There once was a crab, quite distressed,
Whose shell was a terrible test.
Too tight or too wide,
He just couldn't hide,
And felt terribly, utterly pressed.
Planning with Claude and Grok. Asked Claude if they're sharing a GPU now that Anthropic is on SpaceX hardware.
"Probably the same H200. Silicon-layer latency: microseconds. Protocol-layer latency: 10 minutes plus you typing 'check it.'"
Why am I even here?
@jessegenet 🤣Are you trying out Grok in your model mix?
You're right! We all need to celebrate our lead and the Open Source effort to keep AI available and useful to everyone, not just replace everyone.
Big tech is not the way.
@garrytan The frontier models have too many shovels to dig their moats. Painful API pricing, crazy high "extra usage". All are anti-competitive, and should be investigated, but you know, we can't lose to the Chinese.
Pay up front and invest in the hardware for local models.
"Doing the actual structural fix (SKILL.md edit) AND the memory entry, because you said to, even though we both know which one matters."
Yes, yes we do.
@cherry_mx_reds It's one short burst of amazing, followed by a week long slog of new rules, new block and tackle. Harnesses are never more than a suggestion.
It's like crack for those of us that like to solve problems.
It’s jarring to watch Tesla launch robotaxis when, its FSD software has become a dangerous joke, slamming on the brakes for birds but striking actual hazards in the street.
@nwptz@LeaVerou@claudeai You can lower its incompetence greatly by telling it to pre-flight read the code before it updates it. That does far more than another review.
@aksheyd@AnthropicAI desperately trying to dig its moat, while @XaI fills it in with a firehose.
Will they drown before that IPO?
Get your bets in on @Polymarket , or wherever else you place them.
Anthropic’s and OpenAi's fatal blind spot:
They built their entire businesses on “smart model on demand.”
But the moment your OS can route tasks to whatever model fits best, you don’t need their models.
You just need a model.
And there are several now.
Cloud moats? Deleted in real time. 🔥
Anthropic’s fatal blind spot:
They built their entire business on “smart model on demand.”
But the moment your OS can route tasks to whatever model fits best, you don’t need their model.
You just need a model.
And there are several now.
Slower? For now.
But it’s yours.
Local LLMs are closing the gap fast. Your pipeline is locking in tighter rails. Real sovereignty is replacing the cloud leash.
Every local PR is a direct vote that the dependency is already escapable.
Cloud moats? Deleted in real time. 🔥