Snickersnack

[Almer Sad, CFA @Kutekians]

Kenapa Trump sering tiba2 keluarkan GOOD NEWS ketika UST 10Y yield naik/ada di area 4.3%–4.5%? COINCIDENCE? Saya rasa tidak.

Saya coba backtest 17 event tariff pivot Trump sejak Jan 2025.

Dari 17 kali Trump bikin "good news" (tariff pause, trade deal, soften stance):
• 12 dari 17 event (71%) terjadi saat yield ≥ 4.3%.
• Saat yield di atas 4.5%: 3/3 kali Trump langsung lunak, yield turun semua.
• Saat yield di bawah 4.3%: Trump santai, jarang ada konsesi.

Bukti paling jelas: 9 April 2025.
• Seminggu sebelumnya yield spike dari 4.0% ke 4.4%.
• Bond market mulai jual US Treasury besar-besaran.
• Hari itu juga Trump umumkan 90-day tariff pause untuk 75+ negara.

Utang US = $36 triliun.
Setiap yield naik 1% = tambahan $360 miliar/tahun bunga.
US harus refinance $9–10T utang tiap tahun.
Setiap 10bps lebih tinggi = +$9B biaya permanen selamanya.

BOND MARKET IS THE REAL BOSS.
Trump boleh bikin kebijakan apapun.
Tapi kalau bond vigilantes mulai jual, dia balik badan.
Yield 4.3%–4.5% bukan sekadar angka teknikal.
Itu batas toleransi fiskal US.
Manage your portfolio accordingly.

[Ronin @Ronin_Network]

Ragnarok Landverse: Genesis land mint is NEXT WEEK!

Start a new journey in Midgard ⚔️

RONers.

You’re seven days away from owning one of a thousand land plots in the iconic MMORPG beloved by millions of gamers worldwide.

Here’s what’s happening 🧵👇

[Ronin @Ronin_Network]

Tired of missing mints, trends, and launches?

The Ronin Wallet Discovery Page will help YOU keep up.

• Track upcoming game launches
• Follow trending tokens, NFTs, and dApps
• See highlighted community content

Explore our Discovery Page 👇

🔗 : https://t.co/kQOwsGHjjr

[Snickersnack @Snickersnack40]

@SpotZero_Game lfg

[email protected]

[Sabine Hossenfelder @skdh]

I genuinely don't understand why some people are still bullish about LLMs.

I use GPT, Grok, Gemini, Mistral etc every day in the hope they'll save me time searching for information and summarizing it. They continue to fabricate links, references, and quotes, like they did from day one.

I ask them to give me a source for an alleged quote, I click on the link, it returns a 404 error. I Google for the alleged quote, it doesn't exist. They reference a scientific publication, I look it up, it doesn't exist.

Happens all the time.

Yes, it has gotten somewhat better in the past 2 years in that with DeepSearch and chains of thought about 50-60% or so of the references exist. By my personal estimate currently GPT 4o DeepResearch is the best one. Grok in particular often doesn't include references even if asked. It can't seem to link even to tweets. It's hugely frustrating.

Yes, I have tried Gemini, and actually it was even worse in that it frequently refuses to even search for a source and instead gives me instructions for how to do it myself. Stopped using it for that reason.

I also use them for quick estimates for orders of magnitude and they get them wrong all the time. One thing they do save me time with is unit conversion and collecting all kinds of constants. You'd think though that this shouldn't take a 100 million++ LLM to get done.

Yesterday I uploaded a paper to GPT to ask it to write a summary and it told me the paper is from 2023, when the header of the PDF clearly says it's from 2025. I don't even know what the heck is going on there, but intelligence ain't it.

I sense that a lot of people now think knowledge graphs will fix the LLM-issue, but no, they will not. They cannot.

Even in the case that knowledge graphs would prevent logical inconsistency 100%, there are a lot of text-constructions that are perfectly logically consistent but have zero relation to reality.

Companies will keep pumping up LLMs until the day a newcomer puts forward a different type of AI model that will swiftly outperform them. On that day, it will become apparent that a lot of companies have been hugely overvalued. It will be a very bad day for the stock market.

[tong @tong0x]

https://t.co/E2aS4ZOHyp

[generational fumbler @genfumbler]

THE REI NETWORK THESIS

crypto’s good, new thing (s/o jez)

for better readability, I've posted this on substack as well: https://t.co/cPKQVq6ft0

After the crypto AI bubble popped in January, the consensus has shifted towards a flight to fundamentals and quality. Specifically, market participants are allocating mid-term capital to revenue-generating DeFi protocols (Aave, Maker, Ethena, etc.), but I believe that this philosophy should extend to all crypto projects establishing moats.

When analyzing the advancements and directions forward by traditional AI companies (especially given MCP + open-source tooling), it’s clear that the competition for moats are fought at either end of the AI stack. Companies are primarily aligning themselves at either the LLM level (e.g., OpenAI, Anthropic, DeepSeek) or at the consumer-facing product level (e.g., Cursor, Manus).

If speculative capital flows back into Crypto AI, I believe the biggest winners will be those focused at either end. However, the current landscape of launched projects primarily consists of vaporous LLM chatbots (with unique quirks) and frameworks/tooling for developers to build upon LLMs. There is no consistent value accrual (launchpad flywheels are reflexive in both directions), and there has been no significant innovations. With AI being commoditized, tooling offers no moat, but the underlying model architecture does.

REI’s Emergence

@ReiNetwork0x is a research organization addressing the limitations within existing language models through developing novel neural architectures that implement principles of biological intelligence. As defined by the team, this includes “persistent memory, adaptive learning and distributed information processing,” or aspects that cannot be replicated by the scaled statistical pattern patching of current language models. For more about neural architectures, Google’s Titans Architecture is a good place to start.

The current transformer architecture of LLMs rely on attention mechanisms with context windows that scale with quadratic complexity, rendering them computationally infeasible for memory-intensive tasks. In addition, (1) the compression from linear transformers can lead to performance sacrifices for efficiency and (2) existing models are inefficient in long-term memory utilization, lacking robust systems for both memory retention and management. Through significant testing, the REI team now believes that these limitations cannot be sufficiently resolved through CoT (Chain-of-Thought prompting) or MoE (Mixture of Experts), as both may increase hallucination presence and/or propagation. These bottlenecks necessitate architectural changes and pave the way for REI’s solution and moat: Core.

The REI Stack

In the team’s words, “Core [at the bottom of REI’s stack] is a cognitive base layer [‘powering’ neural networks like LLMs] that implements semantic memory [to dynamically incorporate historical context with the current state] and neural processing principles… This occurs through biologically-inspired neural processing layers, structured as a distributed system with multi-layered processing and integrated feedback mechanisms across components.” Core is currently closed-access for testing but will soon be available via their API.

Models powered by Core are housed within Catalog, a series of transformer models tailored for different applications. The REI team believes that the future lies in compact, specialized models which can also enable faster inference times for real-time decision making. The first model, Hanabi-1, is designed for financial prediction and has been open-sourced (untrained). Its current iteration will be accessible through Core’s API for enhanced efficiency. Without getting overly technical into the advancements made with Hanabi-1, its built-in assumptions allow for stronger inductive biases to process inputs of limited or newer data more effectively. Teased in Telegram but not yet officially announced, REI is considering opening an HL vault which places trades based off of Hanabi-1 alongside Core.

For everyone (including non-technical users), REI’s Factory will enable individuals to instantiate and continuously operate personal Core-based agents (REIgents) that evolve through a persistent understanding of interactions with the user. Core’s neural architecture allows for genuine adaptation, maintaining consistent behaviors and the development of user-specialized capabilities. The @unit00x0 version of REI is an outdated representation of cognition (also proof through the terminal), but will be updated upon Core’s public release.

The final component of the REI stack is the Oracle Bridge, a modular translational layer between off-chain arbitrary computation (by AI models) and on-chain operations. Through maintaining context awareness, the Oracle’s intelligence layer can handle raw inputs and maintain deterministic outputs for execution. The Oracle Bridge helps overcome the current limitations for agents to proceed with on-chain actions (or even for DeFAI interfaces): (1) the need for pre-defined LLM tooling and (2) LLM memory constraints imposed by blockchain data. REI can therefore operate as a bifurcated system, even leveraging EVM blockchains as an offloaded database for memory storage (see ERCData).

My thesis is that REI’s Core will be the cognitive engine underlying the next stage of AI agents, especially those navigating the blockchain.

Of course there’s execution risk, but after three months of discussions with the team, it’s extremely clear that their technical acumen far exceeds other builders in this space. Founder @0xreisearch possesses the vision and execution ability to reach their goals, along with a willingness to hear feedback for product improvement (as demonstrated when actively refining the quant terminal). REI hosts a balanced team operating in stealth, has years of runway (apart from the public donation event), and consistently overdelivers with each update. The next major update is the starter pistol: Core’s public release at the end of Q1. Additionally, REI has secured the corresponding HIP-1 ticker, which will eventually trade on Hyperliquid alongside the Base LP.

REI is the only crypto-investable company that is pushing the boundaries of AI research with a clear direction ahead to secure their moat, and the credits required for using Core, their trained Catalog models, REIgents from Factory and the Oracle Bridge presents the path for revenue. At ~10M MC, REI is poised for a significant repricing, both in relation to the valuations of other leading AI research firms and, even more so, when compared to every other AI-focused crypto investment opportunity.

This will be my first and last post for CT :)) To all the accounts that I’ve followed throughout the years—thank you for the advice, wisdom and laughs.

Disclosures: Everything expressed in this article is reflective of my own thoughts and should not be considered as financial advice. I have invested my own capital into this project and am not affiliated with the team. I have not been paid by the team at any point. My time horizons are longer than you can stay solvent, but I will sell when the time comes.

[Bloomberg Technoz @BloombergTZ]

Respons Presiden Prabowo Soal Trading Halt IHSG

https://t.co/l7TEZzw7Zi

[reisearch @0xreisearch]

Finally you can query Rei Quant. Remember that this is in alpha, there might be a bit of bugs here and there, so please leave some feedbacks in the discord and we'll get right to it.

[Wazz @WazzCrypto]

Just found out which part of the cycle we are on

[Watcher.Guru @WatcherGuru]

JUST IN: 🇺🇸 President Trump announces $500 billion AI investment, partnering with OpenAI, Oracle and SoftBank.

[Litecoin @litecoin]

Make Utility Great Again.

[Tree News @TreeNewsFeed]

[🌲] Official D.O.G.E. website is live featuring the $DOGE logo https://t.co/ZcW7lZBr2d

[Gojo Capital @GoonKingCap]

Traders win, believers bleed

The highest EV strategy for this bullrun has been to load 50-100k into a trench wallet and wait for the monthly runners to appear

September - Moodeng
October - GOAT
November - PNUT / Chillguy
December - Fartcoin
January - TRUMP

You could have been relatively late to any of these and still made multiples on your initials

[CZ 🔶 BNB @cz_binance]

A new era begins.

[Binance @binance]

From Trump’s huge memecoin debut to Doge’s journey to Mars - #memecoins are more than jokes; they’re humanity in action.

Dive into the stories that prove memes really can move markets.

Read more 👉 https://t.co/zjRK1VBbNO

[Rei @rei_labs]

Starting Strong with the 2025 First Biweekly update including our Q1 To-Do List

(in no particular order)

INTERNAL DEVELOPMENT
Core Tech:
+ API schema definition and internal endpoint (alpha v0.1)
+ Integration of CORE on demand
+ Infrastructure server provisioning for user access and architecture integration

Infrastructure
+ Oracle backed DeFi autonomous framework v0.3
+ Testing Quantum Resonance Memory Framework for high throughput platform (Discord, Telegram)

EXTERNAL ACCESS
Rei Quant
+ Finalized public (gated) alpha 0.5
+ Adding more analysis capabilities and charts

MVP.a
+ Finalizing UI for agents platform

Non-tech Updates
+ Introducing API/SDK v0.5 Model (Business model)
https://t.co/bpziO6jRtl
+ Article. Task Automation vs Cognitive Architectures: Two Agentic Schools of Thought
https://t.co/NzkvTQzHWt

Community Highlights
+ Crypto AI Landscape and value of core architecture (Rei Core)
https://t.co/x0P5onHtMp
+ ELI5 Orchestrated vs. Cognitive Systems
https://t.co/Mkkw3JckAV
+ Rei Network Deep Dive
https://t.co/QsVw2j3YaP
+ Live session with Nader Dabit Notes + Video
https://t.co/i5nWWw556P

[Bitcoin Magazine @BitcoinMagazine]

Memecoins are temporary; Bitcoin is permanent.

BUY #BITCOIN 💥

[Peter McCormack 🏴‍☠️🇬🇧🇮🇪 @PeterMcCormack]

Trump coin is embarrassing.

It is the opposite of what you voted for.

[Hoteliercrypto @hoteliercrypto]

Bitcoin ATH lagi.

Artikel 3 minggu lalu dg judul Bitcoin ATH lagi tertunda gara2 Pan Gosheng. Bilang stimulus tapi ga cair2, dan cairnya baru Today. Jadi jangan salahin analisa saya, Salahin Pan Go Sheng...

Ini WA nya 0813-3919-3847.

Wa buat staycation di St Regis Jakarta