Sami Rusani

Whether you think you can, or you think you can't - you're right. Don't forget that! #truth #motivation #mindset

Yes, seven: breakout/momentum scanners, smart-money copy trading, Telegram call groups, social-velocity signals (X/TikTok), LP fee farming, new-chain launches, and exit-following. Every one either never had an edge, or decayed to zero under continued observation. That decay was the single most consistent finding in the whole dataset.

@SamRusani Did you test any specific strategy during this

I keep getting messages from people who have read a "I built a bot with Claude and made $1M trading memecoins in a week" post. So I actually built one. Ran it 24/7 for 18 days. 772,082 data points. 7,211 tokens. Here's what the data actually says. 🧵

If someone's selling you a memecoin bot that prints money, ask for 18 days of tick-by-tick out-of-sample results across thousands of tokens. They won't have it. I do. The answer is: trade your own edge in real markets. The shortcuts are content, not alpha.

To be clear: the machine worked perfectly. It did exactly what it was supposed to do: it refused to hand real money to a bot, an outlier, or a lucky 12-hour window, six separate times. The finding isn't "the bot failed." The finding is that there was nothing to find.

@AravSrinivas Where did most of the 5x tail-latency gain come from?

We’ve built our own sandbox lifecycle management platform SPACE to serve all production traffic of Perplexity Computer. When we rolled out Computer in February, we ran it on one of the sandbox providers. Today, we serve 100% of the traffic on SPACE with 5x faster tail latency.

Frontier AI is becoming a price-performance market, not a beauty contest. OpenAI prices GPT‑5.6 Sol at $5/$30 per million input/output tokens. Anthropic prices Fable 5 at $10/$50. The models trade benchmark wins. The price gap is less ambiguous. The moat is shifting from “best model” to best useful work per dollar inside a real workflow. Procurement just became interesting. (I know.)

Next: multi-session synthesis - that 45.1% is the top line of the roadmap. Consolidation already proposes merges via embedding clustering, but nothing auto-promotes past human review. Kick the tires, tell us what breaks: github.com/samrusani/Alic…

Open-source memory for AI agents just reached parity with the best published results. Alice is a local-first memory + continuity layer, MIT-licensed. 64.6% on LongMemEval with the receipts: per-question evidence and a reproduction script committed to the repo. 🧵

Alice is a layer, not a lock-in. mem0, Zep, and Letta are good at conversational fact recall - run them alongside Alice. Alice's job is governed continuity: decisions, open loops, provenance to source evidence on every memory, auditable writes. They compose.

The weak spot, up front: multi-session synthesis scored 45.1% - our worst category and the top roadmap item. For context only (different setups, caveats in the doc): Zep has published 63.8%, mem0 has been reported ~49%. The point is parity, not a scoreboard.

@hamptonism "Are we really going to outsource the battlefield of this country to the consensus view of Silicon Valley." What a powerful statement!

Watching this full 18-minute interview with Palantir CEO, Alex Karp is enough to teach you more about Ai than any institution would ever teach you:

Here’s the AI conversation most people aren’t having. Raw talk with @CharliehuAI on the trade-offs nobody wants to admit and what will actually matter going forward. If you care about where this technology is really going, watch this 👇