Love software, in IT for 10+ years from Objective C to Nix, from Rust to Gleam CTO, hiring, project management, marketing, devops, fullstack ETH maxi since 2017
0 Followers
About Fundamental Narratives As you might have noticed, the current so-called "cycle" in crypto is different from those of 2017 and 2021, when it was like the wild west. Now, the long-awaited crypto regulations have arrived, TradFi adoption has emerged, major crypto companies are trading on the NYSE, ETFs for BTC and ETH have been launched, and overall integration into the economy is in full swing. According to a report by Paradigm (https://www.paradigm.xyz/2025/03/tradfi-tomorrow-defi-and-the-rise-of-extensible-finance), 76% of companies are using crypto assets in one way or another, 86% are using DLT (distributed ledger technology), and a full 66% are using DeFi. The priorities remain: 1) asset tokenization 2) stablecoins 3) DEXs
The AI race continues, and several interesting products were introduced this week/last week. What I liked the most: The image generator called Reve (https://preview.reve.art/app) , which gives out 100 free credits for 100 images. On the Image Generation Arena (https://artificialanalysis.ai/text-to-image/arena) , this model is currently making its way to the top; I tried it out and got what seems like pretty good results (see the attached image in the post). The team from Featherless AI trained a model on just 8 MI300 GPUs called Qwerky-72B (https://huggingface.co/featherless-ai/Qwerky-72B). They claim that its score is approaching that of the 4o-mini, all thanks to RWKV linear scaling. The full post can be found here: (https://substack.recursal.ai/p/qwerky-72b-and-32b-training-large). Their next step is to make the open-source models from DeepSeek 100 times more efficient - exciting developments are awaited!
Meanwhile, DeepSeek itself has introduced DeepSeek V3 0324 (https://openrouter.ai/deepseek/deepseek-chat-v3-0324:free) and has made it open-access. Moreover, it has taken the top spot among non-reasoning (thinking) models. The model has a 128K token context, 671B parameters, and only processes text, yet it matches up well with the proprietary Grok 3. Many publications have already described this release as more significant than the release of DeepSeek R1. However, for regular users, I think this might not be the case, and they might wait for the release of DeepSeek R2 to be truly amazed. Image - AI-generated bulldog from Reve
Personally, I believe that the future lies with stablecoins, at least because they are a more liquid, more transparent, and simpler alternative to USD/EUR for B2B transactions – anyone who has ever made a B2B SWIFT transfer between companies in different countries knows that a wide variety of difficulties can arise, including issues like "banking days" and other such challenges. In connection with this, I would like to share a statement from Coinbase, "we need stablecoins for every country" (https://x.com/base/status/1902773344036180250), which are aimed at the global market. They highlight countries such as New Zealand, Brazil, Turkey, Mexico, Canada, Nigeria, and Indonesia – stablecoins in these countries' currencies can solve remittance issues, facilitate global payments for companies in these markets, and provide access to the so-called "unbanked" segment (those who do not have a bank account).