I try to be positive, but the internet is the internet
trying to stay out it
Joined March 2026
- Tweets304
- Following49
- Followers5
- Likes91
Tok/sec means nothing if you dont state your KV cache quants and context size
Feels like “tok/sec” is losing its meaning here.
I keep seeing numbers that are nowhere near achievable in real-world tasks.
I get that X’s algorithm rewards this kind of hype, but it’s a shame to see some genuinely great custom inference engines putting their reputations at risk just to chase impressive numbers.
1
20
Qwen 3.8 27b was, in many ways, on par with Opus 4.6, so I assume they are aiming for Opus 4.7/4.8 levels.
3
As an American, I would like to apologize to the rest of Earth for making the world unsafe. Our leaders are fucking idiots, and we were stupid for electing them.
🇯🇵 I’m Japanese.
I have a serious question for Americans. 🇺🇸
The conflict between the United States and Iran continues to raise fears of a wider war.
But as a Japanese person, there’s something I genuinely want to understand.
What is America’s ultimate goal in this conflict?
Is it to prevent Iran from developing nuclear weapons?
To weaken Iran’s military capabilities?
To change Iran’s government?
Or to establish lasting security and stability in the Middle East?
And what does Iran ultimately want?
What would victory actually look like for either side?
More importantly…
How many lives might be lost before both sides decide the fighting must end?
I’m not asking who is right or wrong.
I want to understand what both countries are trying to achieve and whether peace is still possible.
Americans, what do you believe would be a realistic end to this conflict? 🇯🇵🇺🇸🕊️ xitter.gq/IRIran_MiliarY/status/…
1
This is why a dual GPU setup is invaluable to me. I dont want to sacrifice either.
7 months since i last logged into my Steam account.
Local AI completely took over. No more games, my GPU runs models now.
I even deleted most of my games to make room for models.
Crazy how much things have changed lol
Any sacrifice like this you have done to accommodate local ai anon?
1
Actually Qwen 4 is coming out this week so this is probably going to age like milk.
Anyone saying open weights models like GLM 5.3 or DeepSeek 4.1 flash or Qwen 3.8 flash next ... Are good enough, that infra and sysadmin /sre roles are at threat, are either all people being paid to say it, or themselves have not used these models. Or they are people who don't understand this space.
Open weight models like these are barely able to even do the most simplistic stuff. They are generations away from being even marginally component.
31