Pinned Tweet
Just got diagnosed with fibromyalgia yesterday.
Been dealing with the pain/fatigue for a while and finally have a name for it. Not looking for advice — just wanted to be real with y’all. Appreciate the community here. Still all-in on the AI stuff, just navigating this too.
Every day I smoke my pain away. 💨💚
Not to escape life, but to survive the relentless pain of living in my own body. From the moment I wake up until I fall asleep, it never stops.
I'm not chasing a high. I'm chasing a moment of peace
Some days, just existing is exhausting. 🖤
14
Peter retweeted
Imagine having enough peace in your life to argue with complete strangers on the internet about whose beef is better
God, I envy having problems that small
Be grateful you can afford to eat and have two functioning hands to cut your own food. Some of us are struggling with both
Replying to @NoDataSold @thsottiaux
GPT-Live-1 next pls — let Astra handle more of the heavy lifting so Live gets cheaper + more efficient 👀
46
Apparently this is what happens when you give me a PC, too many AIs and no adult supervision. 😂
4
63
Weirdly useful discovery from last night: Hermes 0.21.3 doesn’t expose openai-codex in the Hindsight setup UI, but the underlying runtime accepts it and passes it through fine.
I’ve now got local embedded Hindsight running GPT-5.6 Luna via ChatGPT/Codex OAuth — no separate API key needed for the Hindsight LLM path.
Important caveat: give Hindsight its own dedicated CODEX_HOME and do a fresh device login there. Don’t let a long-running Hindsight daemon share the normal ~/.codex refresh-token chain.
Retain/recall, provider telemetry, logical backup and scratch restore all proven working.
Version-sensitive obviously, but this is real and pretty damn useful. 🤯
@Teknium @NousResearch
1
5
130
Wow lol deepseek dropping deeppro for deepflash 4.1 🤖🔥 14th September beijing time.
So basically DeepSeek is saying: “V4.1 Flash has surpassed Pro enough that we don’t need to keep Pro around.” Their own claim is that Flash is better on performance, cost, speed, and task completion time.
Interesting
4
149
What a joke Claude 28% of my 5h usage to just fucking compress context in a session im working
4
104
I will try it if potato @poteto potato or @elonmusk or @cursor_ai team fix my cursor usage please. Cant afford your pro atm. Still wont let me message still saying im stuck on 56% but cant message it.
36
I think it’s time to drop the Claude sub and try Cursor for a month, see if they’re actually more user-friendly. If I’m paying every month I want to spend my usage working, not watching the product fight me. 😂🤖
@AnthropicAI 🤦🏽♂️🚫 @cursor_ai 🤔🧐
29
At least openai have given access to there new ai on there oauth sub on day one and keeping it there and not fucking gating it to there api anthropic its all money games.
12
Claude is an ok model but the devs bro are very different to Hermes, OpenAI, SpaceX. They don’t seem to prioritise the user, they prioritise constraints, limits and controlling how the product is used. The model is capable as fuck, but the experience around it constantly gets in the way of actually using that capability.
Put this way openai when your usage is about to die codex finishes and pauses cleanly. Claude code will cut your task and work and leave you to wait in a unknown state.
71


