toly 🇺🇸 mert | helius.dev Thoughts? 🍿 "These tests seem to confirm our basic understanding that Solana does not fully achieve consensus. In this paper we show how a single malicious validator, once elected as leader, might be able to halt the Solana blockchain."

2.5 Pro Deep Think gets an impressive score on 2025 USAMO, currently one of the hardest math benchmarks. It also leads on LiveCodeBench, a difficult benchmark for competition-level coding, and scores 84.0% on MMMU, which tests multimodal reasoning. #GoogleIO

googlers are going to be so unbearably smug when everyone realizes the best way to use ai coding agents is in a large monorepo with all dependencies vendored into third_party and very strict coding guidelines. y'all better get ready to learn bazel

Both Claude 4 models are state-of-the-art on SWE-bench Verified, which measures how models solve real software issues. As the best coding model, Claude Opus 4 can work continuously for hours on complex, long-running tasks—significantly expanding what AI agents can do.

Only considering Hyperliquid’s open interest -Exponential extrapolation using the compound growth rate over the past 6 months: Hyperliquid would surpass Binance in just 4.5 months -Linear extrapolation using the slope from the past 25 days: Hyperliquid would surpass Binance in

the last week of launches has highlighted a few things for me: - Progress in LLMs has been amazing but incremental capability gains are clearly closer to log than linear while the corresponding cost for those gains is closer to exponential than linear. scale still works but the

If agents are inevitably going to replace apps, then SaaS companies will evolve into “curators of domain-specific ground-truth data” / the Data Protocol Layer. Providing high-quality, structured, domain-specific data thus becomes the core value of SaaS. The new SaaS stack can be

Hyperliquid's dominance is clear • HyperEVM is nearing $1B TVL with 20k+ daily active addresses. • Unit is clocking ~$600M+ in weekly spot trading volume. • Protocol fees have topped $70M/month. • Reached $78B ATH in weekly perps volume. • Leading 71% of DeFi perp volume.

Is an encrypted and decentralized environment conducive to post-training? -Post-training is the primary factor for models to differentiate themselves. -Post-training intensifies the competition for high-quality data sources; high-quality data is superior to RL Reasoning. -Models

AI products must be built around humanity’s primary driving forces—the core motives that propel us. These include, but are not limited to: survival and health, emotional connection, gaining skills and recognition, self-improvement, turning ideas into tangible value, uncovering

Introducing our first free report of Macro & Markets. This extensive report covers: • Our BTC Top Signal Model and what it's showing • Why miners face structural pressures despite strong prices • How tariff policies will reshape crypto markets • Why Bitcoin ETFs are surging

USDhl lending is live on Felix with a 1M Cap. Earn borrower interest and HYPE rewards. Lend USDhl: usefelix.xyz/vanilla-earn

Our engineering & research team put together a deep dive on the multi-agent system that powers our Research capability in Claude[dot]ai; lots of fun details architectural diagrams, and prompting learnings here: anthropic.com/engineering/bu…

x.com/i/article/1933…

🚨 New market map: AI video generation So much has happened in this space over the last few months - it's hard to keep up with all the new models. This covers text-to-video and image-to-video companies + those that make characters talk. Will be on TBPN later to discuss!

Introducing 🧵 dAGI Summit - October 24, San Francisco Digital feudalism or cybernetic economy. A cybernetic future worth building. Here’s why we’re gathering ↓

-For professionals, the essential skill in the AI era is learning to symbolize their knowledge and experience. -For AI, the challenge is to transform these symbols back into applicable experience, either to be imparted to humans or to augment/replace human labor. -For general

Data is one of the holy grails for AI. At Delphi Ventures, we’re actively backing this thesis 🤖 - Grass: historical and go-forward public data at scale - ORO AI: proprietary user data from every top app - 375ai: linking the real world to AI via stationary cameras

A broad ecosystem will only emerge from open-source models. The closed-source path does not have ecosystem breadth and diversity, and closed-source models fundamentally don't care if a diverse, open application ecosystem exists on top of them. Their primary purpose is to lock in

Building Agents with old, complex software paradigms—or creating them to be like Apps—is a path to failure Fixed functionality is the essence of an App, while the essence of an Agent is being goal-driven and dynamically adaptive Generative AI is essentially a database; the