This thread is for anyone working on personal projects to share their progress, and hold themselves somewhat accountable to a group of peers.
Post your project, your progress from last week, and what you hope to accomplish this week.
If you want to be pinged with a reminder asking about your project, let me know, and I'll harass you each week until you cancel the service
Jump in the discussion.
No email address required.
Notes -
Ah, how did I ever live without modern AI. Especially the ones with massive context windows where you can throw in absurd amounts of text.
I recently decided to do a case presentation on a very complex patient. As I'm congenitally lazy, I opted for throwing large amounts of (anonymized) clinical encounter records into Gemini 2.5 Pro, and then asked it to turn it into a coherent case summary. It did a bang-up job, no hallucinations whatsoever. I put no effort into categorizing anything, as models these days are capable of figuring out user intent. (While I'm perfectly capable of doing this myself, and was familiar with the case, I believe in better living through technology)
That's nice enough, but what I really enjoyed was the ability to simply ask it to come up with the kind of thorny questions that a senior clinician might throw my way, to put me on the spot. This is an act that, for reasons unclear to me, is known as "pimping". It's a favorite pastime of many a consultant, aiming to pop the ballooning hopes and dreams of med students and residents alike. It did excellent, coming up with all kinds of interesting lines of questioning, asking me to justify my own suggestions about future management decisions. (Something I did myself, for example, I'd noted unusually rapid decreases in benzo dosage, my other boss is rubbish at this, especially given that he asked me for my suggestions on that front in a different case)
Ah. Feels good. So much scut-work saved.
The actual case presentation went great. My supervisor didn't ask me anything remotely as difficult as the worst-cases I'd prepared for.
Do you not feel an issue with context lengths? I have not used Gemini but with Claude I don't feel the context length is that amazing for my purposes which is programming documentation.
I love the idea of massive context but in actuality I have to manage this quite actively.
Gemini has an enormous context length. I think 2.0 Pro could accept 2 million tokens, and 2.5 Pro is a regression back to a million. That's still way ahead of the competition.
There's very little I personally need to do that requires CLs that long. I think the most I've ever used in a natural manner was when I was translating a novel and used up 250k tokens. I've used more for the hell of it, but never actually maxed it out before I got bored.
Models are also getting better at making good use of the additional context, but there's still performance degradation that's not captured by benchmarks like needle in a haystack tests.
Give Gemini a go, preferably through AI Studio. I think it'll do much better than Claude. I know some documentation in programming can be enormous, but probably not 400k words enormous? You could alternatively enable search if the docs are publicly available on the web.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link