Last Notes
When all your income is in bitcoin and so you just got a 10% pay cut in two days while everyone is celebrating
https://relay.utxo.one/adadfca06d6946031a51feeaf35cb6ac166aba04155e2a965fae8151043ba140.mp4
At this point the only thing I am comfortable recommending is people move from Windows to Linux
Everything else shieeet idk bro
Doubt that, not sure trying to stay humble, but if models keep getting better that will make hardware more expensive I think
Don't understand why anyone would use Gemma, it is inferior to qwen in every possible way
If I can get glm 5.3 running locally, legitamitely will never need a cloud AI again
Deepseek and qwen are sick as private agents but not quite at that Claude level, but glm 5.3 actually is there
The new ones look dope for real, old ones kinda fruity ngl
Trust the leader forks, don't ask questions
Everyone's doing this pattern after buzz
Grok bot
Hermes bots
Slack
#nevent1q…zqm5
I strongly recommend anyone on opencode tries using prime-agent for a few sessions and see if it doesn't both produce better output and save you a lot of tokens
Many people on nostr have 😘
Everyone I know who made millions on bitcoin, none of them bought lambos
Price action is looking scammy and fake as hell
Who the hell uses 23 apps
Unless the entire network all trusts the same single party
Good morning
A signer app has an equal chance to have an exploit as a client
A signer app being a magic bullet is like saying an air gapped calculator Bitcoin signer will definitely keep your bitcoin safe
Nothing is safe, risks exist everywhere
Ok good clue. Which os are you on?
Is this old or new?
https://relay.utxo.one/2d71d311a37f40ee27acc951550333c63d130283d3600fb32da8869ce11cab3e.png
How I feel grabbing Uber eats outside my door
https://relay.utxo.one/20d0392594022b942ddcd134beba1291d86d6b15a4f527aea06dd23673ef5c2c.mp4
Sparks also scale in a nice way, speed and concurrency improve the more you add and don't need a whole rack to do it, other than the racks of cash 😭
Stripe Buying openrouter is very bad for openrouter users
Stripe is a ban happy KYC as fuck company
Nvfp4/mtp3 - and consuming less than 80w each. Older GPU not nearly as efficient, shit runs nice on Blackwell, ampere ghetto
Cryptocurrencies :wisp_lol:
Qwen3.8 pushing 310+ tok/sec with concurrent requests on my dual spark setup
Single chat not so impressive but if you wanna run dozens of agents it's actually insane for the price
https://relay.utxo.one/3c1596b01c46729053d85059dd34f3ef1b83251310ae7f2fe08642765657ccd1.jpg
Appreciating what you have honors what others have lost
Good timing
#nevent1q…39as
Claude usage limits will drop by 50% at the end of August btw
Didn't learn to stay humble
https://relay.utxo.one/d392368bacf99354567b88dc720df8a2dcaf725f71009d11c11911d0c9b7185e.jpg
Go to your interface settings in wisp and disable auto load media - will cut it down like 90%
GPUs in your home brother, you can run qwen3.8 for about $2k and it's actually good
Try using prime-agent it's way better with ctx management
@nprofile…6chp gaining a lot of followers today
Just the way they trained it I guess, it's super annoying to actually work with live so I'm guessing that's why other models don't do this. But if you leave it on overnight it's pretty nutty
This is why qwen scores so high in benchmarks, it actually is insanely smart, if you can stomach waiting for reasoning to complete
So be sure to adjust for that in your workflows
https://relay.utxo.one/d719fec37172be2dbfb7fe4812e0b9810b64f06792a6b6cceaf2db66491c363d.jpg
Let qwen run all night on some codebases I made with opus 4.6 - 4.8 and it found several bugs that fable confirmed are real
Those reasoning tokens are slow and annoying, but they work
Dedicated to old millennials
https://relay.utxo.one/ee37325b9b253e919ec71e4ee14abf8b351057510854f9be3a4e6c5932c910dd.mp4
I hope they see this note and do the right thing (doxx)
