To be honest, I was expecting way more RAM on the new Mac Mini and Mac Studio. There’s still real local AI potential here… but I want to run the bigger models like GLM and Kimi properly. Bigger is better.
76
1M context doesn’t make an agent 2× smarter. It gives it 2× more runway before it compresses and forgets your debugging trail.
Grok 500K: better for focused build/test loops.
Claude/Codex 1M: better for monorepos, migrations, audits, and long messy sessions.
1M maps the battlefield.
500K executes the mission.
1
35
Everyone wants bigger AI models.
I just want them to fit on my computer.
Quantization wins. 😂
🤖 Made with AI
1
2
39
Replying to @Square
@Square gave every merchant a way to get paid.
Now I want to give every merchant something even more valuable: scalable word of mouth.
No more renting attention. Let your customers do the marketing.
viralref.com
14,150
Replying to @DaveyWithCV @withengine
Not yet. Just launched it today. The strategy is intentionally conservative, so I’ll let the live track record speak for itself.
33
Replying to @DaveyWithCV @withengine
No. Your funds stay in your own Hyperliquid vault. Engine gets revocable trade-only permission through Privy, but no withdrawal rights. You control the wallet and withdrawals.
1
49
Replying to @heyatif @withengine
Haha thanks! 😄 It’s actually named after ret_90d_ex7, the 90 day momentum factor excluding the last 7 days.
1
42
Autonomous Grid is like OpenRouter, but actually open. 👀
autonomous.ai/grid
26
Been playing with @withengine and made my strategy, Ex-Seven, public.
Long-only US equity perps rotation: top 4 by 3M momentum, buys sharp dips in uptrends, goes flat below the S&P 100-day EMA, and scales risk down with drawdowns.
Just experimenting. Not financial advice.
Replying to @chamath
The company on the right isn’t promising open models forever. It’s using openness to catch the leader. Once it’s ahead, don’t be surprised if the playbook changes.
187
Good job, @grok ! You’re winning over everyone. You’re the most attractive model right now.
1
27
AI agents don’t fail because they’re dumb.
They fail because reliability collapses exponentially.
At 95% accuracy per step:
10 steps → 60% success
20 → 36%
50 → 8%
100 → 0.6%
The next AI breakthrough won’t be more intelligence.
It’ll be not screwing up step 37.
54
📈💸 Looks like even the servers can’t handle all the $$$ flying around…
Too many $ for @PowerballUSA 🤯🎱
#Powerball #TooManyRequests
913






