Joined April 2021
- Tweets 761
- Following 46
- Followers 661
- Likes 436
39 Photos and videos
In the near future, normie hardware will hardlock to approved LLMs. Distributed bittorrent-style bootleg LLMs will run like piratebay for jailbroken nodes. Nation states will form massive botnets for on-demand GPU's to run super-LLMs.
1
89
Jun 15
Swapping back momentarily to linalg for mid-freq after dealing with SDE for high-freq is such a heavy burden lifted ๐
1
254
Jun 15
Built my mid-freq cryptoverse dataset over the last day. So much faster than a few years back building same stock universe manually. Feels a bit like years of wasted time coding, almost. The existing hardened spec seems to really ground the agent.
9
952
Jun 13
Pointless to proceed w/ HFT sim w/out Fable, which might sound like nonsense to some. It was my first taste of semi-autonomous novel-ish development. It really was better than a team of actually good PhD's. I'll just integrate my MFT, HFT and web3 codebases until it returns.
1
38
5,078
Jun 13
You gotta be kidding me. Now the US gov't has taken Fable from us.
1
1
488
Jun 13
Seems like now is the time to redirect the old mining rig to a massive local AI model rather than HFT sim. Rules for me, but not thee will accelerate from here.
377
Jun 12
Fable helped pull me out of the HFT sim rut. I don't want to count how many 100's of million of tokens, but it was worth it.
Now I'm waiting on my own GPU half the day for sim runs. Time to fire up the old GPU mining rig for this?
7
943
Jun 12
X quant bot farmers: Academic papers are useless for real trading, try this. Also, MIT/Harvard paper reveals top hedge fund secrets, click to join.
I want to exist in the space where nuance matters.
2
385
Jun 12
I'm getting comfortable with the understanding that passive algorithmic market making alone -- good physics, but no alpha/edge/forecast -- is pure survival. A mechanical harness to safely execute that alpha.
1
10
1,874
Jun 10
Fable is 30 minutes and 3 million tokens into trying to fix my HFT sim code ๐
6
5
2,248
Jun 10
70 million tokens and counting. More than I've used with many months of Gemini Pro. But, oddly, I don't have the usual sinking feeling of spaghetti code I've gotten in the past.
535
Jun 10
HFT sims really are the final boss. Essentially one must replicate all market players, the aggregate, multiple assets, one's own decisions, game theoretic reactions, and the entire exchange matching logic...One small bug and it all collapses. That's just to start the modeling.
3
59
28,148
Jun 10
So it took a few hours and 40 million tokens to get me out of a 2 week rut. Tempted to dump another 100 million tokens into this...More effective than my previous team of 6 human Ph.D.'s ๐คจ
3
471
Jun 10
Watching how it operates (still going 1 hour)...first time I've felt an inkling of fear about AI ๐ง. We'll see how the final result looks...
393
Jun 10
Uh, still going at 11M tokens in 1 prompt. Feel like I'm shoving pizza under the door of a room of quants to keep them going.
341
Jun 10
Exactly how I feel, but it's unavoidable.
Jun 9
Doing HFT sims ๐ตโ๐ซ๐ตโ๐ซ๐ตโ๐ซ
317
Jun 3
To his 387k followers: just unfollow all 25k influencers. You are being manipulated. Actually, just get off social media. Period. Nobody is putting all that effort in to simply "share knowledge to give back and help" ๐.
245
Jun 2
Built the physical HFT book/tape ingestion and oms. To get a true intuition for physics of it, an end-to-end synthetic data -> emulated processing -> simulated exchange is going to be a necessary tool. These things tend to escalate.
194