when I have to kill the long running session that has all the good info and a constructive back and forth because claude just starts getting context window dementia
@JustinBleuel@ChatGPT I have pretty in-depth traces showing that this is reproducible and occurs across different Android devices dm me and I can send you everything I had gpt go deep on it
@JustinBleuel@ChatGPT Also when recording voice notes it auto stops at 10 minutes...
Another annoying thing is the heap is fixed and has poor refreshing logic, each tool call counts as a conservation node, so even with little chatting, but alot of tool activity for big requests, the app will crash.
@JustinBleuel@ChatGPT On Android the playback functionality is incomplete. There's no consistent way to rewind or forward a voice message after a certain length. There should also be a highlighting setting that highlights words as their spoken, and the ability to skip to different sections.
Many people know that zero partial correlation does not entail conditional independence. But do you know that conditional independence also does not entail zero partial correlation?
I illustrated this in a manuscript with a friend!
@lonehiking@thsottiaux Also creating reusable workflows with skills/ agent .tomls, also install token killer on GitHub, there's so much optimization you could do. It honestly should be your #1 priority there should never be friction on building
@lonehiking@thsottiaux What helped me was analyzing my actual token usage, I found what killed me was too many receipts and artifacts that semantically were saying the same thing, and a dirty environment. Spent a whole day cleaning/deduping creating a context registry n now my usage is back to normal
@Un_Ortho_Doxxx@grok@MTSlive There's no real way to understand these things like ur 5 w/ AI When I was using AI to learn harder math concepts I'd ask it to use no analogies just use pure deduction. If that's hard to understand I keep going down deductively, until it's something I can figure out, go back up.
You don’t need to be an entrepreneur to have career compression. If you’re ambitious, high-agency and are a go-getter, you can absolutely have career compression in a corporate environment.
Contrary to beliefs, it’s not about titles, it’s about taking on projects that force you to learn beyond your years and to learn beyond your experiences.
Career compression is a function of your learnings. Your learnings are a function of exposure.
In every well established firm, there is a nicely paved road for the median employee to climb and feel a sense of progression, designed to not overwhelm someone looking for a traditional career.
Since your exposures in this path is capped, so will your learnings.
If you truly want to be radically far ahead of your peers, you will have to be bold and take on “impossibly difficult” projects beyond your years. Seek them out, find them, volunteer for them. If they don’t exist, create them - propose them.
When you learn how to do the impossible, you gain the necessary knowledge, worldview and skills for bigger and better things.
You don’t need to be an entrepreneur to do this - it may be the purest expression of it, but you can certainly do it anywhere.
@teortaxesTex@puckberley@SomethingOnSnow It's interesting why when people see information that challenges their own representation, they don't see it as an invitation to potentially enrich it. China's push towards LLMs and open sourcing it vs. the US genuinely introduces that invitation.