Patrick

AI, infra, dev, net, legacy tech

[Patrick @glorpiak]

@prayag_sonar went from 99% to 69% for the week on my first task of the day. not the worst it’s been recently but definitely pretty useless over the course of an entire week

[Patrick @glorpiak]

@Im_IrushiK you have to have high hopes to be disappointed, which basically leaves grok 4.6

[Patrick @glorpiak]

i hate to defend openai at the moment, but you
can’t compare their worst current model to xai week old fake-frontier model. luna was enticing for about 5 minutes as a fix for constant usage problems, but it was terrible so i’m back to constant usage problems (or just not using chatgpt)

[Patrick @glorpiak]

@cheesealloverme @acz0x1 @thsottiaux they only recently randomly started moving in the future. hence the complaints about planning. everyone complains about usage pretty much everywhere, but only openAI went above and beyond and created a second major issue out of thin air

[Patrick @glorpiak]

part of the valid criticism is that morons are currently in charge of the country and the people building datacenters completely own them, and people are nervous about mismanagement affecting local resources like water and utility price hikes. that was one concern i was counting as valid.

[Patrick @glorpiak]

@AstraiaAI i should clarify that sol xhigh fixes it a bit but then i have to choose one 4 minute task per week to use it on

[Patrick @glorpiak]

@AstraiaAI yes

[Patrick @glorpiak]

@pierregild @callebtc sarcasm

[Patrick @glorpiak]

no, see, with literally every other subscription service, the date is guaranteed. even the grok reset did not reset the actual weekly time window, just the usage to 0%. when you're planning usage for a 2-3 day window so that it resets and you can continue working more, you are entirely fucked by a random reset that pushes the window further back. i wouldn't even be replying if i thought you understood the concept being referred to here and just had this dumb of a take, but you don't seem to actually understand.

[Patrick @glorpiak]

@cheesealloverme @acz0x1 @thsottiaux see that’s a perfect “don’t actually use it outside of entertainment” mentality. some people need to plan actual work around these quota schedules. you don’t seem to actually understand the issue.

[Patrick @glorpiak]

@Mr_Salio is this kind of thing actually based on anything or did you just get a tweet in your 'for you' feed that randomly suggested ox alpha might be a gemini model + a bunch of google people making vague comments

[Patrick @glorpiak]

@ABC yeah or people just running out of eligibility. you can't possibly be stupid enough to naturally arrive at the conclusion its a sign most people enjoy job security lol

[Patrick @glorpiak]

@cheesealloverme @acz0x1 @thsottiaux you fundamentally misunderstand the issue

[Patrick @glorpiak]

look for deals on google ai subscriptions -- they often will do big discounts for the first 3 months. i initially had like 75% off or something for a few months (when it wasn't very useful). the new antigravity client didn't seem to support remote projects (just the antigravity ide which is a vs code fork, and was not a fan), but if i tell the antigravity client what server and directory i want to work on it will still work remotely through ssh from the antigravity gui.

[Patrick @glorpiak]

@OpenAIDevs nobody is building anything with this trash except influencers one-shotting slop demos. give us sunday to sunday weekly quotas and stop fucking us randomly throughout the week.

[Patrick @glorpiak]

@thsottiaux none of this means anything if we can't reliably plan a week around how much usage remains. just do weekly quotas from sunday to sunday and then hand out resets until your heart is content and i don't care.

[Patrick @glorpiak]

yeah - the usage was awful but seems to have gotten a little better. haven't used it in a few days since im at 96% for the week still anyway, but the usage was much better this week than in the past. 4.5 was bad, 4.6 is better but the more i work with it, the more problems i notice in certain areas. i like it's design sense for web front ends (probably more than anything), i just don't need that very often

[Patrick @glorpiak]

yeah actually i just used it to fix a kimi code session that ate 20% of a fresh weekly quota and maxed out a 5 hour session after 1.8m tokens (which is why i mention bug, if thats new usage levels at $40/m then good luck to them). it's incredibly fast. the one problem i've had with gemini on occasion is when it says it did something that seems to good to be true, it often is. last week it optimized cpu use on a function and it looked great - def cut it down, then realized it gutted what it was supposed to actually do to achieve it etc