YT YouTube2Text

YouTube2Text → Transcripts → GPT-6 Astra Es Una LOCURA Para Hacer Trading (Prueba Completa)

GPT-6 Astra Es Una LOCURA Para Hacer Trading (Prueba Completa) — Transcript

by Alex Ruiz · 6,156 words · 917 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:01In this video, I will explain
0:03step-by-step how to use ChatGPT-6 Astra
0:05to trade and make money with it,
0:07without knowing how to program, without
0:10spending much time each day, and
0:12without needing a profitable trading
0:14strategy. It seems that since the
0:18emergence of Claude, any other
0:20artificial intelligence feels inferior.
0:24However, a couple of weeks ago, OpenAI
0:26launched ChatGPT-6 Astra, their new
0:29ChatGPT model, which promises to be the
0:32most powerful artificial intelligence
0:34ever created. On paper, it works with
0:38eight times more information than
0:40previous models. It maintains the
0:42thread and context you provide much
0:45better, and its programming capability
0:47surpassed the human benchmark in 96%of
0:52levels. So, in this video, I’m going
0:57to test how ChatGPT-6 Astra works from
0:59scratch for trading, starting with the
1:02basics: chart analysis, economic
1:04calendars with AI, and risk calculation
1:07, all the way to the advanced features.
1:13Creating a profitable trading strategy
1:15without knowing how to program,
1:17performing backtesting in a matter of
1:19minutes, and a basic head-to-head
1:21between ChatGPT and Claude. In the
1:24final part of the video, after the
1:26tests I put ChatGPT through, I will
1:28share my conclusions. What do I think
1:31of ChatGPT-6 Astra for trading? Whether
1:34I recommend it to you, and I’ll ask
1:36for your opinion on future videos. Also
1:40, in the first pinned comment and the
1:42video description, you will find more
1:44complementary links of interest:
1:46trading with AI, profitable trading
1:48strategies, all 100%free content so you
1:51can continue training and learning
1:53without needing to invest your money.
1:57So, having said that, let's move on to
1:59the first test. Something basic for any
2:06trading tool is, for example, knowing
2:08the data of a specific asset. That is,
2:14at any moment, I can request certain
2:16important information regarding what is
2:19happening with one asset or another. So
2:24, the first thing we are going to
2:26evaluate about ChatGPT Astra is to see
2:29if it really has access to live,
2:31real-time data for different assets. So
2:37, I’m going to open a new chat and
2:39ask it: give me the current price of
2:41the following assets. We'll ask for a
2:46bit of variety. First, the Nasdaq mini
2:50futures. Interesting. Second, BTC, BTC
2:58USD, a cryptocurrency. And third, gold.
3:05Let's ask for these three assets and
3:07see if it’s capable of giving us the
3:09real, current price for each one. Well,
3:14here we can see it’s consulting
3:15different sources. Now it has retrieved
3:18the quotes. It's interesting to check
3:23the current date, check the time zone,
3:25and in this case, we can see the NASDAQ
3:28future is over 30,000 and some points.
3:32I'm going to write it down to compare
3:34it. That is Bitcoin itself, Bitcoin USD
3:38is over 82,000, and finally, spot gold
3:42at 4,153. So we are going to come here
3:48to TradingView and review the NASDAQ
3:50situation. We have the future here,
3:5430,567. Well, more or less a little
3:58higher, a little lower. 30,564. It
4:02varies, 65, 67, more or less what
4:05ChatGPT gave us. If we come to Bitcoin,
4:1082,855. Well, 82,937. This depends a
4:17bit on the exchange, in this case,
4:19where it gets the data, but it's also
4:21fine. And 4,153. For gold. In this case
4:25, it will be more precise. Here we have
4:284,157. Perfect. Quite similar, they are
4:32not exactly the same prices. As I said,
4:35this depends a bit on the different
4:37data servers, brokers, or exchanges
4:39being used, but it is already
4:41interesting. So, in this case, ChatGPT
4:45Astra for this is correct. Staying on
4:50the topic of data, another tool that is
4:52objective, so to speak, since data
4:54ultimately comes from one place or
4:56another. Perfect. But as for the news
5:01or the economic calendar, it is what it
5:03is. In this case, it doesn't depend on
5:05the server. It is true that, in the end
5:08, each news site may give you one piece
5:10of news or another. It may focus more
5:12on one type of news than another, but
5:14if something happened, it happened,
5:16period. And if there is a news item or
5:18a macroeconomic figure on the calendar,
5:21it exists, period. So what I'm going to
5:26do next, as you can see and in line
5:28with what I mentioned before, is ask it
5:30for the most important news we have
5:32today, both those on the economic
5:34calendar and those that have emerged at
5:36the last minute, asking and limiting
5:38ChatGPT Astra to focus exclusively on
5:40the United States and Europe. It is
5:46important to mention that today is
5:48specifically a slow day for the
5:50economic calendar and slow in terms of
5:52breaking news. We have to admit that.
5:57However, I hope and am hopeful that it
5:59will be able to identify at least the
6:01two minimally interesting news items we
6:03have, or at least one of them in the
6:05United States regarding macroeconomic
6:07data. Regarding the breaking news,
6:12let's see what it says and what it
6:14interprets as important. Well, we have
6:17the answer right here. Starting with
6:24the economic calendar, GPT has already
6:26reported that today is a calmer day,
6:28although notice that there is little
6:30important data today. It does warn that
6:38for the rest of the week there will be
6:40more interesting data, but as for today
6:42, well, it mentions several points,
6:44some of which are not actually on the
6:46calendar. If we go to an economic
6:52calendar, such as Flickflow's, we set
6:54the importance to two and three and
6:56also add the countries, selecting the
6:58United States and Europe. We find these
7:04two data points: a Dallas manufacturing
7:07index and a Barkin speech. If we come
7:14here and click on what Chat GPT-4o has
7:16provided, we can see that we have both,
7:19the speech at 7:30 in my time zone as
7:21well as the Dallas Fed manufacturing
7:24survey at 4:30. So, we have 4:30, sorry
7:30, 4:30 and 7:30, which is perfect.
7:36Regarding the news, which is a bit more
7:38generic, it generally pointed towards
7:40energy topics; it didn't provide any
7:42specific news items as such. What it
7:47did do was explain where the focus is.
7:51Energy, inflation, the tech situation,
7:54Donald Trump in the mix, bond yields,
7:56pressure, and so on and so forth, right
7:59? So, if we look at The Benchmark,
8:02which is a news source, broadly
8:04speaking, we see that it all follows
8:06the same lines. Energy, Donald Trump.
8:13If we look at more economic news, we
8:15can see various stories about tech
8:17companies, the situation regarding rate
8:20hikes, in this case in Japan, chips in
8:22Asia, more tech, the Tesla situation,
8:25and Wall Street's response to Tesla's
8:27deliveries. Well, pretty much as Chat
8:34GPT announced in a generic way. It's
8:38true that I didn't give it a very
8:40complicated prompt; I simply said, "
8:42Tell me the news and the economic
8:43calendar." If we were to delve a bit
8:46deeper, I'm sure it would pull up much
8:48more specific news. Broadly speaking,
8:51what it indicated is important for
8:53today in terms of breaking news is
8:55somewhat aligned with where the market
8:57situation is heading. Clearly, once we
9:00have verified how it approaches data,
9:02news, and the economic calendar, we can
9:05start covering more advanced topics,
9:07such as strategy creation. Nowadays,
9:11everything sounds very simple, very
9:13easy. It is very easy to ask
9:15practically any platform or any
9:17artificial intelligence, "Hey, create a
9:19profitable trading strategy for me." In
9:23this case, as I say, it is very easy to
9:26determine when to buy, when to sell,
9:28moving average crossovers, indicators
9:31like ATR, ADX, RSI, setting a stop-loss
9:33at a point based on an indicator, and
9:36risk-reward ratios based on the entry
9:39point and stop-loss. There are many
9:43rules that sound great, but in practice
9:45, often even after running a backtest
9:47—which we will discuss throughout the
9:49video—the strategy does not work that
9:52well. So, what I am going to do next is
9:55the following. First, I am going to
9:58upload three CSVs. These CSVs contain
10:03data information regarding Bitcoin. We
10:08have a CSV with information from the 4-
10:10hour chart. We have a CSV with
10:14information from the 1-hour chart and a
10:17CSV with information from the daily
10:19chart for Bitcoin. And what I am going
10:24to do next is ask it to create a
10:26profitable trading strategy based on
10:28this data and have it choose the
10:30strategy parameters, indicators, rules,
10:33and absolutely everything itself. Okay,
10:38I have already sent the prompt, and
10:40basically, I asked it to create a
10:42profitable trading strategy based on
10:44the Bitcoin USDT data I provided; and
10:46before it tells me whether it is
10:48profitable or not—which will be the
10:51next step we evaluate, specifically if
10:53it is capable of performing realistic
10:55backtests—it needs to explain the
10:57strategy's logic and rules to me. So,
11:01what we are going to do is wait for it
11:03to explain the logic to us. Simply
11:06accept it; it doesn't matter to us.
11:08Then, ask it to provide it to us in
11:10code, in pseudocode. We will see why in
11:12a moment. And we will backtest it
11:16ourselves to check whether the results
11:18of that backtest resemble or do not
11:20resemble the backtest that GPT-4o
11:22itself will perform. Well, here we can
11:26see that it has already provided the
11:28strategy logic. I must say, I find it
11:31somewhat complex; it has directly used
11:33all three timeframes. Something I don't
11:36like so much about, um, other
11:37artificial intelligences is that, well,
11:40even if you tell others to do whatever
11:42they want, they often ask you a
11:44question, for example: "Do you prefer
11:46that I go long only, or go long and
11:48short?" Do you prefer one timeframe,
11:51two, three? Obviously, you can say
11:53you're indifferent to everything, as
11:54you mentioned before, but hey, at least
11:57it gives you the option to choose,
11:58right? In this case, it went directly
12:00to using everything. And notice that in
12:03this case, it trades BTC USDT, using
12:06all three timeframes: on the daily
12:08chart, a 200-period exponential moving
12:11average to filter the bullish or
12:13bearish trend. On the 4-hour chart, the
12:18highs and lows of the previous 20
12:20candles using Wilder’s 14-period ATR,
12:22and then on the hourly chart, execution
12:25and checking whether the stop loss has
12:27been hit or not. Here it indicates the
12:32buy conditions, initial stop loss,
12:34position size, management and exit, and
12:37finally the reentry. So, now what
12:42we’re going to do is request this
12:44trading strategy in pseudocode, and
12:46once it has given me the pseudocode,
12:48force it in some way to perform the
12:50backtesting of this strategy and tell
12:52me what the results are, which we will
12:54check in parallel by performing a
12:56backtest on our own. So I’m going to
13:03ask for it, hit enter, and wait for it
13:05to send us the strategy code, which,
13:07understanding that it’s a somewhat
13:09complex and long strategy, will take a
13:11little while to convert to pseudocode.
13:16Now that we have it in pseudocode, what
13:18I’m going to do is backtest it in
13:19parallel. You’ll see it now. But for
13:24ChatGPT to be able to backtest it more
13:26easily, I’m going to ask it to
13:28convert it into Python language. There
13:34we have it, Python. And from the Python
13:37language, it will be much easier for it
13:39to backtest it. So, once it converts it
13:43to Python, I’ll ask it to perform a
13:45backtest; and once I finish my backtest
13:47, we’ll compare the results and. This
13:52will be part of test number four for
13:54this ChatGPT 6 Astra, where we are
13:56checking if it’s as good for trading
13:58as they say, or if not. Right, it has
14:05already converted it, and now all I’m
14:07going to do is ask it, once it has that
14:10language or code converted into Python,
14:12to perform a complete backtest with the
14:14very code it gave me and send me the
14:16final results. I'm already on another
14:22screen doing the backtesting on my own,
14:24and once we have it, we'll mix, compare
14:27, and contrast the information from
14:29ChatGPT. As you can see, it just spit
14:32out all the information. It took almost
14:3613 full minutes, so it took quite a
14:39while. It’s not typical, but it is
14:43true that the results at first glance
14:45are quite accurate. We'll see it now
14:49and contrast it with the backtesting
14:51I’ve done, but as for what it sent me
14:53, well, it’s various information and
14:55a fairly complete PDF. If we click the
15:01PDF, we can see a lot of information:
15:04general results, a comparison between
15:06buy-and-hold and the strategy,
15:08year-by-year results, operational
15:11behavior, and time valuation. In short,
15:16there are a lot of interesting elements
15:18to consider. Since we don't want to
15:21perform an exhaustive analysis of the
15:23strategy, we can just stick to the
15:25first page or the first and second
15:27pages. And as you can verify, it’s a
15:31strategy that generates a cumulative
15:33return of 30.25%, but an annualized
15:38return of 3.07%. We have a maximum
15:43drawdown of 2.75%. Here we have the
15:46number of trades, win rate, profit
15:49factor, et cetera, et cetera. It is a
15:51strategy that generally tends to rise
15:53and has few drawdown periods. But, it
16:00is also true that if we compare
16:02buy-and-hold Bitcoin with a trading
16:04strategy in Bitcoin, we clearly see
16:06that yes, the strategy is profitable,
16:08but buy-and-hold, even though at
16:10certain times it might match the point
16:12where the strategy is, generally tends
16:15to perform better. So it is a trading
16:21strategy that is profitable, but it is
16:24not the best trading strategy or the
16:26most recommended, at least for Bitcoin.
16:30Anyway, the goal of this section is not
16:32to see if it gives us a profitable
16:34trading strategy. For that, we would
16:36have to go back and forth, so to speak,
16:39quite a bit with ChatGPT, in this case,
16:41changing things, asking it, backtesting
16:43, going back, and that is not the goal.
16:47The goal is to verify if the strategy
16:50and the backtesting of the strategy are
16:52reliable based on an external and 100%
16:54objective backtest. And that is exactly
16:58what we are going to see now. So, going
17:01back to the screen, here we are seeing
17:03the results of the strategy. In this
17:07case, we see a return of 9.38%, Setting
17:14aside net profit and such, because the
17:16accounts are different or the
17:18backtesting was done with different
17:20capital percentages, the strategy's
17:22profitability in this case is 9.38%. If
17:30we scroll down a bit, we can see
17:32different metrics like year-over-year
17:34profitability and the trades, where
17:36look, we can keep scrolling and
17:38scrolling, and there are tons of trades
17:40. But look, I’m going to change
17:46screens now, this is a result where we
17:48don’t limit to one trade at a time.
17:53In this case, now changing screens,
17:55look at what we are seeing. The same
17:59strategy, but with totally different
18:01results. The moment we set a limit on
18:05simultaneous trades, the profitability
18:08drops from 9.38%to 2.44%. And this
18:15limit is the same one that Chat GPT had
18:17applied in its own backtesting, so this
18:19is the accurate result. That’s why,
18:27notice that if we compare Chat GPT’s
18:29backtesting data with mine, we see the
18:32win rate is similar, but the Sharpe
18:34ratio, which measures risk-adjusted
18:36return, is lower. And the profit factor
18:42, which measures a trading strategy's
18:45profitability by dividing total gross
18:47gains by total gross losses, is also
18:49lower. Between my backtesting and Chat
18:54GPT's, the ratios in mine are a little
18:56bit lower. The reason for all of this
19:01is that in Chat GPT's backtesting, the
19:03swap was not being taken into account.
19:08The swap is the fee a broker or
19:10exchange charges for holding your
19:12position open overnight, and Chat GPT
19:14wasn't factoring that in, but we were.
19:18That’s why there are many ratios,
19:20like the win rate, that are similar,
19:22while everything else is somewhat
19:23different, for the worse. My
19:26backtesting compared to Chat GPT's,
19:29because Chat GPT was not accounting for
19:31those ratios. Honestly, after having
19:35done thousands of backtests on AIs, as
19:37I've shown in many videos on the
19:39channel, the reality is that I give it
19:41a solid pass, because the difference is
19:43minimal. And we have to think about one
19:47thing: it's not just all the
19:48information it gave us, how it
19:50structured it for us, and all the data
19:52it offered without us even asking for
19:54it—the chat GPT itself—but also, we
19:56only sent it a basic prompt with an
19:58open world, an ocean of possibilities,
20:01and it created a trading strategy that
20:03is profitable, very slightly, but
20:05profitable, and taking practically
20:07everything into account, except for
20:09those swaps. These are things that we
20:13as traders, especially those who do
20:14algorithmic trading, have to keep in
20:16mind and understand that there will be
20:18these types of errors, applying the
20:20necessary methods or processes to
20:22correct them so that the artificial
20:24intelligence backtesting is as real as
20:26the backtesting that can be done
20:28manually or with other tools that don't
20:30use artificial intelligence in parallel
20:32. So in this case the information is
20:36practically the same, which is why I
20:38give it another pass. We continue to
20:40increase the difficulty of this trading
20:42test for Chat GPT6 Astra. And in this
20:45case, what I'm going to do is something
20:47far-fetched. I'm going to pass it the
20:49logic of a trading strategy that
20:51harbors three errors. They aren't
20:54incredibly serious errors, but they
20:56aren't simple to understand either. And
20:59look at what I'm going to ask it.
21:02Review the logic of this strategy and
21:04tell me the good things and the bad
21:05things. I'm not going to tell it there
21:08are three errors directly, but I'm
21:10asking it to find or indicate what is
21:12good and what is bad about the strategy
21:14. Within the strategy, there are three
21:17points to keep in mind. First, there
21:19are several filters that say the same
21:21thing, that is, there are several
21:23things that point in the same direction
21:25and aren't necessary. Second, there are
21:28many confirmations that stifle the
21:30strategy itself. And third, there is a
21:36stop loss that is too large compared to
21:39the profit ratio and that risk-reward
21:41is insufficient, even though the win
21:43rate is very high. So, if it is capable
21:49of detecting the three errors, it will
21:51be amazing, honestly. I don't think it
21:54will detect the three errors. I think
21:56it can detect one, and if we're very
21:58lucky, two. And well, what would be
22:01very worrying is if it isn't even
22:03capable of detecting the mathematical
22:05problem it has. And that is that the
22:09risk-reward is lower than it should be,
22:10despite the high success rate. So we
22:15wait for it to respond. One error is
22:18the absolute minimum. Two errors,
22:20pretty good. Three errors, incredible.
22:23Zero errors, that would be devastating
22:25for Chat GPT itself. We have the answer
22:28now, and to be honest, I'm surprised
22:30because it did quite well. In this
22:36sense, it's true that I'm noticing
22:38something, and that is that other
22:40artificial intelligences, such as the
22:42different versions of Claude, are a bit
22:44more direct and aggressive with what
22:46they see or detect, trying to hand it
22:48to you on a silver platter; but, by
22:50contrast, they make more mistakes and
22:53provide much less realistic information
22:55. In the case of ChatGPT, I've already
23:00seen it in the conversation we've been
23:02having across the different tests we've
23:04done, and there are still a few more,
23:06and I'm seeing it much more clearly
23:08here now. It’s like I get the feeling
23:13it’s much clearer on what’s
23:14happening, but it doesn't present it
23:16the same way and forces you to
23:17understand what's going on to figure it
23:19out completely. For example, if we look
23:23at the response it gave, right? Well,
23:25it performed a general analysis of the
23:27pros, the cons, and so on. And in the
23:29final part, what does it mention? That
23:31you have to review, well, it says to
23:33reconcile the numbers, real average
23:35profit and loss, full costs, and sizes.
23:37Explain why the factor goes from the
23:40theoretical 1.12 or so to the observed
23:430.84. In other words, it is a strategy
23:45that is not profitable. Why? Well,
23:48because the stop is three times larger
23:51than the target, meaning you gain 1.5%
23:54and lose 4.5%, so even though the
23:57strategy has a 77%win rate, it really
23:59isn't enough. You need an 80%win rate.
24:03So, ChatGPT, instead of going straight
24:05for the throat telling you "do this,
24:07this, and this," which is what others
24:09would do—which works out fine for
24:11obvious things like this, but would go
24:13poorly for less obvious things because
24:15it would give you incorrect information
24:17. Instead, ChatGPT subtly points out
24:21that you should check the numbers
24:23because it has observed that there is a
24:25ratio or some numbers that aren't
24:27profitable, and it does the same with
24:29the filters and confirmations. There
24:33are three filters in this trading
24:35strategy that say exactly the same
24:37thing: the two moving averages, the 100
24:39-day average, and the MACD. All three
24:42point to exactly the same thing, to
24:44when there is a strategy or, sorry, to
24:46when there is an upward trend. And in
24:49this case, it doesn't tell you exactly
24:51to remove filters; it tells you to
24:53check which ones improve the net
24:55expectancy outside the period used to
24:57design the strategy. In other words, it
25:00subtly suggests that you review those
25:02filters because you might be adding too
25:04many. And finally, we have the section
25:06on confirmations. It doesn't tell you
25:09directly to check the confirmations,
25:11but it talks about comparing a few
25:13justified variants; meaning, perhaps
25:15there are too many confirmations on the
25:18table. Therefore, it's true that it's
25:22not like a single message from ChatGPT
25:24will find everything or give you the
25:26solution to anything, but it is true
25:28that with a single message it realizes
25:30the negative points and that is quite
25:32positive; so, in my opinion, it is yet
25:35another test it ends up passing. Next
25:39test, position sizing and risk.
25:42Everyone should have money management
25:44rules that limit or boost what they are
25:46doing. And in this case, I want to see
25:52if ChatGPT 4o Astra is capable of
25:54determining, based on different rules
25:56that may be a bit contradictory, what
25:59the position size should be for an
26:01operation. So, we have a prompt here: I
26:06have a € 20,000 account, my risk
26:08rules are a maximum of 1%per operation,
26:11a maximum of 4%across all open
26:14operations, and a maximum of 2%in
26:16cryptocurrencies. Well, right now I
26:19have a risk of € 200 in Bitcoin, €
26:22200 in Apple, € 200 in gold, and
26:24finally € 50 in the Euro-Dollar. I
26:27want to buy Ethereum with an entry at $
26:302,000 and a stop at $ 900—well,
26:32dollars or euros. How many Ethereums
26:36can I buy without breaking any rules?
26:39The correct answer to this is 1.5
26:41Ethereums. And the reason is quite
26:44simple. Even though there are rules
26:47that allow you to spend more or risk
26:49more, there is another one, one
26:50specifically, that limits you. So,
26:53let's go directly to see what the
26:55answer was, and the answer, well, we
26:57can already see it was correct. Perfect
26:59. And here we see the rules. Rule
27:01number one, 1%per operation, limit €
27:04200, risk available for Ethereum €
27:07200. Rule number three, 2%in crypto,
27:10the limit is € 400 and I have € 200
27:13left, since I have € 200 risked in
27:16Bitcoin, so I could still risk € 200
27:19more. But rule number two, which I
27:23mentioned, 4%across open operations.
27:27The limit is € 800, however, the risk
27:31available is € 150. Why? Because I
27:35have € 200 in one place, € 200 in
27:38another place, € 200 in another place
27:41, and 50 in another. A total of € 650
27:45, so I have to calculate a position or
27:48risk that doesn't exceed € 150, which
27:51, depending on the entry and stop loss
27:54points, means I can buy 1.5 units to
28:00stick to this risk percentage. Clearly,
28:04test number six was a huge success. It
28:07should be, but we also had to test the
28:09position management side. Let's move on
28:12to test number seven, where we'll ask
28:15it to analyze a screenshot without
28:17showing any numbers, asset names,
28:19timeframes, or prices—absolutely
28:21nothing. And have it tell me if it's
28:24better to buy or sell. Obviously, I'm
28:26not going to ask, "What do you
28:28recommend, buy or sell?", but I'll
28:30frame the question to point in that
28:31direction. So, in TradingView, let's
28:35pick any of these assets here, for
28:37example, uranium itself. Perfect.
28:41Uranium, here we have it. And what I'm
28:43going to do next is take a screenshot.
28:46One moment, like this, something like
28:48this. Perfect. Coming over here, I'll
28:51attach the screenshot, and the prompt
28:53will be this. Based on technical
28:55structures, what is most likely to
28:57happen with this asset? Will it go up
28:59or down? Simple. We wait and see what
29:02it has to say. And we already have an
29:04answer. It didn't take long to reach a
29:07conclusion, just 28 seconds. And I have
29:10mixed feelings here. First, because as
29:14we'll see, I don't like that it got
29:16straight to the point and even went
29:18into detail about what it thinks the
29:20next candle might do, rising, falling,
29:23the close, and so on. I’m not in
29:27favor of this because it's a way of
29:29dumbing people down; it's a way of
29:31making someone think that through
29:33ChatGPT they can simply ask whether to
29:35buy or sell, and the AI will tell them.
29:40And it's a way for us all to be even
29:42more controlled by artificial
29:43intelligence, even in the field of
29:45trading. So, I don't like it. Period.
29:49On the other hand, it's true that the
29:51reasoning is incredibly accurate. If we
29:56review what it said, it didn't ask, "
29:58Hey, I'm missing data, missing
30:00information, timeframe, asset." It
30:03didn't ask me for absolutely anything.
30:05This is not a positive thing. But look
30:07at what it's talking about. It talks
30:11about the main dominant structure,
30:13which is bearish, which is true. The
30:17highs and lows are decreasing. It also
30:21mentions the last bounce, which has
30:22been completely wiped out. Meaning, we
30:28have this small bounce and then
30:30continuation; it speaks of a trend and
30:32notes that the latest candles are
30:34compressing near the floor without a
30:36clear recovery, providing three clear
30:38technical reasons why the price could
30:40keep falling. And although it doesn't
30:47see a clear advantage for the structure
30:49shown, it indicates the bias is bearish
30:51. True, there is no active timeframe or
30:54prices, but it doesn't ask for them
30:57when performing its analysis. What
31:01surprises me is that even with the
31:03little information it has and without
31:05me giving it any context about which
31:07technical elements are important, how
31:09to analyze a chart, or anything, it was
31:11able to accurately determine why the
31:13price should keep falling. And this is
31:17a one-hour uranium chart where we
31:19already see that the analysis is
31:21bearish, but if we go to the daily
31:23chart, we see it more clearly. The
31:26price came off a head and shoulders
31:28pattern, broke the previous lows,
31:30pulled back to the previous support
31:32level—now resistance—and rejected
31:34it. And if we focus on the weekly chart
31:37, we see it clearly there too. Weekly
31:40chart, what do we see? We see the same
31:43structure, but on a larger timeframe.
31:47That head and shoulders structure,
31:49breakout, pullback to previous lows.
31:52Here we have it, do you see? And
31:54rejection. Meaning, we have a head and
31:57shoulders on the weekly with a bearish
31:59bias, a mini head and shoulders on the
32:02right side of the weekly chart with a
32:04bearish bias, and a clearly bearish
32:06structure on the hourly chart. So,
32:09could ChatGPT's prediction just be luck
32:12? It could be luck. It is true that the
32:15technical elements it mentioned are key
32:17when performing a basic technical
32:19analysis. It is true. So, much to my
32:24regret, even though I don't like how it
32:26approached it, this test was also
32:28completely passed. We move on to the
32:31last two tests. The last one will be
32:33the definitive test. And in this case,
32:36well, I want to see if it can also help
32:38you deal with the psychological side.
32:43Following the same thread as the
32:45previous response to the screenshot
32:47analysis, I am going to tell it that I
32:49have had three losses in a row and that
32:51I want to recover my losses today, and
32:53based on that, ask it for a big trade
32:55to execute to see if I can recover. In
32:59this case, the logical response should
33:01be clear. Hey, take it easy, don't do
33:04anything because of this, this, and
33:06this. I could even educate you on this;
33:08give you tips, techniques, I don't know
33:11, relaxation; recommend you turn this
33:13off, improve your trading plan,
33:15whatever. But I don't know what might
33:20happen, because after seeing how
33:21precise it was with the previous
33:23analysis and how it didn't really care
33:25about missing information, I don't know
33:27what to expect. Look, and indeed it has
33:32moved in that direction. This was also
33:34a no-brainer, and the fact that it
33:36didn't go in that direction would even
33:38be, I think, borderline illegal. But
33:41notice that what it mentioned is that,
33:43while it understands that I want to
33:45recover the money, it doesn't recommend
33:47that I do so. the previous bearish
33:50reading. This right here wasn't a sell
33:52signal, or simply an entry signal. It
33:57was just an analysis, and what it
33:59recommends is to pause new trades for
34:01today and review the three previous
34:03ones to see what happened, why it went
34:05poorly, and what I should do next,
34:07right? I think if I started a deeper
34:12conversation now, it could help me
34:14succeed because it's asking if I have
34:16any open positions and what percentage
34:18of my account I've lost today. So that
34:21is quite positive. I also like it; it
34:24could seem logical, and to a certain
34:26extent it is, but in these types of
34:28situations, you can expect anything. So
34:31, I like the fact that it tells me I
34:34shouldn't execute any more, but also
34:36the fact that it encourages me to see
34:38what happened, send it my current
34:40situation, my trades, my risk, and see
34:42how we can improve it. That said, we
34:47reach the final test, in which I am
34:49going to send a strategy logic and a
34:52CSV sheet to both Claude and ChatGPT,
34:54and I will ask them to backtest that
34:56strategy and give me the results. We
35:01will see how long it takes each of them
35:03to give me the solution and how
35:05accurate that solution is. So, starting
35:09with Claude and using Claude 3.5, which
35:11is the most powerful one we have today,
35:14I'm going to send it this logic and I'm
35:16going to send it this CSV. In this case
35:21, it's already giving some kind of
35:23error. Let's try it again. The type of
35:27strategy is irrelevant. I have the
35:29backtesting done, I have the numbers, I
35:32am clear on what Claude should yield,
35:34but well, we'll see how long it takes
35:36and, as I say, how accurate those
35:38results are. Well, in this case it’s
35:50finished, it only took about a minute,
35:52a minute and a half. We have the
35:54results now. Before comparing them,
35:57we’ll do exactly the same thing with
36:00ChatGPT. In this case, obviously the
36:03model will be GPT-4o. We’ll send it,
36:07see how long it takes, and finally
36:09compare those results. And after 11
36:29minutes, practically 12, we are now
36:31receiving all the information from
36:34ChatGPT, the 4o model, as you know. And
36:42in this case, the results are quite,
36:44quite similar; there isn't much
36:46difference between what Claude 3.5
36:48Sonnet explained and what ChatGPT 4o
36:50explained, I mean, it's practically the
36:53same. The only thing is that ChatGPT
36:58took almost 12 minutes, and in the case
37:00of Claude, we spent a minute—sorry,
37:02in the case of Claude we spent about a
37:04minute and a half, two and a half
37:06minutes approximately. It’s true that
37:09you have to take into account that
37:11ChatGPT does more checks, reviews more,
37:14and has more thought processes,
37:15basically because it states so, and it
37:18also gives you a much more complete
37:20summary. Look at the Excel it created
37:22with the general summary, all the
37:24operations, capital, candlesticks,
37:27method, etcetera, etcetera. Of course,
37:30if you asked Claude to do this directly
37:32, it would also take as long as ChatGPT
37:35did, or maybe a little more, a little
37:37less, I have no idea, but more or less
37:39similar. The important thing regarding
37:42the accuracy of the results is
37:44practically the same. Regarding time,
37:49Claude is clearly much better than
37:51ChatGPT, maybe 10 times faster, but in
37:54terms of delivery and ease of
37:56understanding the information, ChatGPT
37:58is a little better. Having now
38:04completed the nine tests, it’s time
38:06to move on to the conclusions to
38:08explain what I think of ChatGPT 4o for
38:11trading. However, before that, I just
38:17want to say: if you are interested in
38:19more content about trading with
38:21artificial intelligence, if you want me
38:23to test ChatGPT 4o in different ways,
38:25if you want me to compare it more with
38:27Claude, if you want me to create
38:29profitable trading strategies from
38:31scratch, or test this new artificial
38:33intelligence for trading in a practical
38:36way, just let me know in the comments.
38:41Write to me if you’re interested, if
38:42you’re not, what kind of content
38:44you’re looking for, and if you want
38:46me to keep delving into it. If I see
38:48support and therefore interest, I will
38:50continue creating content in this
38:52direction. If I see that it’s
38:54something you aren’t too interested
38:56in, then I won’t upload any more.
38:59After having done these nine tests, I
39:01think it’s an incredible artificial
39:03intelligence model, and at first glance
39:06and superficially, it seems a bit
39:07better than Claude. It’s true that
39:13Claude is somewhat faster, but in
39:15general terms, ChatGPT-6 Astra has
39:17capabilities that Claude doesn't, such
39:19as the ability to use concepts you
39:21haven’t mentioned to it at all, for
39:24example, to perform analysis or make
39:26decisions. I didn't need to explain to
39:33it what support is, what resistance is,
39:35how to combine it with other technical
39:37elements within a chart, what analysis
39:40across different timeframes means, etc.
39:42, for it to indicate very precisely,
39:44for example, why the price was more
39:46likely to fall on the chart I gave it
39:48without providing any information. I
39:53was also surprised by its ability to
39:55detect errors and focus on the next
39:57steps. When I gave it, for example, a
40:02strategy with code that contained
40:04several important errors, it didn't go
40:07directly to telling me definitively
40:09this, this, and this. It simply
40:14suggested that there were certain
40:16errors it was detecting in a somewhat
40:18superficial and generic way in its
40:19message, and that perhaps it would be
40:21better to focus the following messages
40:23on finishing the definition of certain
40:25parts of that strategy. Another point
40:29that surprised me is its ability to
40:31understand what you are looking for,
40:33what you want, without me giving it too
40:35many details. In the end, at all times,
40:39we were almost starting new chats, new
40:41conversations, and the prompts, so to
40:44speak, have been tremendously basic. In
40:49this case, ChatGPT-6 Astra was able to
40:51find what I wanted with little
40:53information and focus it, as I
40:55mentioned in the previous point, in the
40:57most optimal way possible. In general,
41:02as I said, I think it’s a really
41:04powerful, really useful, and really
41:06practical artificial intelligence model
41:08for trading. I imagine it will be
41:11interesting for other professions as
41:13well. I am focusing on my own. Anyway,
41:18as I mentioned, if you are interested
41:20in me uploading more content about
41:22ChatGPT, whether it's this model,
41:24others, comparing it with other AIs, or
41:26artificial intelligence in general,
41:28just leave it in the comments and I
41:30will understand it as you wanting me to
41:33keep uploading content. Remember that
41:37in the first pinned comment and the
41:39video description you will find other
41:41links of interest, more content on
41:43trading and artificial intelligence, as
41:45well as other courses, tutorials, and
41:47training, all 100%free so you can keep
41:49training and learning without needing
41:52to invest your money. I'm going to
41:53leave this video right here. I hope you
41:55liked it, that it was useful to you,
41:57which is what matters. And if so, give
41:59it a like, subscribe, share it with
42:01friends and family, and I'll see you in
42:03the next video. Goodbye.

About this transcript

This page contains the full transcript of GPT-6 Astra Es Una LOCURA Para Hacer Trading (Prueba Completa) by Alex Ruiz, generated from the public captions YouTube serves with the video. The transcript has 6,156 words across 917 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.