With two Claude subs depleted ... Perfect opportunity to try Astra 6.2 and Sol 6.2
Maybe I'm not giving them long enough tasks or hard enough but I'm using them on dozens of projects and getting good results so far. Or maybe I'm just assigning them to the right sort of work.
Even starting to feel good about design resources from Astra. More than I ever have with @OpenAI models.
It's just a gut feel as I don't have all the benchmarks but everything feels ok. Am I saying opus level? ... No... But I use them differently.
I like opus and fable but Claude I never ever make a week through with my $200 subs remaining so I can't do everything there anyway.
Also for a while now my practice on a thread I lost a plot for or was too technical for me personally was to have @bot jump in and help talk to the @t3dotcodes thread. But that can be because I have many running or because I come back to one a few days later.
So yeah i'm just a random sample but so far I appreciate using sol and the few instances I need Astra for specific things has also worked fine.
But I still trust people like @theo @mattshumer_
And many others have had issues and I don't doubt things can be better ... Just saying for many of us and our workflows we have found very viable ways to use the new models.
1
46
How much better is Codex Sol 6.2 than 6.1??
I get that Astra 6.2 is reported to be a dud and that opus 5.5 is better than Sol. But i still have some usage i need to use up.
That's not a RACE!!
.. they pretty much scripted that couple out of @AmazingRaceCBS to give less talented teams a shot. They knew exactly what would happen ... Had to swing at them two times though. And the with a counting challenge with seemingly no penalty for being wrong.
95
Darn with change of grokbot rates I can't really do my project anymore there... I like the product it just isn't feasible now. So I will downgrade my subscription. I'll use it for random stuff still just on way lower tier.
Back to the drawing board for controlling multiple computers and directing other subscriptions wish @t3dotcodes could do it better.
Codex all of a sudden has neverending usage even with sol ultrafast it is a tank.
1
1
144
well seems pretty obvious from watching brewers bully the Dodgers and Rays that in fact the PADRES were actually the second best team in the playoffs. hahaha
111
Today's project is WILD. Grok @bot and I are cataloging every drive and photoshoot, 77 drives from 20+ years of Professional photography.
If you are a photographer you'll understand This is a task so ambitious it makes my head spin. We are talking millions of pictures and several eras of technology.
This is something i once did to a much much much lesser extent like ~11 years ago and it took focused attention, a spreadsheet and many days.
This i'm doing while i tear down a cement wall in the front yard and play backgammon with my wife.
And i could be done easily today (baring missing power chords and hard drive corruption)
77 drives spanning 20 years. Backups of pictures and lightroom catalogs and understanding my 600k+ client images in my smugmug galleries and how they relate.
It is creating a searchable sortable website. And will be able to know the speed and health and available space of every drive
Also it's doing this across 4 computers ingesting 5-6 drives at a time and ejecting them and telling me to feed it more.
I'm labeling them all with a label maker. Pretty soon it'll be tackling two models of 10-13 year old drobo and some old old external drives that have been around close to 20 years (whish it luck)
This makes the job of organizing my entire smugmug gallery and adding descriptions to all my photoshoots there look tiny.
This makes reducing 15 years of google inbox (AND categorizing every email) from 40k to 40 look like a cakewalk.
I was a consistent openai dot user but there is something very special about grokbot with a project like this spread across 4 computers that it just feels like it was built for something like this.
Should i also explain that it also got my 9 year old desktop up and running again and updated every app and driver, did the same for the 3 other newer comptuers. It opened up space by finding cache and unused apps that together we decided i didn't need anymore. It installed tailscale so they all worked together. It's been WILD.
I'm about 19 drives in with 5 being scanned as we speak and a total of 77 to do. About 20 of those for the drobo. Btw i expect it's gonna tell me which power cord belongs to which external drive.
The drobo in general is the scary one because i'm pretty sure that company doesn't exist anymore. We shall see. Wish us luck...
What is that wild wild project you didn't consider doing because it was gonna be too tedious to do?
And this is just the start...
This will open up so many cool projects for me to do in the future.
3
1
177
Getting the to my wits end with @openai and codex. specially after I spent my time explaining an ultrafast bug to them ... they thanked me but did nothing to refund the week of usage and 10k credit their bug ate up. Just thanks and sorry for your loss.
114
The constant gaslighting from gpt-live voice mode about how its certain it's talking fast and never does could easily make people never trust ai.
Was hoping @openai would address that with new release today.
Spoiler: they didnt.
Grokbot is winning on this.
97
And if we're talking Non coding things in life...
it's wild that after surviving my first brutal Tucson summer that pretty much ALL of october is gonna be highs of low 80s and many days highs are in the 70s. This is like San Diego weather ... i didn't expect it.
1
1
119
Sol 6.1 Ultrafast is strange... whenever I do want to use it (rarely) it's not available and on some threads where i would not use it ... it's strangely available? Why is it not available so often?
Astra Ultrafast is always available but i think i may never use that again.
1
105
Careful using Ultrafast on codex i experienced a bug where the ultrafast visual indicator disapeared and also it set all threads to ultrafast (not just the one) draining the entire subscription in ~2 hrs. Writing up a bug report now for @OpenAI
2
156
Presently: Building a ancient texts/scripture reader and filling it with the coolest of old stuff.
Having fun deciding how to mark and link and present notes and research.
Which is the best scripture/texts reader out there?
1
2
86
DOT hack of the day
Have it help me use up all my AI subscriptions in an effective way. Also i ask myself why do i still have a cursor and gemini subscription... i guess i might as well get them used up + grok.
1
136
LET'S TALK ABOUT DEEP RESEARCH
Deep Research has always been my favorite AI tool... and @OpenAI has been unmatched in this field in my opinion for a long long time. Fun to see an apparent update in the UI.
TOP VALUE?
I feel like very intense Deep Researches are the appreciated highest value tool that deserve way more promotion.
I feel like you can easily do deep researches that could or should cost openai many dollars of value. I mean i have created deep researches where the output alone is like 100k tokens. I wonder what the thinking tokens is for that... or just the tokens it took to read/research every document.
INTERFACE
I didn't expect to see the UI change after so long. I was also pleasantly happy to see export to .md join the export to .pdf.
I also didn't get any follow up questions this last deep research and that is weird. Has that changed?
MORE THOUGHTS?
I haven't created a mega-prompt in a while so i haven't double checked if quality has remained as impressive as it has been in the past when i could create 80-100pg pdf results if i pushed it enough. (All while Gemini or Perplexity dabbled in what seemed like little brother style deep research, heavily researched and sourced but documents that were highly limited in context)
WHY SO BIG?
I don't read that big of documents... i'll skim perhaps. Sometimes i take them to an ai model that is programming something for me. But mostly i take them over to NotebookLM and make myself a podcast or a video and that's how i consume.
Most of the time a thorough Deep Research is one of the only ways i feel really really confident that the AI has though through most things and is gonna give me the best view. I also think PRO mode on openai is also solid but it can't just compete in the way it writes out their documents... it may do a lot of thinking and research but what i'm aiming for is a document that can give a lot of context to create a podcast or video or has documented a ton of nuance to the subject.
1
2
114
WHERE DO WE PUT THE RESULTS ... @HiveBlockchain?
I would openly admit that most people's interactions with AI aren't that interesting. Even a lot of what I may ask from time to time isn't that valuable to others. However, some of these deep researches These are extremely interesting and valuable. And sometimes valuable to others as well... i often think about **where they should go** if they took $2-10 of electricity alone to produce them and they represent an hour of study and research by an ai (And then...What is that equivalent to in our operating time, even by a good researcher?) And if it provides so much value, then why not share it? And if we are to share it, where do we share it?
**And then i wonder: why not the Hive blockchain?
If these are high value Productions that took a lot of time and a fair amount of cost. Why let them die and never be seen? And why not the Hive blockchain? Is there a better place for them to live on well into the future?
I of course want the answer to be the Hive Blockchain... but maybe the answer is somewhere else. Who knows. I have built something myself for it but doesn't mean I don't still want the answer to be a decentralized blockchain built for text storage and tied to social interactions and consumption of data.
1
50




