Iâve seen a plugin for ChatGPT where you can connect your trading broker and Chat GPT can provide analysis and answer questions on your portfolio.
You can just dump CSV of your portfolio and for each holding also add current financials from Yahoo Finance or similar. Then use that as the base for questions.
Seems to be an easy feature that Freetrade could implement. Good for hype and getting more users on onboard.
EDIT:
I understand that LLM are not good at numbers. The idea is to use the LLM for aggregation & summarization of recent publications, product announcements, sentiments etc.
Interesting idea though, im guessing well see some sort of AI type integration with brokers in the future to help with various things.
I doubt youâll ever see an AI integration giving investment advice though. But for something to automatically find and pull the latest information on a company? maybe
could be more difficult than you think. integrating an AI is a bit more than turning on an API. Also these AI companies are charging massive sums of money to use their product, its unlikely to be worth it
Could be a hint of mission creep getting into those kinds of features. ChatGPT does have limitations on its recency of information, would bet given the nature of stocks that something more tailor made for stocks would be a better fit.
ChatGPT (or any LLM) canât actually do mathematical analysis (adding up numbers, percentages, interest, etc.) reliably so this is not a very good idea.
Yeah but the idea is not for it to pick stocks for you or to calculate financials but more to aggregate and summarize information on recent developments around your portfolio.
Go to ChatGPT right now and insert the following text I provided below. Itâs exactly 100 words, no hyphens no special characters, easily verifiable using many tools (such as https://wordcounter.net/). Now ask ChatGPT how many words are in the text. You will see it has widely different answers each time you ask and 90% are off by a vast margin (as much as 60%). Now think about what it means. LLM models such as GPT4 canât actually do math, the way they sometimes seem to do math or analysis is deceiving, what they actually do is generate text that âfits rightâ (according to the statistical distribution of all text seen in training).
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Aliquam in finibus felis, sollicitudin vestibulum enim. Duis leo augue, dignissim id sem rhoncus, faucibus pretium sapien. Sed condimentum varius nibh. Fusce feugiat eros a sem ultricies hendrerit. Vivamus ac lectus vitae lorem molestie tincidunt. Integer convallis eget mauris non efficitur. Nam rhoncus fringilla ante. Donec a ex semper, consectetur nisl non, imperdiet odio. Curabitur tellus nibh, dictum nec tortor quis, tincidunt aliquam ligula. Vivamus quis dignissim orci. In cursus aliquet sodales. Donec bibendum erat ac malesuada lobortis. Donec sit amet finibus libero, sed posuere arcu. Quisque consequat urna at nisi placerat.
As an ML engineer, I can tell you that the idea of using language models (on their own) for any analysis/recommendation involving mathematics is a very very bad idea.
As I just mentioned before, the idea is not to calculate numbers. Instead, use it to aggregate & summarize articles/publications/sentiments for portfolio holdings.
Yes, that could be a potential use case. Itâs not completely void of issues (similar to those in math/numerical analysis) though, but at least itâs much better.
The thing a lot of people donât seem to realise is that these "AI"s are essentially a fancy pattern matching tool. They are not intelligent. they donât have a model of the world in their âmindâ like we do. the are basically predicting a reasonable string of words as a response to the input. The massive amount of data used to train the model means they can spit out something very convincing. but they are not thinking and are not in any way sentient.
Youâre missing the point of LLM models. Itâs not whether they are sentient or not - itâs the size that matters.
In the end of the day, think it as an encoded database. Itâs a huge knowledge graph. And thatâs the benefit. Itâs like going to a library, asking a question and getting the answer (most optimized for your question) without needing to search through books and summarize yourself. And because the size of information encoded, it can utilize connections to other topics or find patterns you might not have realised.
Now, I feel itâs extremely useful to aggregate and summarize business related news on portfolio firms.
Rather than run your investment decisions for you, I do think LLMs could help reinforce investor behaviour that results in better long term investing outcomes. Coach/encourage you to stick with the plan, donât over trade, rebalance calmly etc.
Thereâs clearly also an opportunity to automate some customer service. Eg hereâs Octopus energy:
âWe started trialling [AI] with a handful of customer emails with a person supervising it [in] February. By the end of April, it was answering 34% of all customer queries. Thatâs the work of 250 people in the UK alone â and it is doing it with an 80% satisfaction rating. Humans get 65%.â
Yeah, that could be a problem. It might offer you a stock to buy, then harass you with hard-sell for much better investments that only a select few will have access to and only for a limited time - but that you canât miss out on, because it will be much much better than the previous one they suggested. All of this would consume hours of time watching videos that say absolutely nothing except âgive us your moneyâ.
I would then unsubscribe and leave it as a lesson learnt!
Imagine AI hands with all those extra fingers, except replaced by AI Motley Fool emails and âexclusive member eventsâ and endless âextraâ marketing!
I am imagining the pain - then the joy of pulling modules from Hal 9000 one by one to end the suffering for all of humanity.
I canât understand FTs strategy here, as with many of their recent decisions. Are they experts at ML/AI development or integration? No. Is this what the community, users or investors have been clamouring for? No. Will it improve key metrics or business performance? No. Will it cause issues and additional workload? Yes. Is every other feature that could be expected of a broker available? No. Are basic features missing? Yes.
It feels like the crypto thing again. No real strategy, just doing something because itâs the hip new thing. There isnât even a web version yet!
I agree. It may just be that it would be a powerful marketing tool and very much topical from non-FT users perspectives.
It may be that integration would be a relatively simple process whoâs marketing outcomes would outweigh the staff input. FT users have not asked for more advertising as far as I am aware - but it is still an important aspect of the business and brings benefits including increased users, income and therefore increased capacity to develop user-requested features.
Maybe @Freetrade_Team can offer their rationale on this decision?
