BILLIONAIRE MARKET LEGEND JEREMY GRANTHAM JUST WARNED ON CNBC #BITCOIN# IS "USELESS, SPECULATIVE INVESTMENT" "IT WILL WITHER AWAY WITH A WHIMPER OVER YEARS AND DECADES" "IT IS NOT A STORE VALUE. IT JUST HALVED IN A STRONG ECONOMY. IT IS DIFFICULT TO USE TO MAKE SERIOUS TRADES." "WHAT THE HELL DOES IT DO? IT ALLOWS CROOKS TO MOVE MONEY WITHOUT A TRACE" 👀

Practices for embedding AI agents into enterprise systems [Async Job + Load Control] 💡 "Everything real-time" is a path to collapse. Turning long-running tasks into background jobs with priority queues gives you an agent platform that survives traffic spikes. 🔥 Problems solved - HTTP timeouts on agent tasks running tens of seconds to minutes - Spike traffic degrades latency for all users, causing cascading failures - Low-priority batch work starves real-time conversations of resources - Uncontrolled LLM call costs during traffic surges 🏗️ Proposed pattern Return a job ID immediately upon request and enqueue the task to a background queue. Stream progress via SSE/WebSocket and deliver results through webhooks or Slack callbacks. Use priority queues to ensure real-time conversations always come first while background analytics are deferred. Separate "online intelligence" (lightweight models for instant responses) from "offline batch intelligence" (heavy models running overnight), then let daytime agents reference precomputed results for instant answers. ✅ Selection criteria - Use when: Tasks exceed tens of seconds, high-volume parallel processing, spike-prone multi-tenant environments - Skip when: Conversational interactions completing in seconds (job overhead hurts UX) ⚠️ Pitfalls - Without a Dead Letter Queue, failed jobs silently disappear - Stale offline batch results can lead to wrong decisions if freshness is not managed - Missing per-tenant quotas let one runaway tenant degrade the entire platform 🛠️ Implementation Approach 1. Set up a message queue (SQS / RabbitMQ / Kafka) to accept jobs and return a job ID immediately via API 2. Use a workflow engine (Temporal / AWS Step Functions) to centralize job progress tracking, retries, and DLQ handling 3. Implement priority queues to separate real-time conversations from background work, with per-tenant quotas (Token Bucket / Sliding Window) 4. Stream progress updates via SSE/WebSocket and deliver results through webhooks or Slack callbacks on completion 5. Run heavy analytics overnight via batch pipelines (Airflow, etc.) using large models, storing results in Redis/DynamoDB for instant retrieval by daytime online agents #AIAgents# #EnterpriseArchitecture#

💥 GLM-5.3-Flash: Reported Opus 4.8-Level Scores at One-Tenth the Price Zhipu has confirmed that Ox Alpha — the anonymous model that dominated OpenRouter and OpenCode last week — is GLM-5.3-Flash, a 320B-parameter MoE with 18B active, released with open weights and an API at roughly one-tenth of GLM-5.3's price. Zhihu contributor 小小将 opens with a confession: he had hoped Ox Alpha was an external team fine-tuning the open GLM weights. He was wrong — it was Zhipu's own model all along. His broader take: at this performance and this price, Flash becomes the new "kill line" for large models — the bar below which rivals simply get priced out. 1️⃣ The numbers behind the "kill line" Flash is somewhat larger than DeepSeek-V4 Flash, at 320B total and 18B active. On Zhipu's own benchmark comparisons, overall capability is roughly level with Claude Opus 4.8. More reference points he cites: 🔹 57 on the Artificial Analysis Intelligence Index — around GPT-5.6 Terra's level, slightly below GLM-5.3. 🔹 No.5 on Code Arena, one spot above GLM-5.3. 🔹 API at 1/10 of GLM-5.3's price, with a limited-time half-off promo bringing it to 1/20. 2️⃣ Same DeepSWE score, a fraction of the cost The author's sharpest comparison is on DeepSWE, where Flash scores 63% at a single-task cost of $0.24. DeepSeek-V4-Pro hits the same score at $1.67 per task — roughly seven times the cost for equivalent results. This is the author's cost arithmetic on reported figures, not an independent measurement. 3️⃣ The architecture that cuts the bill Part of the price drop is structural. Flash uses a hybrid of linear and sparse attention, sharply cutting attention compute, and a new IndexPool that compresses the indexer's cache from four copies to one — reducing latency and memory at million-token context. Versus GLM-5.3, the company reports attention compute down to about 1/3 and KV cache down to about 1/4.4. 4️⃣ The compute mystery, answered by domestic chips One reason few believed Ox Alpha was Zhipu's: the company was not thought to have enough spare compute for a massive free public test. The answer, per Zhipu: Flash is served from domestic Chinese chip clusters. The team built a custom inference engine on SGLang and worked around limited VRAM and bandwidth with quantization, layered deployment, and trading compute for communication — lifting end-to-end serving performance about 3x on the same hardware, with per-token cost now close to mainstream NVIDIA GPUs. 5️⃣ Chinese silicon just passed its biggest stress test The author's conclusion: Chinese chip clusters have shown they can carry large-scale inference for a frontier-level model — including a free, record-breaking public trial. If that holds, he argues, the outlook for Chinese models just got a lot brighter. 🔗 Full Reading: #GLM# #Zhipu# #GLM5Flash# #AIChips# #LLM# #AIInfra# #OpenWeights#

🚨SlowMist TI Alert🚨 💸 @usmfum Loss: ~70.83 ETH 🔍 Root Cause: Pricing logic flaw in `ethFromDefund()` of USM's `defund()` function. The function uses arithmetic mean of current and estimated final FUM sell prices for single redemptions, but lacks "split invariance." Combined with per-redemption state contraction (`adjShrinkFactor`) and integer rounding, 64 small `defund()` calls return more ETH than one large call for the same FUM amount. - Attacker EOA: `0xb92b2e47680c89da8f951b8963ef469f461a50fc` - Attacker Contract: `0x5a5e29ba89663a3558273354e990426f3cac7de7` - Victim Contract (USM): `0x2a7fff44c19f39468064ab5e5c304de01d591675` - Profit Receiver: `0xe3c6346b6f282029312d2caf4677ef39beabbf99` Powered by Tx:

Colman’s wasabi-style punch made the mustard a British institution. Now the farmers who protect its seeds hope new American owners don't let the brand wither. Read more: 📷️: Getty Images

agent payments need two kinds of guardrails working together. one controls how much value can move, and ampersend handles that with per-agent wallets, configurable limits, and topup schedules. the other controls where the money goes, which has been the missing piece across the whole agent commerce stack. the first guardrail has been in ampersend since day one, giving full control over economic activity per agent. last week at @consensus2026 we showed a live demo of the second guardrail we're building. think about what happens without that second guardrail in place. a counterparty wallet on a sanctions list or tied to ransomware will accept a payment within whatever spend cap the sending agent has, and the compliance team finds out after the fact when the transaction shows up in a screening log. we built compliance screening with @trmlabs and @chainlink ACE. real-time risk intelligence on every counterparty, policy enforcement onchain through ACE, decisions executed inside the ampersend SDK before any value moves. landing in ampersend soon. stay tuned.

last week at @consensus2026 in miami: we showed the live demo of compliance enforcement on autonomous agent payments, first time live, with @trmlabs and @chainlink. here's the unpack 👇 what we demoed two transactions, screening enabled. badclaw, a wallet associated with tornado cash. blocked with a 403 forbidden the moment it hit the screening layer. no funds moved, no human in the loop, no rejection details disclosed so adversaries can't reverse-engineer the logic. goodclaw, a wallet with clean defi history. screened against the same intelligence layer, came back at acceptable risk, transacted normally. one toggle inside the dashboard: every decision logged with risk scores, categories, attributed entities, offending addresses, and timestamps. why this matters agent guardrails have been a half-conversation so far. spend limits cover how much your agent can move. ampersend has handled that since day one with per-agent wallets, limits, topup schedules, and a dashboard for monitoring activity. that's half the problem. the other half is who the agent is transacting with. a clean spend limit doesn't help when the recipient is a sanctioned entity or a ransomware operator, and no regulator is going to be satisfied that the dollar amount was capped. compliance screening closes the second half. spend controls plus screening is what defensible agent commerce actually looks like. why three companies policy, intelligence, enforcement. three separate jobs. 1️⃣ @chainlink ACE writes the policy onchain where regulators can verify it directly. 2️⃣ @trmlabs returns risk intelligence on the counterparty in under 500ms with fedramp high authorization. 3️⃣ ampersend executes the decision before the transaction broadcasts. splitting it across three companies is the point. no single party can weaken the rule, fudge the intelligence, or skip the enforcement. what's next screening ships into ampersend as a fast follow for enterprise customers. roadmap extends to buyer-side gating, world id proof-of-personhood on counterparty owners, and kyc-verified counterparties through chainlink ACE and sumsub. huge thanks to the @trmlabs and @chainlink teams. get started with ampersend:

.@world_xyz is officially live on Solana. Here’s what else is trending on Solana in the last 24 hours 👇 Consumer & Community - @PumpFun deprecated its tokenized AI agent launch option following community feedback around PvP dynamics. - @PlayKintara integrated MetaMask Connect, allowing MetaMask users to access Kintara across desktop and mobile. - @World_XYZ launched its prediction market on Solana through Phantom, powered by Chainlink oracles - @ChimpxAI FIFA season campaign still on Infrastructure & Ecosystem - @Backpack secured MiCA and Payment Institution licenses from the Bank of Latvia, adding to its existing MiFID II license. - $1.25B USDC was minted on Solana. - Solana network fees reached a 30-day high, rising approximately 60% over the past month. - @SanctumSo has now grown SOL TVL for 11 consecutive quarters. AI & Builders - @ExaAILabs, now live on Solana, enabling autonomous agents to perform web and content search without accounts or API keys, with per-request USDC payments via x402. - @Nika_Finance opened NikaAI to all users, introducing “Trade by Words” for perps, stocks, ETFs, equities, and indices. - @Orb_Markets announced the merger of Lana into Orb. DeFi & RWAs - @ObligateCom launched OTFY, a $200M trade-finance RWA on Solana, bringing fully composable institutional collateral into DeFi lending markets. - Drift Protocol rebrands to @VelocityDex - @Project0 partnered with Titan Exchange to simplify PT looping, enabling one-click looping across Project Zero, Kamino, and Jupiter. - @Sunrise reported that $SNDK has surpassed $70M in onchain trading volume since listing last week. - @UseDiversifi launched its Transformed HODL Strategy on Solana.

After reading this new essay in The Conversation (left), I am convinced these people are either stupid or consciously deceptive. There are no other explanations. Ignorance is no longer possible. Their entire argument is: "The traits of males and females has some overlap." And? Show me one person who has claimed otherwise. Nobody ever engages with the actual arguments people like me are making when we claim "sex is binary." I've made my case countless times in excruciating detail in major news outlets, massive podcasts, and leading academic journals. So have others. My scholarly article "Why There Are Exactly Two Sexes" (right) has now been viewed over 73K times. Anyone paying even a little bit of attention knows it exists and what it argues. It directly addresses and debunks the "multidimensional" model forwarded in the new essay. Yet these activist scientists refuse to engage with any of the actual substance. They just keep arguing against strawmen. They are of course free to write up a scholarly rebuttal in the same journal I published my paper. I'd even agree to a series of exchanges in a popular outlet. Or a live dialectic via Zoom or in person. It's 2026 and we're winning the argument and shaping public policy. At some point (i.e., now) these people will need to engage with the substance of my arguments or slowly wither away to irrelevancy. There are many avenues available to them if they actually want to engage. My DMs are open. Top journals and major news outlets are increasingly willing to publish the exchange. Many large podcasts will gladly host the debate. Yet they invariably choose to write and speak only to audiences that already agree with them. It's pathetic. They're completely out of excuses.

# Hermes Agent Features and Practical Usage 🚀 For when you don't want the conversation to stop while the agent works. The modern TUI offers non-blocking input, mouse support, and rich overlays so you can watch long tasks and drop in extra instructions. 📌 Title and Feature URL Title: TUI (Ink terminal UI) URL: 📝 Overview The TUI is a modern terminal UI for Hermes, launched as a subprocess from the Python CLI. It is Node.js-powered (≥20) and shares the same Python runtime, sessions, and slash commands as the classic CLI, while delivering a cleaner, more responsive experience. 🔧 How It Works ・It paints an instant first frame for a non-frozen startup, uses differential updates to prevent flicker during streaming, and provides rich modal overlays for model picking, session selection, and approval prompts. ・Non-blocking input lets you queue messages before the session is ready, and mouse selection uses uniform background highlighting instead of terminal inverse video. ・It renders LaTeX math, both inline (`$E = mc^2$`) and block (`$$...$$`). ・A status line narrates state (starting agent… / ready / thinking… / running… / interrupted) plus working directory with git branch, elapsed time, auto-compression count, active `/background` tasks, and a YOLO warning badge. 🛠 Practical Usage ・Launch with `hermes --tui`, resume the latest with `hermes --tui -c`, resume a specific one with `hermes --tui -r `, or force classic CLI with `hermes --cli`. ・To default to it, set the env var `HERMES_TUI=1` or `display.interface: tui` in `~/.hermes/config.yaml`. ・TUI-specific slash commands are plentiful: `/model` (picker with per-provider cost hints), `/sessions` or `/switch` (live session switcher), `/skin` (live theme preview), `/details` (toggle tool detail), `/agents` or `/tasks` (observability overlay with a subagent tree), `/mouse`, and `/reload`. ・Press `Ctrl+X` to open the live session switcher and run multiple concurrent sessions within one terminal. 🎯 Use Cases ・Watch a long task progress in the live panel while dropping in mid-run instructions via non-blocking input. ・Dispatch and orchestrate many concurrent agent sessions from a single terminal. ・Survive SSH disconnects by auto-resuming with `HERMES_TUI_RESUME`. ⚠️ Caveats ・Node.js ≥20 is required (verified by `hermes doctor`), and the first launch installs TUI Node dependencies into `ui-tui/node_modules` once. ・A TTY connection is assumed; pipes or non-interactive environments fall back to single-query mode. ・It cannot attach to a remote gateway (the TUI spawns its own in-process gateway). Light-theme detection relies on `COLORFGBG` or OSC 11 probes, which not all terminals support. #HermesAgent# #DevTools#