This is true for some classes of strategies. At the same time there are strategies that can be profitable on longer timeframes. The two worlds are not mutually exclusive.
Yes, but LLM can barely cope with following the ordering of complex software tutorials linearly. Why would you reasonably expect them unprompted to understand time any better enough to trade and turn a profit?
Exactly. If it can't distinguish between a basic repeat after me ordering how is it going to even get a simple output order correct? Let alone pull apart the strategies themselves
This is very thoughtful and interesting. It's worth noting that this is just a start and in future iterations they're planning to give the LLMs much more to work with (e.g. news feeds). It's somewhat predictable that LLMs did poorly with quantitative data only (prices) but I'm very curious to see how they perform once they can read the news and Twitter sentiment.
I would argue that sentiment classification is where LLMs perform best. folks are already using it for precisely such purpose - have even built a public index out of it
sorry dude. tried going down the rabbit hole but I'm too lazy and uninterested in it. read about it month ago or so. perhaps Daily News Sentiment Index uses LLMs, not sure. if you go long enough through https://quantocracy.com/ you should be able to find it
Not just can i guarantee the models are bad with numbers, unless it's a highly tuned and modified version they're too slow for this arena.
Stick to using attention transformers in better model designs which have much lower latencies than pre-trained llms...
I saw this idea implemented in the book "Extreme privacy: macOS devices". The author also provides importable profiles that you can switch between, e.g. to enable/disable security updates. I haven't tried them yet, but I am now more motivated to do so.
What is even more shocking is running an Android simulator in the same context. Literally dozens of Little Snitch prompts before the OS even boots to the lock screen. Not defending Apple here, but when I was developing a mobile app in both Xcode and Android Studio I noticed a marked difference in the amounts of phoning home.
On the other hand, this particular argument also gets overused. Not all compute-bounded parallel workloads are easily solved by dropping into multiprocessing. When you need to share non-trivial data structures between the processes you may quickly run into un/marshalling issues and inefficiency.
This sounds a lot like my own story. Your last sentence in particular sends chills down my spine, because that was exactly where I was stuck for perhaps 15 years. (I was diagnosed at 14 and CGMs would not be available for more than a decade still.)
Please don't give up! It's absolutely possible to break through this mental block and improve your results and mindset step by step. It is a tough journey, but it can be empowering.
I don't know where you're at now, but for me the change started with honestly tracking my data and looking the truth in the eye. I learnt to review the data regularly and think about my results, and the possible interventions I could do. And, well... it's been like that ever since :) Endless incremental improvements and occasional setbacks, but with a clear trend towards higher TIR and lower BG variance.
Many years ago, I went a similar route and built a small T1D monitoring stack from scratch for myself. It even pretended to be a Dexcom Follow client so I could get CGM from Dexcom in near real-time. When Dexcom eventually changed their internal APIs and broke my data ingestion, I decided to finally give Nightscout a shot. I've never looked back since.
As I see it, the big advantage of Nightscout is that it is a de facto standard interface, with many integrations already existing. And it's easy to build add-on apps on top of its API. I've built about four myself [0], [1] and there is a big community of users and developers building other things such as [2], [3], [4]...
Even though Nightscout is a little bit messy and requires MongoDB, it's surprisingly easy to self-host. I'm using the stock docker-compose file from the main repo with only minor modifications. I run it on a $6/mo VPS. As an alternative, there are two or three hosted Nightscout services costing little more than that.
I highly recommend you to consider going this "standard" Nightscout route because it can save you work (and worries) in the future, and you get to connect with the community around it. In my experience, going all alone from the start was not worth it.
I've recently switched from BasalIQ to ControlIQ on the X2. I very rarely feel like I'm on a "closed loop system". It's super conservative and left to itself it consistently keeps me higher than I'd like. With BasalIQ I could at least set up a low-enough target BG (5.0 mmol), now I'm stuck with 6.1 and the pump is actually happy to leave me at 7-9 mmol for hours, with very feeble interventions.
Long story short, I'm planning to buy a few boxes of Omnipod Dash pods, in a country where I can do so, out of my own pocket – so that I can try iAPS. #WeAreNotWaiting and all that :)
BTW Nightscout should not be required for Loop nor iAPS.
