This GPT-5.6 Trading Bot Is CRUSHING Hyperliquid 24/7 (so far) — Transcript

Full transcript

1. 0:00Hello, hope you all doing well. So, I
2. 0:02wanted to do an update video because
3. 0:04I've been playing around with a lot of
4. 0:05stuff using AI cloud code GPT 5.6 Sol
5. 0:10Fable to try to improve my trading. So,
6. 0:13today we're going to focus on hyper
7. 0:14liquid and I built a fully automated
8. 0:17model that has been really doing really
9. 0:19well for me. So, I wanted to make a
10. 0:21video on it. You can see we started
11. 0:23yesterday and we are already up $170
12. 0:27with a really nice graph here. Uh if we
13. 0:30select this to 7 days, you can kind of
14. 0:32see our spike in the right corner here.
15. 0:35But basically, what we have been doing
16. 0:38is we have created a 24/7 running uh
17. 0:42automated trading system built by GPT
18. 0:455.6 Sol. So, today I was wanted to go a
19. 0:47bit into the details because I think
20. 0:49this exact setup is a really good
21. 0:52starting point.
22. 0:54Uh of course, we have this set up like a
23. 0:56evolutionary mode. So, we always collect
24. 0:59the data and we try to improve it over
25. 1:01time. But uh I wanted to go through how
26. 1:04I set this up today. So, I prepared a
27. 1:07slide just to go through it so we can
28. 1:09kind of explain a bit more about it. So,
29. 1:11I think we just going to get into that
30. 1:12and get started. Okay, so I just created
31. 1:15this slide we can kind of have a look
32. 1:16at. So, yeah, the autumn AI automated
33. 1:19trading system we built. It's not AI
34. 1:21automated, it's built by AI to be
35. 1:24automated. That's the difference. Uh but
36. 1:26basically, this is the setup. It's super
37. 1:29easy and I created this because I want
38. 1:31to learn from it. So, of course, we
39. 1:33check how much we have in our balance.
40. 1:36We scan the market. We score the IDs
41. 1:39that the market finds from our scan
42. 1:41setup. If we pass our
43. 1:44um
44. 1:45parameters, we will open that trade. We
45. 1:47will spawn a manager that keeps track on
46. 1:50the trade and it will close it when we
47. 1:52meet our, yeah, parameters. And that's
48. 1:55basically it. So, we scan this every 5
49. 1:58minutes in the background. So, this is
50. 1:59not using any tokens. It's just written
51. 2:03in code. But, basically here are the
52. 2:06score system we have. You can see
53. 2:08we have plus 20 if it's has is it easy
54. 2:11to trade. This basically means that
55. 2:13there is a lot of liquidity. If the
56. 2:15system has seen like a longer move, so
57. 2:17for example, in our example, SpaceX
58. 2:19declined 1.6%
59. 2:22across 48 price checks. That is plus 15.
60. 2:26And you can see we have plus 15 for the
61. 2:28RSI balance. This kind of measures the
62. 2:31buying versus selling pressure. And we
63. 2:33have a range on that. So, you can see
64. 2:34SpaceX was 47.3 inside inside our range
65. 2:39from 42 to 65. We have a plus 20 score
66. 2:43if the trend is wider. So, like a fast
67. 2:46average below uh the slow average means
68. 2:48prices are broadly falling. And here you
69. 2:51can see we have a gap of .6%
70. 2:54or .60% I guess.
71. 2:57Uh also, we have something called a weak
72. 2:59bounce. So, for this we got 18 points.
73. 3:02Uh a bounce is a brief rise during a
74. 3:04fall. So, basically when a stock is down
75. 3:07and we get like a bounce. And we only
76. 3:09got a weak bounce here, right? Over the
77. 3:11last three checks, we got 18 score for
78. 3:14that. And also, we have something called
79. 3:16a small bounce. We also check this. So,
80. 3:18we ended up with like a 96, but we need
81. 3:2270 or more to actually enter a trade.
82. 3:25So, that was like 26 above. So, this
83. 3:26means that we think this is a good place
84. 3:29to enter. Okay?
85. 3:31So, also, we are running leverage on
86. 3:34this because I want to run this on a
87. 3:35small account. This is of course much
88. 3:37more risky, but uh this is like a
89. 3:39training setup I have, learning setup.
90. 3:42So, we do like 300 times 10. So, we
91. 3:46actually do the position with $3,000.
92. 3:49Uh just because
93. 3:50uh
94. 3:51yeah, I just wanted to do that because I
95. 3:53think it's more interesting and it kind
96. 3:54of reflects a bit more if you had like a
97. 3:57big balance to trade with.
98. 3:59Uh this is the example, so you can see
99. 4:01we had like $385 free and we have one
100. 4:05slot open. We got 96 points as
101. 4:07described. We actually went short on
102. 4:10this at 139 and you can see we keep
103. 4:13watching this every 5 seconds for 43
104. 4:16minutes we actually run this trade
105. 4:18and we closed this at the 137.88
106. 4:22and we made a $25 profit. The trading
107. 4:26cost was uh 50 cents, so we netted 24 on
108. 4:30this SpaceX trade.
109. 4:32So, another thing that this has and this
110. 4:35was kind of our main thing about this
111. 4:37setup is that we want to trade, record
112. 4:40the results and the data and then we're
113. 4:43going to go back to GPT-5.6 now and try
114. 4:46to adjust and improve the strategy over
115. 4:48time based on the more
116. 4:51data we collect and this is what I think
117. 4:53is the big advantage using this setup
118. 4:55because we can collect all the data we
119. 4:57use, we can still try to evolve the
120. 4:59strategy over time
121. 5:01uh
122. 5:02not keeping it stale. And that is what
123. 5:04we're going to do next now because this
124. 5:05has been running for like 24 hours and
125. 5:07we haven't tried to improve it yet, so
126. 5:09I'm going to head over to GPT-5.6 now,
127. 5:12look at all the data we have collected
128. 5:14and try to improve this now. So, let's
129. 5:17just head over to our Codex. I'm going
130. 5:19to switch the model now. I'm going to
131. 5:21switch to solve. Uh let's just do extra
132. 5:23high for this.
133. 5:25And uh or maybe we can just do max for
134. 5:27this. I haven't tried that too much. I'm
135. 5:29not going to do ultra because it's just
136. 5:31way too much. So, I'm going to try solve
137. 5:33max for this setup. So, I'm just going
138. 5:36to put in my prompt here
139. 5:38and let's try to improve our strategy
140. 5:40based on the results that we have gotten
141. 5:42so far. So, I'm just going to keep the
142. 5:44prompt pretty It's We have been running
143. 5:46for 20 plus hours now. Let's look at the
144. 5:48data we have collected so far to improve
145. 5:50the strategy to make it more robust and
146. 5:51profitable, analyze the data, and run
147. 5:53calculations to look for meaningful
148. 5:55improvements in the strategy. That is
149. 5:57all I'm going to give this now, and I'm
150. 5:59just going to keep it running in the
151. 6:00background here, and I'll come back if
152. 6:03we have something. It might not be that
153. 6:05we see any improvements, but if we do,
154. 6:08we're going to try to implement them.
155. 6:10I'm going to run it for a while, and I'm
156. 6:11going to come back to see if we have any
157. 6:14kind of improvement.
158. 6:16Basically, we most likely would not see
159. 6:18this in this video because I plan to
160. 6:20post this today, but let's see what we
161. 6:22can find here in the data we have
162. 6:24collected and if we can make any
163. 6:26meaningful improvements. Okay, so after
164. 6:29Sol Max ran for 25 minutes, we got our
165. 6:31response and didn't really want to make
166. 6:34any big changes here. So, basically,
167. 6:36we're going to keep the 5-minute stale
168. 6:39check. We're going to implement all the
169. 6:41other improvements. These are basically
170. 6:43not strategy-related,
171. 6:45but basically basically some retries,
172. 6:47some
173. 6:48emergency exit setups, and stuff like
174. 6:50that. So, I'm just going to implement
175. 6:53these. Okay, so you can see we
176. 6:55implemented all the changes. We ran 44
177. 6:58pad tests passed. Perfect. And now you
178. 7:00can see we have set up a system that the
179. 7:02data collection will let us test those
180. 7:04properly over the next 200-plus trades.
181. 7:08So, we kind of improved the way we
182. 7:10collect the data, and this means that we
183. 7:12can even improve this more the longer we
184. 7:15run it. So, this is something I wanted
185. 7:16to implement because when I set this up
186. 7:19initially, I didn't really focus too
187. 7:21much on that. I just wanted to see if it
188. 7:23can run,
189. 7:24but now we should have had that up and
190. 7:26running, and hopefully we can improve
191. 7:28this over time. We also did some small
192. 7:30changes to the
193. 7:32the the labeling here. So, now we have
194. 7:35some more clustering here. So, we have
195. 7:37like a hard limit on we can only do one
196. 7:40cluster at a time, but uh might actually
197. 7:43go back on that. You can see we have a
198. 7:45hard limit of one live position per
199. 7:47economic correlation cluster. We might
200. 7:50change that up. We'll see how well it
201. 7:52goes.
202. 7:53But uh this is actually how I am using
203. 7:56um yeah, this case GPT 5.5 5.6 soul to
204. 8:03experiment with this. So, basically uh
205. 8:06you can see we did quite well here. I
206. 8:07think we If you look at the trade
207. 8:09history, we just exited with a $23
208. 8:13win here, but uh you will see some up
209. 8:16and downs.
210. 8:17Uh so, this is not something that is
211. 8:19like perfect or anything. It's not going
212. 8:22to be like a steady curve up to the
213. 8:24right, but uh the main idea for me is to
214. 8:26learn from this.
215. 8:28So, uh this is something I want to use
216. 8:30to actually learn more about uh AI
217. 8:33agentic trading. I even tried in the
218. 8:36beginning to set this up to always run
219. 8:39in uh like a loop in uh GPT 5.6 to
220. 8:44monitor every single, but it kind of ate
221. 8:47up all my tokens. So, to do that, I
222. 8:50guess I can show you the
223. 8:53the prompt I used to run this in like a
224. 8:55loop inside GPT 5.6 to always keep
225. 8:59track.
226. 9:00So, this is what I used yesterday. You
227. 9:02can see I do the {slash} goal, create a
228. 9:05Kanban list as follows. One, sleep 180
229. 9:08seconds. This is just to give it some
230. 9:10time in between. Two, we're going to
231. 9:12check a live position status. And three,
232. 9:15scan for new trades. It's supposed to be
233. 9:17new. And be active. Four, enter if any
234. 9:20trades found. Five, restart this goal.
235. 9:23So, this kept me going in like a
236. 9:25continuous loop, but after a while it
237. 9:27just started eating up my tokens, so I
238. 9:29decided to ditch that and move
239. 9:32everything over to a TeamViewer session
240. 9:34that runs this code 100% autonomous. So,
241. 9:37that was much better, actually. And we
242. 9:39kind of reduced our latency a bit and
243. 9:41stuff like that. So, I'm really glad I
244. 9:43did that. This is something I learned by
245. 9:45just experimenting. Couldn't really find
246. 9:47anything about it online. So, all of
247. 9:51these
248. 9:52AI genetic trading setups I've been
249. 9:54using, basically I've been just
250. 9:55experimenting, always using the models
251. 9:58to run map-based database tests to kind
252. 10:02of confirm what we hopefully want to
253. 10:05achieve. Sometimes it doesn't work out.
254. 10:07Sometimes it does work out. This is
255. 10:10something I also do over on Polymarket.
256. 10:13So, yesterday I did an experiment in the
257. 10:15France-Spain game to try to do some
258. 10:18automated bet placing using AI.
259. 10:22Didn't go that well. I think we actually
260. 10:24lost some money, but we had some good
261. 10:26trades here. You can see we have a 50
262. 10:27plus here, 49 here. So, we actually did
263. 10:31sell out. We built some ladders and
264. 10:33stuff like that to enter. But, I think
265. 10:36we overall we lost some money, I think
266. 10:38on this.
267. 10:39Here you can see we have this ladder
268. 10:41that kind of
269. 10:43yeah, printed over time with different
270. 10:45entry points and stuff like that. So,
271. 10:46that was pretty cool.
272. 10:48Yeah, don't mind the balance too much. I
273. 10:50did like a deposit if you have seen any
274. 10:52of my previous videos.
275. 10:53But, if you look at the portfolio the
276. 10:56last week, we have been doing really
277. 10:57well. 389
278. 11:00in profit. And this is mostly my fully
279. 11:03automated uh
280. 11:055-minute up and down Bitcoin setup that
281. 11:07I've been doing before on this.
282. 11:10So, basically this video was just to
283. 11:12show you and share a bit about my
284. 11:15thought process going into an experiment
285. 11:17like this.
286. 11:18And when I find something that produces
287. 11:21these kind of results, I always want to
288. 11:23double down on it.
289. 11:25And now it's just going to be
290. 11:26interesting to see how well this run
291. 11:28over the next few days.
292. 11:30I promise I will make an update video on
293. 11:33this if you are interested. This might
294. 11:35go to zero and we lose everything we
295. 11:37have gained so far. Most likely we going
296. 11:39to
297. 11:40go down again or I call it reverse to
298. 11:43the mean here because I think this is
299. 11:45kind of over performing a bit, but I'm
300. 11:47excited to see where this goes.
301. 11:49So,
302. 11:50and with kind of the improved data
303. 11:52collection we have done, I think this
304. 11:53could be really interesting. So,
305. 11:55hopefully this gave you some inspiration
306. 11:57to start exploring a bit using these
307. 12:01new models that are so strong at coding
308. 12:03and math and stuff like that and that is
309. 12:05exactly the use case we want to extract
310. 12:07some value from.
311. 12:09And so far, super happy with this one
312. 12:12and
313. 12:13I might make a video explaining a bit
314. 12:16more how to start from scratch on this
315. 12:18to be honest because this was a bit all
316. 12:20over the place, but hopefully it gave
317. 12:21you some inspiration to start
318. 12:24experimenting as I am doing at the
319. 12:26moment because of these models, because
320. 12:28of this technology, no one really has
321. 12:30like a good
322. 12:32perfect way to use this yet. I'm sure if
323. 12:35you have like tons of experience in
324. 12:37quant algo trading, you can use this
325. 12:39better than me, but me as a hobbyist
326. 12:42over a long time, I found this very
327. 12:44helpful at least with kind of the the
328. 12:46database, the math base and running
329. 12:48these
330. 12:49setups.
331. 12:51So, yeah, that was all I wanted to share
332. 12:53today. Hope you enjoyed it. Give this
333. 12:55video a like, subscribe if you want to
334. 12:56see more and hopefully I see you again
335. 12:58in a few days. Have a nice day.