beret4breakfast

Question about system prompts in general. How are they affected by context rot/growing context windows. Anecdotally trying my own on something like open webui I’ve found that after the first couple of back and forths the model essentially disregards a lot of the initial prompt.

Would also recommend soldering wick - it’s just copper braid that you heat alongside the solder you’re removing and it ‘wicks’ away the solder.

There’s an excellent book with the same title by Tom Whipple that id highly recommend.

For the purposes of learning, I’ve built a chatbot using ollama, streamlit, chromadb and docling. Mostly playing around with embedding and chunking on a document library.

Or use calibre web. Which has a less functional but nicer looking interface (with user logins)

It’s a bit disingenuous to have phrases like “start planning for free” when it’s basically just do qualifying questions then hit a paywall. Really needs clear pricing up front, no way to see it without trying to set up a plan. It’s also not clear what paying actually gets you (number of plans etc).

It feels like there’s a significant lack of “ML/AI” based approaches applied to these kinds of problems. I’ve seen a lot of example of RL/GNN papers that do attempt to solve smaller problems but it always feels like the best option is to just pay for a gurobi license and have at it. I’ve been doing some scheduling optimisation recently (close to job shop scheduling) and while there’s some examples of using RL they just don’t seem to cut it. I’ve resorted to evolutionary algorithms to get reasonable solutions to some big problems. Maybe it’s just always more efficient to using OR type approaches when you can formulate the problem well.

I’ve been using obsidian for this for the past couple of years. I have lots of folders and notes for work projects along with notes for learning topics. I keep track of tasks and to do lists for me and people in my team (using #people/name tags).

I have specific updates notes for each project. I add a weekly header (via a template) in reverse chronological order. Pull the key updates into that (I also have a todo for each project that pulls in tasks from all files in that project subfolder - so picks up meeting actions)

I also put them all in a canvas so when I doing updates about projects they’re all in one place.

I do the same for people I manage so I can keep track of updates and todos when I’m meeting with them.

My go to start point is a daily not where I throw in new tasks (tagged to projects or people) and those get pulled into the corresponding project/person file.

Because it’s markdown with lots of plugins it helps me do things like diagrams (mermaid diagrams) etc and excalidraw/canvas for architecting/mind map style ideation.

My Apple Watch. There are very few times where I need instant notification of things and they became distracting - I instead now just check periodically. Granted I probably don’t stand every hour anymore.

I’m speaking as someone who has never had to price anything. To me it feels like a service that doesn’t give me monthly benefit (i.e. frequent recurring usage like my iCloud storage or Netflix etc) as I wouldn’t use it often enough, so paying monthly just feels like a waste to me. The difference between monthly rate if bought as a year and per month is pretty big to incentivise people to pay up for a full year - I personally dislike this strategy, it implies to me that I have to pay a 50% markup just to have the option to change my mind after a month or two (to me this implies I might, so best to get all the money up front).

I think you want to incentivise long term ownership of the product (MRR->YRR) with a focusing of the price at yearly. It aligns more with financial years / tax filings and the general cadence of people looking at their finances.

Now as I say I have absolutely no experience and haven’t done any market analysis. So this is just my opinion.

Pricing is a super tricky issue, especially when its for something related to financials. I think there are plenty of people who wouldn't bat an eye at $14p/m (and maybe they are the target audience), but there are also plenty of people who aren't buying $14 cocktails.

I fall into the camp where I like to play around with spreadsheets (free -£time) or other online calculators (free) just to make loose projections and see where I'm at. I make a good salary (for the UK) but don't pay for any financial advice (and probably wouldn't unless my situation got really complicated). I think if it was priced at £30-50 a year or £4-5p/m I (personally) might go for premium because the tool is honestly really great!

One completely different avenue you might consider is to selling direct to companies to provide to this a service to their employees. I'm not talking financial companies, but more as part of the benefit package. My company added something to their benefits, called bippit - which is awful (or at least not useful to me), but provides virtual one-one financial advisors and some very basic online tools.

This is fantastic, smooth and really very polished. It clearly covers a lot of stuff and you’ve really got a lot covered so great work. I’ve not sat down and tried to fill out all my details as I’m worried I’ll lose all my details and have to start again. But this kind of stuff I find really interesting and most online are really bad/basic.

Main concerns are around price tbh (I’m a cheap bastard) and I think $14 a month is steep for something I’d use loads one evening and then probably use once every quarter after that.

I think the persistent storage is the key thing to make someone want to pay and the way it’s presented feels wrong to me. If I revisit the site you basically say ‘hey I’ve got all your data stored, sign up to premium and you can have it back without entering it all’ to me that comes across a bit of a ransom. I think that’s fine I’d say a couple of days/weeks have passed. But I think if you Accidentally closed the tab or it timed out it might put people off.

I think the real money maker here is likely financial planners who can pay a large fee and share/use it with their clients over zoom or whatever. Or perhaps some kind of insights/ yearly/monthly forecast summaries for a fee.

All in all amazing job, I hope it’s a great success.

I get his point that most work that uses code is the combination of the technical structure/actual code and the creative elements that combines them in different ways. But I think it’s an act of hubris to assume AI (or whatever you want to call it) can’t, now or in the near term, do the creative part.

Hence why 3pa like Apollo are great, it’s video/gif player is (was :’( ) the best.

Wait what? Is that just mobi? Or can other files be used. My USB port broke years ago so I soldered on a qi charger and now just send books over WiFi.

you need to login with the default un/pw at the top of the page

probably a ad/popup blocker