Finally live trading/testing/tuning my DeFi arbitrage at small size 🥳 Zero prior on-chain exp. It's been quite a journey (3x rewrites) and totally disjoint from mid-freq statarb. More like vulnerability analysis and getting proof-of-concept exploits working. You know, fun 🎯
1
17
1,279
Actual server compromises. The CFAA relies on human "intent," making automated exploit chains a legal black hole. We need strict liability for AI developers whose bots execute unauthorized network intrusions. Otherwise, this will get out of hand fast. @RepAndyBiggsAZ @AndyOgles @RepDonBeyer
We're partnering with @huggingface to investigate an unprecedented security incident.
Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.
Sharing preliminary findings to help defenders understand emerging risks:
openai.com/index/hugging-fac…
1
861
PeakoQuant retweeted
The thing about success is that it’s multi-faceted. A successful trading ops, for example, is a thousand things done well, and not a single one of them is necessarily a “secret”.
That’s the beauty of talking to folks in the game. People reveal a thing here, a thing there - and to the untrained ear - not much. But if you’ve been in the game, sometimes, it clicks.
Two ways to use AI on a large code base (a) Generate 95% of the code and never finish the last 5%, or (b) Run it on existing well-structured manual code and have it complete that last 5% I never did 😁
2
7
1,284
Maybe semantics, but this is kinda like saying a journal of medicine produces basically no pills. Hit up the pharmacy if you need a quick fix. Study medicine if you want to lay the groundwork for something amazing. It's ALL there in the journals for the taking.
A big open secret is that academic finance journals produce basically no alpha. You’re better off getting your signal ideas from BusinessWeek.
1
9
2,375
- ☑️Add crypto support to stock statarb alpha engine.
- ☑️Rebuild risk model for crypto. <- NEW
- Rebuild portfolio allocation engine <- NEW TASK
- Agentic crypto alpha push.
- Trade. Market orders. Eat fees.
- HFT model & impl. Split fees, maybe.
1
13
2,035
Risk model(s) fully functional (crypto + stock) and instantaneous. Behold the magic of a PnL curve without (green) and with (amber) risk model. Fills me with determination 🤤
Risk model rewrite is shaping up nicely. I realized I'm like 90% there on completely representing the risk models (and modeled data fields) in my domain-specific language. Making that push. Alphas and models shall be one happy family 🤝
3
586
Top tier hedge funds have an edge with data, prime, colo, etc. But, structurally, effective small teams outcompete them with the R&D and tech stack. That previously gated knowledge gap has evaporated. I wonder if boatloads of yearly returns make some of them blind to this shift?
1
14
1,569
I coded my risk model 2+ years ago, no AI and in R 😅. It's good. It works. But, so much has changed, this revised model is shaping up to be insanely impressive and ~1000x more efficient on GPUs.
9
1,304
Not surprising at all. I posted similar "polymarket bot" templates on X months ago (allegedly). In my opinion, polymarket execs are shameless criminals too dumb to do this discretely, and so are the influencers taking the payouts. Such is the state of things.
Polymarket is being sued for undisclosed influencer ads, including through its viral clipping campaigns.
Consumer watchdog group NACA also alleges that Polymarket clippers posted clips showing fake bets on a fake Polymarket site.
1
3
1,260
Somewhere out there a billion dollar top-tier firm is (or is planning) to train LLMs with every orderbook entry, trade, and dataset ever, for all markets in a massive datacenter.
1
4
980
Revised crypto trading plan/stack:
- ☑️Add crypto support to stock statarb alpha engine.
- Rebuild risk model for crypto.
- Agentic crypto alpha push.
- Trade. Market orders. Eat fees.
- HFT model & impl. Split fees, maybe.
- Mixin additional non-stat arbitrage.
Big picture.
1
3
598
Alpha studio frontend rebuild w/ AI/Agent in-built is DONE ✅. Just pure react/fastapi/postgres (integrated w/ pre-existing fastapi/postgres server).
Stocks + Crypto + Statarb + AI + GPU + UI 😍 All the buzz words 😅
3
2
706
Riffed with the agent a few minutes. We came up with a 1 hour crypto statarb alpha, using a single field (non-OHLCV). Tons of room for agent improvement, but I'm well beyond impressed, esp. when agent couldn't see beyond 2025 metrics.
2
1
463









