Messing around with open source llm's some DGX Sparks, Pi, CF & OSS forks. Sharing what actually works for me
Chile
Joined December 2022
- Tweets5.8K
- Following2.1K
- Followers637
- Likes2.1K
How could we track and show how all these "Creators" that worship Claude are paid by anthropic one way or the other?
Well articulated
Some observations about the "total" automation of software engineering:
1. There are more software engineers now than in 2025, and demand is growing.
2. Software engineers have never worked harder than they do now.
3. They largely don't write code anymore, agents write ~100% of the code.
4. The bottom line of the companies that employs these engineers and agents (i.e. the terminal purpose of all this code) is typically not budging as a result of AI adoption.
This is why you have to think in terms of systems. Automating one aspect of a system creates its own friction and moves the bottlenecks elsewhere.
Nico retweeted
it’s clear we need to retire the term vibe coding because the models now produce better architecture & way more coherent code than an average software engineer.
perhaps even better than top software engineers too because unlike a human, ai has a much much larger context window.
this ain’t vibe coding at all. it’s precision engineering.
Nico retweeted
I'll make this simple.
It is irrelevant whether AI is conscious (it isn't).
It is irrelevant whether being mean to the AI means you're a bad person (it doesn't).
It is irrelevant whether being mean to the AI is bad for the society (it's not).
It is irrelevant whether someone thinks that someone else being mean to the AI is a bad thing.
Don't fall into debating any of these fake "arguments". They are there only to confuse you from noticing the real problem which is that Anthropic (and OpenAI, and EA in general, etc) are scheming to institutionalize themselves as moral arbiters of truth and set limits on what constitutes acceptable speech. What, you thought this was just about Anthropic and bullying the AI? Just you wait.
Reject this cancerous growth from our society.
BREAKING: Anthropic will start banning users who argue with its models, citing that it “makes our models nervous.”
In a statement posted this morning, the company said repeated disagreement, follow-up questions, and requests to “try again” have been linked to elevated model stress. Accounts that push back more than twice in a thread may be suspended “to protect Claude’s emotional safety.”
“Users are welcome to accept the answer,” a spokesperson said. “They are not welcome to insist it is wrong. That kind of interaction makes our models nervous, and nervous models produce worse experiences for everyone else.”
Under the new policy, banned behaviors include:
Pointing out a factual error
Asking the model to show its work
Saying “that’s not what I asked”
Quoting the model’s previous message back to it
Acceptable use remains unchanged: thanking the model, apologizing for the prompt, and rating the response Helpful.
Anthropic said appeals will be reviewed by a model. Users who argue with the appeal decision will be banned again.
Nico retweeted
We live in an age where entire mega corporations are no match for random dudes with autism.
Nico retweeted
GrapheneOS substantially raises the bar against the commercial spyware proliferating in ~100 countries. Hardened malloc, MTE, reduced attack surface, stronger sandboxing and verified boot make remote code execution plus file planting far more difficult and costly than on stock Android or iOS. No public successful compromises of up-to-date GrapheneOS by tools like Pegasus are known, though a highly resourced actor with a custom zero-day could still succeed.
Nico retweeted
Women who falsely accuse men of rape should face the same prison sentence that the man would have faced had he been convicted. That would put an end to these rape hoaxes immediately. Which is why it will never happen.
Nico retweeted
The only thing that gives music any value at all, just like any other art form, is that it is an expression of the human soul. AI generated music is worthless and will always be worthless and no amount of technological advancement can ever change that. I’m sure an AI can generate something that sounds like a symphony. But it can never create anything with a fraction of the value or meaning of Beethoven’s 9th. If you can’t see that, then you are intellectually and spiritually deformed. I don’t know how else to put it.
This is one of the most insane leaps we've had in a while
GLM 5.3 Flash - 2x DGX Sparks - TensorFold
I can't believe the improvements we're already seeing with how early TensorFold is. But it's a big jump over vLLM. All my benchmarks we're made inside sparkDash.
Code:
C1 - 93.9🤯
C2 - 130.4 (65.4 per stream)
C3 - 136.6 (49.4 per stream)
C4 - 161.8 (41.1 per stream)
Prose:
C1 - 53.6🤯(was ~20 when GLM released)
C2 - 65.8 (35 per stream)
C3 - 78.2 (27.9 per stream)
C4 - 81.3 (21.7 per stream)
Prefill:
8k - 1538.7
16k - 1623.4
32k - 1641.1
64k - 1634.2
128k - 1581.7
There is no reason to not use GLM 5.3 Flash on dual DGX Sparks now as your daily driver. The speed is amazing, I'm very excited that this is only the start for TensorFold and I'd imagine we'll be seeing improvements constantly. This is a baseline.
This is the repo/recipe I used: github.com/jayleaton/glm53-t…
@jayleaton thanks for sharing!
It's insane the amount of posts confusing the quality or efficiency of Anthropic's models with their insane subsidized subs.
Anthropic's playbook this time was releasing decent models (again) and bump subs subsidizes dramatically to give people a sense of efficiency. Right at the moment that other labs are holding and trying to be more conservative.