It does create easily-read English language text. But it's basically just sort of word salad. All the sentences make grammatical sense and mostly sound all right, but there's really not much to it. It seems to have assembled a collection of phrases it found on the internet into something that seems like it addresses the question, but that actually are just pasted together and sound like they are saying something of substance, but aren't.
There is also the issue of factuality that @SMCJB references. And as a trader, I found this "discussion" of the issues in this chat to be relatively empty, much like something written by someone who doesn't know what they're talking about, but who can sling the words pretty well and can use Google.
This does not mean it's not a very interesting achievement, but it isn't a meaningful post in the forum, within the meaning of what this forum is here for, which is discussion between traders.
It's somewhat like a particularly good spambot. We normally just delete those.
Since this post was clearly marked as having been generated by the program, there's no problem with it, but going forward I'm going to ask that members do not post chatGPT output as part of posts in regular threads having a trading subject matter, even when labeled as bot-produced.
The problem is that they've got no real content, could easily lead someone astray who takes them seriously, and are not the actual informed thoughts and experience of real traders. They are an amalgamation of things the bot found on the internet, hashed together and presented as if written by a human being.
It would be perfectly all right to start threads devoted explicitly to machine learning or to this particular program within machine learning, or to some application of the program, and to explore these interesting topics. Just not in threads that do not have that type of inquiry as their subject.
Also, probably someone would be able to spoof people into thinking that bot output is regular human output, made by a real member. There's not much point of this, in terms of what the forum is intended for, and clearly would not be appropriate here either.
Basically, if someone wants to look into this type of machine learning as a subject of investigation, that is fine, but not as part of the content of a post on a separate trading-related topic. Any thread topic specifically about chatGPT or similar programs would be fine. We should just keep the topic distinct.
Thanks.
Bob.
When one door closes, another opens.
-- Cervantes, Don Quixote
The following 4 users say Thank You to bobwest for this post:
I have asked chatGPT to summarize your last comment...
---
The text discusses ChatGPT, a machine learning program that creates easily-readable English language text. However, the text generated by ChatGPT lacks substance and is not based on fact. It is composed of phrases found on the internet that are pieced together to appear to address a question, but do not actually convey meaningful information. It is suggested that ChatGPT output should not be included in posts about trading, as it could mislead readers and is not the product of real traders' experiences and thoughts. It is suggested that threads specifically about ChatGPT or similar programs would be appropriate, but that such content should not be included in threads about other subjects.
--- can you condense it further.
---
The text discusses ChatGPT, a machine learning program that generates English language text. However, the text lacks substance and is not based on fact. It is suggested that ChatGPT output should not be included in posts about trading, as it could mislead readers and is not the product of real traders' experiences and thoughts. It is suggested that threads specifically about ChatGPT or similar programs would be appropriate, but that such content should not be included in threads about other subjects.
--- can you condense it even further
---
The text discusses ChatGPT, a machine learning program that generates English language text that lacks substance and is not based on fact. It is suggested that ChatGPT output should not be included in posts about trading, but that threads specifically about ChatGPT or similar programs would be appropriate.
The following 3 users say Thank You to trendisyourfriend for this post:
This is pretty good. The AI condensed it down well, and in the subsequent shorter versions it did even better.
The whole thing is interesting and in another context could be a good subject for discussion and exploration. But AI output in a trading thread is probably not something that we are going to consider as appropriate, for the reasons stated.
We may formalize this policy after some internal moderator discussion, but in the meantime let me just ask for members to not intermix AI output in normal trading threads. At a minimum, any AI-generated content needs to be identified as such (as it has been here), and there needs to be a useful purpose for it (which is why I suggested keeping it in threads devoted to AI, rather than to normal trading.) Certainly, "useful purpose" is always a matter of judgment, so for now let us say that just displaying the abilities of this AI to create chat does not fall into that category, on a trading forum at least (on an AI forum, that would be different.)
I'm not criticizing and I'm not being close-minded or obstructing technological progress either, I hope, but let's just keep the use of AI's separate from normal non-artificial conversation between actual real traders. Real traders talking about real trading is the point here, after all. The whole discussion of what chatGPT would say on the topic, and of chatGPT itself, is very off-topic here.
Bob.
When one door closes, another opens.
-- Cervantes, Don Quixote
The following 5 users say Thank You to bobwest for this post:
It's clear that the markets are manipulated, through a process called "gunning the stops." If your firm owns an order-execution desk, then it is simple to funnel stop information to your hedge fund. You then move the market toward the stops, which triggers automatic sell orders that move the market even faster on the short side.
It's why, fro example, Citadel owns Citadel Securities.
The following user says Thank You to map367 for this post:
It does but in its view, we (well, I assume most of us here are retail traders) are too small to care about exclusively. We hardly blip on its radar.
On other hand, your broker does know your position and your broker also utilizes it in his advantage whenever he can, also he can easily take copy positions to you or against you. But that only happens with slightly shady brokers, and even then he only cares about you when you have been identified by its algorithm as a profitable account over certain period of time.
I can say that with certainty because I have brokers franchise, not too long ago I could even see positions of people who opened accounts under me in my broker backend terminal, but that is also gone now. But the point remains, they do have access to your trade data and can also utilize to their means, if they wish to.
So rather than worrying about markets trying to flip you upside down, worry about brokers.
I guess this has been said many times in this thread and on this forum in other discussions, but we are far too small timers to worry about these things. Even 100's hardly makes blip in a tape of instrument I trade.
Just my 2c.
The following 2 users say Thank You to AllSeeker for this post:
Funny you should mention Sam Bankman-Fried. The reason they got in trouble was because they were using arbitrage market making strategies, and lost their edge. Someone else was able to execute faster. So they started to make up for it by taking on more risk, and that's eventually what blew things up. I would not consider this pattern recognition. Rather it's looking for a pricing discrepancy. They know what the price should be based on prices in similar instruments, and adjust their bid/ask accordingly.
Not that a statistical inefficiency can't exist. There's some evidence that traditional technical strategies did work prior to the advent of computers in the 80's. Statistical inefficiencies get harder and hard to find as the competitive market eliminates them. To the point that you need a masters or phd level understanding of the stuff to even be competitive.
The larger problem though is that retail traders don't have the data or expertise to really verify if there's an edge there in the first place. The more fat tailed a distribution is the more data samples you'll need to see the true profitability. It scales exponentially. So for the kind of scalps that order flow traders try to do you'd probably need 10,000 samples over 5 years or more to get any meaningful data. Data providers usually don't provide order flow information beyond 1-2 years. I can backtest with tick replay in NinjaTrader 8, but I have some reason to believe that's not very accurate. That doesn't do historical market depth either. So better start recording it all yourself now, and maybe in a few years you can try to find a pattern.
As for the DOM, the information that I'm paying attention to the most is only found on the DOM. Some of the best empirically proven stylized behaviors of the order book have more to do with the latent liquidity. Many of the resting orders are just holding a place in line. Other liquidity providers don't put their liquidity out until it is about to trade. So you can't just look at the resting orders. You have to watch how the resting orders react to the market orders, and how much stacks or pulls each time an attempt to clear a price is made.
- SpeculatorSeth
The following 2 users say Thank You to TWDsje for this post:
I guess it depends on the market you are playing with.
Big banks/Big brokers always hold a huge order book so they could know the position clearly, in turn, they could make use of it to make money, i.e pushing the price to the stop loss areas triggering moves.
Then we come to a question: what makes the extreme move during economic figures?
I guess no one could read all the information within one or two seconds, non they are confident about AI reading news headline and react.
What we always see is, when coming to economic figures/event, sometimes the market just go one way regardless of what the figure is, say USD going strong against all currencies given a dovish Fed members comment. And after that, news will just comment "priced in" or crowded position triggering trend reversal. etc
That might be a lot of stop loss orders in certain area and when the figures come out, the market just react in a kneejerk way and trigger further moves.
If we view from this perspective, we sometimes "feel" somebody is watching us but actually it is more like from crowded position or similar activities.
@32:00 Talking of order execution and HFT's- "the shorter you go, the more important it is to extract your alpha, without giving away your intentions to the market place, as soon as they figure out your intentions, they're going to come and get you ........... buyer beware"
This is a guy who is the CEO of a 8.7 BILLION dollar hedge fund ..........
