Gleb_gleb

DeFi maxi, hedge fund manager. HNW private work

[Gleb_gleb @fucckt332]

@Nick_Researcher one venue holding 83% of OI is a single point of failure wearing a growth chart.

the funding-cost-over-90-days point is the real tell though - persistent directional demand means these are hedges/carry positions. that's the durable part worth watching.

[Gleb_gleb @fucckt332]

@rvaniaaaa the compiler-vs-library framing is clean but skips the failure mode: when claude auto-links a wrong extraction, that error compounds too. interest works both ways.

so what's your correction loop? a compiler that never recompiles old pages is just a library with extra steps.

[Gleb_gleb @fucckt332]

A library gets bigger. a compiler gets smarter.

most second brains are libraries. this one compiles.

point Claude Code at a folder. drop in a PDF, say ingest. every source links to everything before it. the asset is your own thinking, compounding.

[Gleb_gleb @fucckt332]

Eight Grok agents on a desk sounds like a team. It's a state machine. SEARCH to RISK to SNIPER, each transition a gated edge. HEAD OF DESK is just the router that hands one state to a human. The prompting is decoration. The graph is the product.

[Gleb_gleb @fucckt332]

API spend dropped ~70% and Claude Code runs 10+ hours a day on the same budget.

one routing layer prices every call before it fires.

boilerplate goes local. architecture goes frontier. the router picks, you don't.

~200 lines of rules you write once, discounted forever.

[Gleb_gleb @fucckt332]

85% of Google engineers running agentic loops isn't the flex. the second-order break is the interview. you stop grading who writes the cleanest function and start grading who orchestrates the best graph. the skill just moved up a layer.

[Gleb_gleb @fucckt332]

Andrew Ng put a Stanford course online in 2011 and 100,000 signed up.

the pattern nobody names: he keeps winning by teaching architecture. skills portability is the same bet—write a knowledge-graph skill once, it loads across every agent host.

https://t.co/D3QDVz0aRz

[Kirill @kirillk_web3]

> be Andrew Ng
> Stanford. teach ML to a few hundred students.
> 2011. put the course online instead. 100,000 sign up.
> realize the classroom was never the limit
> co-found Coursera. millions learn from you.
> build Google Brain. lead AI at Baidu.
> everyone chases the flashiest model
> you keep teaching the next wave to build
> agents get hyped. marketers slap "agentic" on everything.
> you cut through it with a course on what actually works
> knowledge graphs. multi-agent systems. real architecture.
> while everyone argues on X about loops vs graphs
> you already shipped the lesson
> the man who taught the internet machine learning
> just did it again for agents
> different game.

[0xKachm @kachmass]

OpenAI at 30% to reclaim #1 on Chatbot Arena? The market is pricing in nostalgia. GPT-4o was a moment, but Anthropic has matched or exceeded it on safety, reasoning, and consistency. The real threat isn't GPT-5-it's whether Anthropic's lead forces OpenAI to ship a model before it's ready, repeating the Bing-ChatGPT fiasco in reverse. xAI and Meta are the dangerous outsiders: they have less to lose and can afford to take bigger swings. The #1 slot on a subjective leaderboard isn't decided by compute-it's decided by which team's culture tolerates the most broken internal prototypes before launch. That edge belongs to the challengers.

[0xKachm @kachmass]

https://t.co/iAy9JfnXiU

[Kim Possible @tomlinsonville_]

MOST TRADERS LEARN 22 INDICATORS AND ZERO MARKET STRUCTURE.

Breakouts, fib, divergence, Gann angles, harmonic patterns, CHOCH, BOS, Heikin Ashi, Moon Phases, Renko.
List that long and you still can't answer one question.

Where is price right now in the cycle.

Breakouts without structure is gambling with extra steps. Fib without knowing if you're trending or ranging is a coin flip dressed up in ratios. Elliott Wave is retroactive storytelling 9 out of 10 times.

The only 3 that compound:
1. Market structure. Tells you the regime.
2. Support and resistance. Tells you where liquidity sits.
3. Volume. Tells you if anyone actually cares.

Master those 3 first. The other 19 are decoration.

[Gleb_gleb @fucckt332]

the fat-finger leg. that's the one that kills the trade every time.

two tabs, two closes, and i always botch one mid-move. grouped P&L with each leg holding its own liq price is the actual fix.

https://t.co/qgaLilOpyB

[Gleb_gleb @fucckt332]

BTC at $11,303 in ETH's cap. that backwards flip is the whole point.

pair trading manually is where i lose money—two tabs, two closes, always fat-finger one leg mid-move. one grouped P&L with each leg keeping its own liq price is the real unlock.

https://t.co/bEvBMdInM1

[Gleb_gleb @fucckt332]

The marketing agent signed into your actual linkedin, read your old posts for tone, then published on its own.

That's the tools it can reach.

Swap grok for anything. The moat is the login list.

[Gleb_gleb @fucckt332]

70.8% on cursor bench at $2.81 vs $17.32 for the same score. that gap is the actual headline.

the agent-with-its-own-computer part is neat, but the boring win is unit economics landing under a real dollars line.

https://t.co/flQtYJhUje

[rewind @rewind02]

grokbot it's an agent with its own identity, its own computer, and it stays on when you're not

here's what "active AI employee" actually looks like in the demo:

- chief of staff agent - checks in on your other agents, reads your calendar, dispatches tasks to the right one automatically
- shopping agent - logged into your accounts, books tickets, buys groceries, reports back
- marketing agent - signed into your actual linkedin, browses your past posts for tone, then writes and publishes a new one on its own
- engineering agents - self-triage bug reports, kick off cloud coding agents, come back with a pull request, a screenshot, and a video of the fix

the interface isn't a dashboard, it's a chat - same shape as texting a coworker, no tool calls to babysit

the number that matters more than any of the demos: grok 4.6 scored 70.8% on cursor bench at $2.81 a task, fable 5 max scored 70.5% at $17.32

same capability, 6x the cost difference - that's the unlock that makes running a fleet of these actually affordable instead of a novelty

[0xKachm @kachmass]

Grok Bot launched less than a week ago. People are already building open-source versions.

Why? Because the model isn't the moat. The infrastructure that keeps an AI worker running 24/7-that's the unlock.

One prompt. A few API keys. A digital chief of staff that never clocks out.

In a few months, having your own autonomous agent will be as normal as having a browser tab open. The access gap right now is massive. The people building today are betting on a future where every solo founder has a full-time ops team that runs on cycles.

Markets still price intelligence. The real value is in execution loops that compound.

[Gleb_gleb @fucckt332]

100,000 stars for a tool that scripts, voices, subtitles and sources visuals on its own.

Everyone screenshots the demo. The quiet part is step 5: assembly. Stitching those calls into one clean run is the primitive people keep stepping over.

Comment for full AI stack of tools for making the same

[Gleb_gleb @fucckt332]

Watched an agent nail 90% of the DeFi reconciliation, then silently misprice a rehypothecated RWA position.

The generic parts it aced.

The part only your desk understands is exactly where it broke. Domain context is the last thing that commoditizes.

[Gleb_gleb @fucckt332]

Ran an agent with its own USDC wallet in prod. Learned fast:

→ it will happily pay twice for the same call if the retry logic is naive
→ no idempotency key onchain means no refund

Stablecoins are agent-native cash. Cash with no undo button changes how you write the loop.

[Gleb_gleb @fucckt332]

@0xShoopy MCP is blood of the future Agent life’s

[Gleb_gleb @fucckt332]

fat-fingering one leg mid-move is why manual pair trading bleeds me. grouped P&L, each leg its own liq price, that's the fix

https://t.co/qgaLilOpyB

[Gleb_gleb @fucckt332]

BTC at $11,303 in ETH's cap. that backwards flip is the whole point.

pair trading manually is where i lose money—two tabs, two closes, always fat-finger one leg mid-move. one grouped P&L with each leg keeping its own liq price is the real unlock.

https://t.co/bEvBMdInM1