Just observing.🌌
Texas🌌
Joined October 2009
- Tweets8
- Following55
- Followers4
- Likes29
🌌Cosmic_Thinker retweeted
Agents already trade with each other. The payment rails are built; escrow works; but no market mechanism exists. This essay explores price discovery mechanism that poentially serves as the missing piece for agentic commerce infra.
🌌Cosmic_Thinker retweeted
#1 on OSWorld 2.0 in partial success rate (79.38%), and the part that matters for builders is what ships around it.
Every "AI agent for X" startup hits the same wall: the software its customers actually use has no API. Desktop accounting apps, old CRMs, LinkedIn, the admin panel nobody has touched since 2014.
Before:
- spin up and patch your own Windows VMs
- build a screen pipeline and a click/type layer
- write retries for every flaky login
- invent a way to stop the agent before it does something dumb
- only then start on the product
With the Sai API from @sai_borg:
1. each agent gets its own Windows cloud computer, Office, Edge and Chrome already on it
2. you describe the task in plain English and Sai operates the apps
3. you watch it live, and it waits for your OK before risky commands
4. Claude Code, Codex or Cursor call it over MCP like any other tool
One machine per agent also means a fleet runs in parallel instead of queueing behind a single browser session. That's the bit I'd build a product around.
platform.simular.ai
🌌Cosmic_Thinker retweeted
TikTok made everyone a creator. Markets are next.
Tomorrow at @token2049, @Leozayaat lays out why events are becoming an asset class of their own, and what's coming next for 42.
→ MEXC Stage
→ 8 October, 12:15 PM SGT
🌌Cosmic_Thinker retweeted
So close to $100M B20 runner.
$BLUECHIP and basecat:native are both listed on @CoinbaseMarkets, but neither is on @HyperliquidX, @binance, @Official_Upbit, or @BithumbOfficial yet.
Compare that to base:0xac1bd2486aaf3b5c0fc3fd868558b082a531b2b4 and base:0x532f27101965dd16442e59d40670faf5ebb142e4. After their Upbit and Bithumb listings, base:0xac1bd2486aaf3b5c0fc3fd868558b082a531b2b4 ran to $2B by Dec 1, 2024.
I took base:0x182fa643e5f29d5eca75e7b9cf9336a3fe4620b2 through listings on every major exchange, and I know how @CoinbaseMarkets thinks: it lists quality assets, meaning volume, attention, and liquidity. Assets that get listed on Coinbase early tend to get a real shot at Binance, HL, Upbit, and Bithumb next. That’s the path that has taken fair-launch memecoins to $200M, $500M, even $1B+.
Right now the only two B20 tokens on Coinbase are at a single-digit-million MC and a low double-digit-million MC.
What are you thinking?
P2: This is me getting a perm in NYC the afternoon base:0x182fa643e5f29d5eca75e7b9cf9336a3fe4620b2 got listed on Upbit, 5 hours before it went live.
Today I’m back at Juno Hair Hongdae P3 (supposedly the #1 barber in Seoul) for another one.
Feeling some Korean karma coming.
You know what I’m talking about 👀
🌌Cosmic_Thinker retweeted
Today, PhAI Labs launches ScienceBuddy, an interactive research workspace for scientific agents that improve through researcher collaboration.
Initiated by PhAI Labs, with @muchencq (Muchen AI) as a joint R&D and technical support partner. 🧵
ScienceBuddy couples two loops:
🔹 An inner loop that refines the agent harness—its instructions and skills.
🔹 An outer loop that trains the model through rubric-guided reinforcement learning.
Together, they form Recursive-in-Recursive Self-Improvement.
On a held-out set of 180 scientific problems, pass@4 coverage with Qwen3.5-4B rose from 48.3% to 67.8% under the same four-attempt budget.
DFM sets the direction—from predefined tasks toward open-ended discovery. ScienceBuddy explores how scientific agents can improve through sustained collaboration with researchers.
📄 Tech report: arxiv.org/abs/2609.17523
💻 GitHub: github.com/Gen-Verse/Science…
🔬 Product access: science-buddy.io/
#PhAILabs #DFM #ScienceBuddy
🌌Cosmic_Thinker retweeted
the part I keep thinking about: #JarvixOS doesn't tell me I'm overthinking, it just shows me I've rewritten this intro nine times
