tectonic geopolitical shift, it is just Some Nerds Over There saying "Hey we figured out some cool shit, here's how we did it, maybe you would like to check it out?"
That last sentence was so real. Like why can't Deepseek R1 be a cool breakthrough that researchers are just excited to share with each other? I know you're probably thinking "YB you're so naive, this has CCP written all over it". And yea maybe you're right. But based on the facts I see so far, it seems to me that a bunch of smart researchers in China had an amazing breakthrough and shared all the details with the rest of the world. All major American players will replicate this in a matter of months into their training flows and we're back to an even playing ground.
As someone who loves learning the history of tech, Yishan's analogy of this Deepseek moment to Google's S-1 filing in 2004 made a ton of sense to me.
Basically, in that era, in order for companies to scale they had to buy larger mainframes (similar to buying more GPUs). But then Google came out of nowhere and fixed the issue by revealing that they didn't need to scale vertically anymore because they built the largest supercomputer through distributed networking.
Eventually, Google published their MapReduce and BigTable papers and everyone else in the industry followed suit. Of course, they had a huge competitive lead and didn't open source their findings right away.
Deepseek basically gave us the equivalent hardware reduction moment but also decided to open source how they did it from the get go.
With that being said, what's the key takeaway here?
Deepseek R1 lets the small players finally compete and form their own wedges.
This snippet from Daniele sums it up:
Deepseek R1’s arrival underscores a pivotal shift: open-source AI no longer demands exorbitant compute or licensing fees. By offering robust reasoning at a lower cost, it paves the way for broader adoption, from small-scale dev teams to large enterprises.
To be clear, this is also a huge win for anyone excited about decentralized AI. I'm not saying everything is solved now but there's some new found hope that the vertical's biggest bottleneck (centralized hardware clusters) might be getting smaller over time. Of course, the capital that the big dogs have is still going to pour into making their general models 10x better, but that doesn't mean distributed players can use less capital to focus on specialized, niche models.
This is a half joke from Ansem but I'm sort of with him on the acceleration of open source AI.
Anyways, this is where I'm at with all of the Deepseek stuff. Overall, I think the hype will die down pretty quickly until we start seeing cost reduction data from American players with the new Deepseek training method.
Unrelated to Deepseek, yesterday a token called VVV launched and hit a $350 million market cap in a day!
The $vvv token is for a product called AskVenice that has been around since May '24. AskVenice is basically a ChatGPT UI but focused on privacy.
Here's the rundown:
Venice makes it easy to use open source models for people who don't want to download from Hugging Face and run models like Llama locally
Your requests are encrypted and the actual conversations you have only exist in your browser
They host multiple open source models: Nous theta, Llama 3.1 405B, Dogge 70B, & a few others for image generation. Over the weekend, they added in DeepSeek as well
400k registered users & 50k DAU
Venice uses multiple decentralized GPU providers, so it's not possible for any compute provider to have your entire history***
The only thing Venice knows is your e-mail & IP address
Started and run by Erik Voorhees, Teana Taylor, and a small team
The token vvv is on Base and there's an airdrop for wallets holding Virtuals ecosystem tokens (i.e. aixbt, vader). Check the claim page here to see if you qualify (only 1 month left!)
Worth noting that Coinbase listed the token on day 1 of it being launched. Not something they usually do, could be a signal of them leaning into the new meta. Imagine if Kraken listed vvv before them, would be a major flop
Devs an access the API for $149 / yr which is the pro plan.
That's pretty much it. The product itself is simple - if you've used ChatGPT before then this isn't anything new. Rather it's the fact that the models you can pick are open source and uncensored + the key point that they're going for privacy first.
***BUT! I want to call out an important point in the marketing pitch that's a bit misleading.
This tweet below by Eric is a bit harsh but a really important callout in regards to the privacy part.
Notice how in the bullet points above, I mention the fact that Venice uses multiple decentralized GPU providers? Sounds great! But the key point worth emphasizing is that this means privacy in the sense that no one provider has the entire context on you. But, any single thread is available to the compute provider processing your request. So, if there's any sensitive information in one chat with Venice, that's fair game to be used against you in the same way that centralized players can.
Eric gives a solid counterexample of someone uploading their job offer letter that includes their name, salary, company, etc. that the GPU provider then has access to.
The only reason I noticed this caveat is because I actually went through the docs and found the mention in one of their blog posts. Venice's marketing needs to make that crystal clear for the average user!
In regards to the VVV token above, many people were upset about the fact that Coinbase listed the token on the first day it launched. Specifically, the community was calling out the fact that $VIRTUALS (which at one point hit $4b mc) hasn't even been listed yet so this clearly feels like Coinbase playing the "favorites game".
I agree with that take for the most part. Look, I'll be the first to say that as a public American company, Coinbase faces the most headwinds in terms of regulation restrictions. But the Virtuals argument feels like a legitimate counterargument. It's less about the fact that Virtuals hasn't been listed yet and more so that there's no transparency on how the process actually works.
But, it seems like the Coinbase team is taking this a lot more seriously as hypertokenization is only going to increase over this year.
Jesse and Paul Grewal (Coinbase chief legal officer) are going to have a live discussion on the topic this Thursday, so I'll keep you all posted on what happens there.
That's all for today's post!
See you all on Friday :)
If you've been enjoying my content, please please share TOC with your friends in crypto...really trying to get to 10k subs asap!!
- YB
Over 9.5k subscribers
It's not a geopolitical red flag but an open source green flag 3 upvotes, submitted by @timdaub.eth