Register and share your invite link to earn from video plays and referrals.

Jesus Rodriguez
@jrdothoughts
Co-Founder, CTO-CPO of @SentoraHQ (fmr IntoTheBlock), Co-Founder of @layerlens_ai, @neuralfabric( acq by Cisco) and The Sequence, Teaching at Columbia-Wharton
1.6K Following    6.6K Followers
AI models got smarter this week. The money got nervous. In the latest Sequence Radar, I look at both sides. Google’s Gemini Omni 1.2 Flash adds conversational video editing and stronger reference consistency, alongside a 30% price cut. OpenAI’s GPT-6 Cyber targets vulnerability discovery, validation, and patching for approved customers. Thinking Machines’ Inkling 2 Small combines image inputs and adjustable reasoning with 17B active parameters. Apache 2.0 weights are planned, not yet released. On the research side: provable LLM inference, formal verification of agent-written software, and evidence that careful context management can compete with much longer context windows. Meanwhile, the market selloff put infrastructure spending, financing, and value capture under scrutiny. The distinction matters: cheaper, more capable intelligence can be great for users without producing attractive returns for every company supplying it. Technical progress and investment returns are related. They are not the same thing.
Show more
this is the way. Every vault should have embedded loss protection
Loss protection is increasingly built into DeFi vaults that manage digital assets. Today, Lido Earn extends its own: Firelight can absorb eligible losses first, then remaining loss can be absorbed by the Lido DAO’s loss protection mechanism.
Show more
Another announcement today. This time from @SentoraHQ An institutional Earn program needs a simple interface backed by serious infrastructure. @SentoraHQ’s Smart Yield( just added two important pieces: embedded wallets and vault screening powered by @HypernativeLabs. Embedded wallets make yield easier to bring into existing products. Screening adds another layer of risk visibility around the vaults users interact with. Together, these upgrades are an important step toward institutionally ready Earn programs, connecting distribution with stronger tools for evaluating risk. The goal is to make sophisticated onchain strategies easier to access, understand and integrate into financial products at scale. We’ll be sharing more about where this is heading very soon.
Show more
Building a yield product in-house takes serious engineering, risk and compliance work. With Smart Yield Widgets, any app can offer Sentora’s @Morpho @upshift_fi & @veda_labs vaults in three steps, now with compliance screening by @HypernativeLabs .
Show more
Super excited about this partnership: Today, @Firelightfi brings onchain cover to @LidoFinance’s Lido Earn: EarnETH and EarnUSD. DeFi has gotten very good at compressing complexity into a deposit button. Underneath, the dependency graph keeps running. EarnETH puts ETH and stETH to work across established DeFi protocols. EarnUSD accesses USD-denominated strategies with transparent asset selection and risk controls. Protection extends to the vault contracts and the integrated protocols where they deploy capital. Eligible events include contract exploits, oracle failures, governance attacks, mechanism-driven bad debt and depegs, and redemption failures caused by protocol defects, subject to coverage terms and limits. Scope, price and capacity are registered onchain and refreshed every 30 days as allocations evolve. Capacity sits at the vault level. Exposure is monitored continuously. An independent Risk Consortium validates events. This connects two ideas I keep writing about: a vault is a risk graph, and serious DeFi infrastructure should make both deposited capital and covered capital visible. The vision for Firelight is programmable protection built into financial products. Defined failure modes. Visible capacity. Capital standing behind the cover. We already made yield composable. Now we need to make protection equally easy to integrate. That is how DeFi becomes infrastructure people can build lasting financial products on.
Show more
Just published an essay in The Sequence: Useful AI increasingly depends on matching the model, its inputs and its feedback to the job. I look at three recent releases through that lens: - @GoogleDeepMind's Gemini 4 Argon turns inputs like photographs into finished artifacts through tools and subagents. - Gemini 4 Carbon targets long engineering and cybersecurity work, where reliability compounds: over 50 steps, 99% per-step reliability gives about 61% task success, while 99.9% gives about 95%. - @MistralAI's Timestral brings zero-shot forecasting to numerical series, with quantile outputs that express uncertainty. The takeaway: pair specialist models and generalist agents through clear, checkable interfaces.
Show more
Sentora Research (@SentoraHQ) just published its incident report on the Binance hot-wallet exploit: On 29 October 2026, starting at 20:05 UTC, an attacker drained about $6.36 billion from @binance hot wallets across ten chains. Issuers responded within roughly 40 minutes: @tether, @circle and @Ripple froze affected stablecoins, and @aave froze its DAO-run V3 and V4 markets. As of 08:09 UTC on 30 October, about $2.8 billion had been frozen or seized, while roughly $3.2 billion, almost all in BTC, ETH, XRP, TRX and SOL, remained free to move. Sentora had no direct exposure to Binance. The impact on its vaults was indirect and showed up as liquidity stress rather than losses. Its two main @Morpho vaults, RLUSD Main and PYUSD Main, saw net outflows of 11% to 13% of assets within hours and met every redemption in full. Sentora-curated markets recorded one liquidation of about $35k, no bad debt, and no liquidation of any Sentora-supervised position. All @chainlink oracles behind Sentora's markets operated within their published parameters. USDe and RLUSD briefly traded below par and recovered within hours. Sentora continues to monitor potential sales of the attacker's unfreezable BTC and ETH, and is reviewing health-factor floors for supervised positions and per-market oracle behaviour.
Show more
I’m not a fan of the “every catastrophe is an opportunity” sermon. 🙄 If your strategic vision arrives after somebody else gets wrecked, perhaps it’s just a content calendar. The best thing to do in moments like this is remain humble and recognize just how much we still need to upgrade our on-chain infrastructure. 🛠️ But the activity following the Binance hack reinforces why we built Firelight: on-chain protection needs to exist before the panic. 🔥 During the on-chain chaos, @SentoraHQ’s protected vaults and @FirelightFi performed exceptionally well, demonstrating the value of infrastructure designed for precisely these moments. 🛡️ In interconnected markets, losses can travel. So can fear. 🌊 Protection against covered losses can help preserve confidence when everyone is suddenly reassessing what “safe” means. DeFi needs more than attractive yields and emergency group chats. It needs capital prepared to absorb shocks. 💪 Ideally, before somebody writes the inspirational post about the opportunity. 😉
Show more
Aave markets are now re-enabled for new positions. Withdrawals and repayments remained available throughout the incident. Aave was unaffected, and all positions remain overcollateralized with no bad debt.
Show more
Quick update: @SentoraHQ is not affected by the Binance exploit. Every Sentora vault has liquidity for withdrawals, and all Earn funds are safe and available to withdraw on the normal schedule. Our team will keep monitoring markets overnight and will share updates as we learn more.
Show more
An AI may have just knocked out one of computer science's load-bearing assumptions. Here's the part worth following. 👇 OpenAI says an internal model disproved the Exponential Time Hypothesis (ETH), a belief researchers have leaned on since the late 1990s. What ETH says, in plain words: take a logic puzzle called 3-SAT. Given a pile of true/false conditions, can you set the variables so every condition holds? The best known methods are only modestly better than trying every combination, where the work roughly doubles with each added variable. ETH bets that this explosion can't be escaped. What OpenAI says it did: its model found a disproof on Oct 16, and the team formalized it in Lean, a proof checker, on Oct 17. The write-up and the Lean files are public. Why it matters: 1️⃣ ETH is a foundation. OpenAI says hundreds of papers use it to show that problems in graph theory, biology, databases and geometry can't be solved much faster. If ETH falls, those "can't be faster" results lose their footing, and the question of how fast these tasks could get reopens. 2️⃣ It's not a practical speedup, and it doesn't settle P vs NP. OpenAI says turning this into better algorithms needs significant further research. Anthropic has since posted a similar claim from Claude, and its own paper calls the result purely asymptotic. 3️⃣ The review may be the bigger story. ETH's original proposers (Impagliazzo, Paturi) and MIT's Ryan Williams are examining it, unpaid and with no NDAs. They say their involvement is not certification. And Lean checks a proof, not whether the formal statement says what you meant. AI is now producing results that experts must check in the open. This is a test of how we do that. OpenAI: Williams:
Show more
An internal model at OpenAI has produced a claimed refutation of ETH! Russell Impagliazzo, Mohan Paturi, and I have been asked to help organize an open and independent examination of this claim. 1/4
hear me out...A VAULT IS NOT AN EARN PROGRAM
A simple money market vault is not an earn program. Proper earn programs use multiple vaults with multiple collaterals. They also don’t require a costly liquidity sleeve. If you are looking to add an Earn product where you can make a profit, dm me.
Show more
Proud moment for the @SentoraHQ team: @UpholdInc just launched RLUSD Earn, powered by Sentora. Uphold clients can now put RLUSD to work in an onchain vault curated by Sentora and built on @upshift_fi. Rewards accrue in RLUSD, and every position can be verified onchain at any time. One of Sentora's core principles is that DeFi yield should feel institutional. That means risk-managed, transparent and auditable, without asking anyone to become a protocol expert. Seeing it live inside a platform like Uphold is exactly the bridge we've been building toward. Stablecoins are becoming real money rails. The next step is letting that money earn, in the open, with risk managed by people who live and breathe it. And this is just the start. The roadmap for Uphold Earn is very ambitious. Stay tuned. Huge thanks to the @UpholdInc and @upshift_fi teams. Check out Uphold Earn:
Show more
Uphold has launched RLUSD Earn, powered by Sentora. @UpholdInc clients can now allocate RLUSD to an onchain vault curated by Sentora and built on @upshift_fi. Rewards accrue in RLUSD and the position can be verified onchain at any time.
Show more
The thing about AI is that the GPUs show up before the revenue does. Financing the future would be a wonderful business if the future would agree to arrive before the interest payment. So credit bridges the gap, and lenders are getting pickier about what they'll bridge. There's a CoreWeave bond yielding roughly 15%. There's a five-year loan against customer contracts averaging three years, which is a meaningful allocation to optimism. Three customers make up 72% of revenue, so a lot of different loans quietly depend on the same people paying their compute bills. Meanwhile Meta, a formidable cash machine attached to a slightly more formidable shopping list, is selling equity. SpaceX has a roughly $40B GPU lease commitment, which is committed financing, not $40B sitting in a bank account. The bear case doesn't need AI to be useless. It just needs the cash to arrive late. New in The Sequence Opinion, "The AI Credit Crunch" 👇
Show more
The AI demo gets the applause. The infrastructure gets paged. 📟 Behind every blinking cursor, GPUs crunch tokens, sandboxes run code, and some poor engineer makes sure the agent can't email the production database to a stranger. In this week's The Sequence Learning Loop, we pop the hood on last week's agent stack: ⚙️ DeepSeek's DeepInfer opens up its production inference engine, including DSpark, a scheduler that decides how many draft tokens are actually worth checking. That's a 4–7% throughput gain near production, plus a refreshingly honest regression at low load. 💤 Google's Agent Compute Engine gives agents computers that know when to sleep: checkpoint the workspace, free the compute, wake up right where you left off. 🛡️ NVIDIA builds the infrastructure around the whole agent, from running models to enforcing the rules. The big takeaway: the next gains in AI come from running models cheaply and trusting where they act. 👇
Show more
More about recursive self improvement in TheSequence today: Your AI coding assistant learns a new API mid-conversation and uses it perfectly. Then you open a fresh session, and it's gone. 🧠💨 Showing a skill in context isn't the same as acquiring it. If we want AI systems that improve through experience, the real question is how experience should change the model, and who decides. In today's The Sequence Knowledge, we dig into SEAL (Self-Adapting Language Models), which hands part of that decision to the model itself. It writes its own "self-edits," which are training data and update directives. It applies them, and reinforcement learning rewards whichever edits actually make the updated model better. The result in one knowledge setting: QA accuracy rises from 32.7% to 47.0%. Training on the raw passage gets you just 33.5%. 👀 We also cover: 🔹 Remembering vs. learning: context, external memory, and weight updates 🔹 Why a training example is really a choice about computation 🔹 What LoRA solves, and what it doesn't 🔹 Nested Learning and Titans: adaptation across timescales 🔹 Catastrophic forgetting, the tax every persistent learner pays The big idea is models that get better at becoming better. Read it here 👇
Show more
You know an S-1 is good when the company gets hacked and nobody cares 😂
Amazing to see that the NeuralFabric platform is powering within Cisco. So proud of that team:
.@HubSpot reported 306,446 customer accounts at the end of June 2026. Across those organizations, weekly pipeline reviews, sales follow-ups, and support queue handoffs depend on fast answers from the CRM. As teams begin connecting AI agents to HubSpot, the first questions they ask are the same ones managers ask before a forecast meeting: how many deals are open, or how many support tickets are still waiting. Getting those numbers right requires more than reading a row count. Every answer depends on a rule inside the records, such as which pipeline stages count as closed. If an agent misreads a stage condition, a sales team risks reviewing an inflated pipeline or overlooking an active queue. This week, LayerLens is testing how agents handle those record conditions inside a simulated HubSpot account. Each day examines one CRM question, the known answer written down in advance, and the saved tool calls that show how the agent got there.
Show more
@aave v4 is one of the most innovative DeFi protocols of the new wave. @SentoraHQ just released a very comprehensive catalog of risk signals for Aave v4: Monitor liquidations, whale activity, inflow and outflows and dozens of other signals. Give us feedback.
Show more
Excited about this
The largest lending protocol opening up for externally curated markets. @SentoraHQ Liquidity Hub is now live on Aave V4.