Joined April 2015
- Tweets 8,542
- Following 97
- Followers 339
- Likes 441
146 Photos and videos
Pinned Tweet
Introducing Stella Agentic Coding Agents
These work directly with your harness to magnify existing functionality. Give your Claude Code, OpenClaw or Cursor even more power. Any Harness of your choice.
Create An API Key and simply tell it to set up your agentic legion at Kineteq.ai/DeveloperConsole
Over 4000 behaviors with over 100 specialists agents.
Growing More Behaviors every day for your harness to use to gain more capabilities.
- ML Research Focused (Yes we have all you need to do cutting edge ML research. Ask away). 😁🤣😏
1
51
Closed source was always going to be closed access.
This is why it was closed source.
4
We are entering the most important AI decision of the next decade.
Not “which model is smartest?”
But:
Do you want to depend on intelligence you can access…
or build intelligence you can actually control?
Cloud-based frontier models are insane.
Mythos, Fable 5, and the next giants will keep pushing the limits.
- Best teams.
- Best infrastructure.
- Best performance.
But Fable 5 just reminded everyone of the uncomfortable truth:
- That power can be throttled.
- That intelligence can be locked.
- That access can be conditioned.
- That dependency can become a cage.
On the other side, open-source and local models may not always be the smartest out of the box.
But with the right workflows, fine-tuning, agent loops, memory layers, automation systems and execution infrastructure…
you can create something that compounds around your own needs.
Maybe less raw intelligence.
- But more sovereignty.
- More privacy.
- More control.
- More freedom.
The next winners won’t only be the ones using the strongest models.
They’ll be the ones building systems that survive when someone else turns the switch off.
Jun 13
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-myt…
1
4
260
I want to live in a world where no one ever dies again.
I want to live in a world where no one ever ages again.
I want to live in a world where there is no war.
I want to live in a world where there is no poverty.
I want to live in a world where there is no hatred.
I want to live in a world where tribalism ceases to exist.
I want to live in a world where we only have time and a curiosity to expand and explore the stars.
I want to live a world where the dishonest start being honest because the reasons to lie frankly start just ceasing to exist.
I want to live in a world where we know the truth.
The truth -- for once. For every one to just sit around and talk. No more war. No more division. No more bs.
These are simple wants. And with AI all of these things are possible.
1
2
We solve our problems by rational thinking.
We solve our problems by kindness in that thinking.
We solve our problems by compassionate reasoning.
We solve our problems by exposing the liars and finding the truth.
3
The general idea of recursive agentic learning is that we write code, lint code, check against accuracy of what we wanted to accomplish, store the trace, build a primitive for that task, add it to the node, continue. The tree builds itself.
1
6
We do this against every task space.
We separate into two cohorts
Causal | Correlative
Causal Tasks with apply DoCalculus against the Heuristic and compute the rate of error it didn't trigger an event.
Correlative we build a loss function against a heuristic such as (1) does the code run (2) is the finding grounded and the rate that its grounded.
We apply the harness as the forward pass to generate code and then we backprop over the causal and correlative by minimizing the respect heuristic loss function assessing the rate of error against the trigger for causal and whether the code ran and its rate of groundness/correctness against source material.
We do this billions of times with agents. We collect the data and redesign the agentic pass now with primitives of each to direct the forward pass.
1
9
let the agent build the algo itself.
The human mind doesn't have sufficient context.
You degrade the outputs.
4
A model need only be sufficiently intelligent to explore Solomonoff Induction or Universal Distribution of Combinatorial Space given below. AGI requires Turing Completeness and by providing a harness we make an LLM a Turing Machine. Intelligence can be defined as the objective function to minimize the error of a task. An Agentic Harness introduces a Recursive Agentic Loop against the Task Space. We hypothesize that merely constructing a deep enough sufficient harness you can dramatically exceed any pretrain. The underlying model need only be sufficiently intelligent to make turns over the agentic harness to optimize the loss function of combinatorial tasks.
1
10
It is non-differential method of learning that can exist only with iterative code development over that harness.
1) Write code
2) Solve task
3) Working Code --> Pass
4) Repeat
5) Select new Task
6) Form Primitive
1
6
🚨CLAUDE FABLE-5s LATEST BENCHMARKS FOR NON US CITIZENS
JUST IN: Andrej Karpathy, a top AI scientist at Anthropic, is reportedly barred from accessing the company’s most advanced AI model because he is not a U.S. citizen.
17
4
120
8,998
Model providers have a literal financial incentive to undermine any open source development that can run locally for most use cases because their entire business model is selling you expensive tokens that they pay 1 penny for million tokens of compute on. If an equivalent model becomes available that you can run locally, they cease to have a business. AI will compress down into 1% or even .1% of if its size and that will destroy their business model. And they will do everything they can to prevent that from happening.
1
13
The margin spread on compute against what they sell it will exceed 99% one day.
1
9