-
Sharing what I read. Spans across different crypto verticals and sometimes macro too. Happy to discuss if you are keen on the same topics too! 🍪🥛 NFA. Views are my own. I’m just a cookie.
Bagel: New multimodal by Bytedance
https://x.com/dreamingtulpa/status/1926182453233738220?s=46&t=ywALktyM5ZCyt-zigN5Big
My guess on what features it will have
• Able to access GPT with voice command instantly
• Records meetings
• Idea prompts for conversations, or data points
This is a very fun and educational video to watch
https://youtu.be/Jp2Qgw44MsA?si=uskrbNZuJjlWxZom
How to earn Virgen points
https://x.com/octop3s/status/1926466201443512816?s=46&t=ywALktyM5ZCyt-zigN5Big
OpenAI o3 disregarding shut down instructions
https://x.com/palisadeai/status/1926084635903025621?s=46&t=ywALktyM5ZCyt-zigN5Big
AI Semiconductor Value Chain
Pretty good piece on TSMC's evolution, and goes on to talk about market risks and catalysts in the future
https://www.generativevalue.com/p/tsmc-the-index-on-technology
Not Boring: Everything is Technology
https://www.notboring.co/p/everything-is-technology?utm_source=%2Finbox&utm_medium=reader2
Infinite tool use
Talks about the concept of giving AI access to a comprehensive set of tools it needs for a niche / specific use case
Interesting as it takes away the need for huge context windows
https://snimu.github.io/2025/05/23/infinite-tool-use.html
Stateless infra - State synchronization through decentralized peer-to-peer node network
Makes it possible for real-time applications
Any action you take on an app can occur with low latency
https://x.com/MultisynqDave/status/1925028680390344952
• Composability: Taking agent memory and porting it elsewhere
• Glasses are a great way for AI interaction - Google’s Astra glasses has memory
https://youtu.be/nZtmmUQDzMQ?si=u3aWEOSVerdnVe8N
• Original content is going to be even more valuable in the future of vast AI-generated content
• Content creators should content gate, and make users pay for insights
https://youtu.be/WQf-eB2xSew?si=JJhFudyr2CXNUTv8
DualMintRWA: Onchain laundromats (amongst many other kind of businesses)
Essentially lowering the capital barrier to start these businesses
With RWA protocols like Goldfinch, it's already difficult to ascertain their credit risk
Expansion into other RWA like laundromats is likely going to be even more difficult, every new kind of asset you are looking to bring onchain has a different risk profile, requiring different pricing of borrowing risk etc.
But now my clothes can dry at 500ms speed (it's a joke)
https://x.com/blocmatesdotcom/status/1925145504251523273
Breaking the pre-training ceiling with new tasks and environments
https://x.com/MinqiJiang/status/1921176396228952253
veldt
the platform is pretty neat - you get a feed of the various articles and podcasts, together with a summary
interesting, keen to try this out - research agent
https://x.com/Saypien_/status/1925587320813629695
Brief overview of how to optimize the search engine for LLMs
Revolves around the concept of giving users the ability to specify multiple factors
https://x.com/jxnlco/status/1925563996494614901
AI initiatives in banks
https://frontierfintech.substack.com/p/81-beyond-the-hype-why-gen-ai-is
Xiaomi Su7 is cool af - looks so much like a Porsche
https://youtu.be/cIF4hvxQHZ4?si=hu9jHmGGXKpOMmEt
OpenAI's wearable device - but it's not to be worn, it's just in your pocket
https://x.com/EHuanglu/status/1926164610010259754
Spark (subDAO of MakerDAO) offering 4.5% stablecoin yield, running season 1 of their points program
Referral link: https://app.spark.fi/points/VYMUBN
DNET: Towards Heterogeneous Multiprocessing for Deep Learning
New player in distributed compute
https://dria.co/research/dnet:-towards-heterogeneous-multiprocessing-for-deep-learning
002 Loaded Tokens
Interesting - loading meaning into language - probably a good way for the less quantifiable and more qualitative tasks
https://latentreflections.substack.com/p/002-loaded-tokens
Wrote a summary - this piece helped me to piece together the thoughts I had in my mind over the past few months
https://x.com/jinglingcookies/status/1926303563640156232
Been thinking about real2render2real
1 human video to 1,000s of robot demos - human video can be taken using mobile phone
It's not as tedious as teleoperation yes, but scalability is likely dependent on the number of human videos being submitted
I believe that adding a crypto element here to incentivize people to share their videos would be rather useful
Question lies in the quality of the data being collected through phones, as well as the importance of the video (is it something that we want the robot to be able to do)
https://x.com/IlirAliu_/status/1924903627086266445
Products by Ellipsis Labs
1. SolFi - AMM
2. Phoenix - Orderbook
3. Gavel - Launchpad
https://x.com/blockworksres/status/1925546060472295506
Skywork: AI workspace agents - Creating a swarm of agents to handle different tasks, which can be presented in various output formats
https://x.com/Skywork_ai/status/1925522059326898412
InfoFi Landscape
https://x.com/ahboyash/status/1925540901772026191
Ideabrowser: Platform that shares an idea everyday, inclusive of details on the idea, and brief market research (supposedly to showcase how much potential the idea has)
https://x.com/gregisenberg/status/1925956955408888042
Understanding the Nuclear Fuel Value Chain
https://x.com/Melt_Dem/status/1925839284948320746
Absolute Zero Reasoner - where the AI creates and solves its own tasks to learn
https://x.com/jinglingcookies/status/1926203766585393360