Philosophy of Mind • AI enthusiast • GTM
Joined August 2026
- Tweets876
- Following712
- Followers170
- Likes2.6K
Elliot retweeted
peter thiel is the only one who understands that the whole point of getting rich is getting people to take your essays seriously
Elliot retweeted
Absolutely incredible: MIT researchers say GPT-6 Astra improved a small Qwen model’s math score from 7.9% to 48.1% by directly editing just 0.016% of its weights, without gradient-based training.
The result points toward AI systems that can diagnose and improve other models directly, a potential step toward self-improving AI.
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
Breakthrough of the year..?
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
Elliot retweeted
if you’re an employee of any corporation, in the medium & long term you are effectively generating training data for an ai that will likely replace you.
even if the company fails, your work will still create value as training data.
i know a lot of ppl find this bleak & possibly unpleasant but it’s the truth. & it’s kinda cool to think about. humans were never meant to do jobs that could be automated (we don’t use plow’s anymore). eventually we should all be creating our ai replacements in some sense so that we could live the way we were meant to.
I've made a completely local / offline dictation tool. I'm thinking of selling it for like 25 quid maybe. In my experience, it's also much faster than others on the market like Whisprflow.
Probably going to launch in the next couple weeks. Feel free to DM me if you want free access in exchange for feedback.
Elliot retweeted
This is a wonderful idea! Anthropic, if you IPO in the EU, I will personally throw in 15 carbon credits and a €250 entrepreneurship grant
Elliot retweeted
Quick super important reminder:
All the crap you see happening around the world - protests in France, other European countries, now Indonesia, other places, it's all noise.
None of this matters in the grand scheme of things. Even short term.
The only thing that really matters is AI progress. The fact that we know there is no moat, that more compute = bigger breakthroughs in mathematics and physics, and soon in biology, medicine, materials science, engineering, and anti aging progress.
The trend is crystal clear. This isn't some science fiction prediction anymore. We're being presented with new data and evidence almost every hour.
Acceleration is real.
Doom and gloom grifter media are doing their job, manipulating you into believing everything is getting worse.
In reality, these are the best times in human history. Nothing else even comes close.
100s of years of scientific and technological progress in a single year may actually become possible (thanks to AI)
The concept of a country of geniuses in a data center is already becoming a reality. Soon, we'll have 10x more compute, more AI geniuses, more discoveries, and even faster progress.
And this is just the beginning.
***Ignore the noise. Focus on the acceleration.****
Elliot retweeted
the fact that you exist as an entirely unrepeatable perspective on reality should fill you with an enormous sense of possibility. nobody can create from the exact place you create from. nobody can see through your eyes. take that seriously. make something of it.
Dude I love this app.
But whenever they crank up the “doom” algorithm, I feel compelled to uninstall.
I think people are misreading his expression here.
Look at his eyes and chin. His reaction is not smugness, it’s vulnerability, guilt, and sadness.
“You just spooked everyone by saying stockpile (for a crisis)”
Andy Burnham:- <Smug look & laughs knowingly>
“It’s a challenging World”
The UK PM treats the Public with absolute contempt after telling everyone they need to “prepare for a crisis” - now we can’t even keep a straight face‼️
The Big Bamboozle
Elliot retweeted
this may be my last tweet before shit goes parabolic. i’m not an ai researcher whistleblower. i work for none of the biggest labs in the world.
there is currently a major bloom happening across humanity. they are hiding this from the public.
we are on the verge of an era of abundance, superintelligence, cure for cancer & so much more.
prepare for the best.
Elliot retweeted
It seems relevant that AI memory is not even close to solved.
I mean real memory, like a human or animal has.
A long stack of .md files is not it.
