

Jorge
9.5K posts




Cerebras is now running Kimi K2.6 – a trillion parameter model – in enterprise trials. At ~1,000 tokens/s, this is the fastest frontier model performance ever measured by Artificial Analysis @ArtificialAnlys.


BITMINE $BMNR BUYS 26,659 $ETH FOR ~$63M, NOW HOLDS 5.21M $ETH WORTH ~$12.11B










🇵🇹 A good time to finally launch portugalstory.co where we anonymously record the incredibly Kafka-esque experiences with bureaucracy and companies here Your stories are kept 100% anonymous If we want to improve things, good to record what isn't good!


Bridges have been the single biggest category of DeFi losses cumulatively. $2-3 billion in hacks since 2021. Unfortunately, much of the story of how we scaled DeFi over the past several year was through bridges. I'd estimate 35% of DeFi TVL today has some third-party non-native bridge dependency today (e.g. a Kelp + L0 style architecture). We adopted third-party bridges for two reasons: 1) we needed to scale via L2s because Ethereum L1 wasn't scaling and 2) L2 native bridges mostly sucked due to pre-zk optimistic rollup tech (e.g. 7 day withdrawal windows). Then we pretended the added complexity and dependencies didn't materially increase risk to DeFi. And so we adopted daisy chained assets like L2 rsETH as practically the risk equivalent to ETH on L1. And we wove this risk into the system. We'll learn from this. We'll price risk more effectively. DeFi will adapt. But this is a painful setback and there will be more to come if we don't minimize bridge dependencies. Scaling the L1 is a security priority. We can't build DeFi on rickety bridges.




