Ayush Saha retweeted
Ramanujan wasn’t skipping proofs. He was hiding his chain of thought to prevent distillation.
Hardy: “Show your work.”
Ramanujan: “Sorry, reasoning tokens aren’t exposed via the API.”
81
711
9,882
994,294
Ayush Saha retweeted
All hijackers are stupid but the extra stupid ones try to hijack a plane full of 180 Israelis all of whom have gone through some of the best military training the world has to offer & an Indian pilot who thinks honor and karma are bigger than his own life. Get well soon Capt'n 🫡
244
5,534
48,984
899,768
Afaik @sarahookr basically called this in 'On the Slow Death of Scaling', brute force compute is giving less and less back and the next wave is models that adapt from real usage
1. Agree unique data is the moat but the real question is what kind of product gets you that data
2. Shipping a bunch of FAFO fad products gets you a lot of noisy data, or,
Going deep on one workflow, let's just say an agentic IDE, gets you really clean signal
3. But I don't think one works without the other for eg clawdbot now openclaw had to walk so instinct and muse could run
the FAFO phase tells you what people want and going deep is where the unique data actually comes from
last 6-8 months have been a good example of this.
so a good product leader should do both? or no?
Prediction: from now on, great product leaders will get $100M comp package, not researchers.
1. 2022-2026 is all about killer models. Right now models still matter, but only for frontier tasks like science, math, trading etc. etc. You either kill cancer or you don't matter.
2. Instead, great products like muse, grokbot make AI truly accessible for people. The proliferation of AI matters way more now.
3. Open weight models make it cheaper and better, and the margin for most models will go down significantly, leading to less profits.
4. Training techniques are copyable, but unique data is not. All things equal(compute, training techniques), unique data is the new oil. Even if your training techniques are second-tier, with unique first-tier data you win.
5. Great products attract sticky users who provide unique data that lead to even better models.
Therefore, product is the one that matters. We are finally entering the gold era of product visionaries.
2
226
This is too good to be true, might as well want to check it yourself
Kudos @roman6301 on the launch!
evening lads!
i've just launched gap133: a cross-venue terminal for prediction markets.
"one event. two venues. one number matters: the gap."
live right now: gap133.xyz
follow @gap133xyz for the gap of the day.
alerts land first on t.me/gap133_alert
keep reading though...
1
4
217
guess it’s RLCD now
{"status": 200, "confidence": 0.98}
After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?
I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev
• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions
AFAICT the shortest path to AI-based economic revolution
150
Ayush Saha retweeted
“Dark Talent” that’s what @balajis calls it
we are serving the infra you deserve 🇮🇳
6
7
50
3,131
HL done. Another leg up HH or LH?
Not really observations that others before me haven’t made, but:
- Despite the seemingly magical nature of LLMs, reflection over a >3 month timescale suggests my total productivity hasn’t increased by over 100%, or perhaps even by over 50%, and a lot of time is actually wasted because LLMs enable me to spend time on gratifying but low-productivity tasks that in the future turn out to not be useful
- Also, capabilities are incredibly spiky and highly correlated with the degree of investment poured into them, which my earlier tweet about math benchmarks implicitly points out
- From the above, it seems that the nature of LLM intelligence is wildly dissimilar to that of human intelligence and we won’t trivially get to something superior to human intelligence in all important respects just by scaling up existing approaches with various tweaks; even if AGI Is eventually achievable, this implies a significantly longer timeline
- Benchmark progress is almost definitionally guaranteed to happen because the process of constructing a benchmark is a direct precursor to the process of constructing a training dataset used for hill climbing that benchmark, but the scope of what can be captured in a benchmark is (at least for now) grossly lacking in terms of its relevance to real-world work, with maybe several limited exceptions
- Progress seems highly gated by data but the nature of model training means that each “next dataset” is significantly harder to assemble than what preceded it; some wins are possible through synthetic methods but those feel more like “patching up gaps” than “pushing the frontier forward”
At a higher level, I guess I’d say there’s a sort of refusal to think carefully about what models are or are not useful for in a rigorous way which I find personally quite annoying, and instead a reliance on some nebulous notion of being “AGI pilled” as a replacement for serious thought. I think people are very quick to anthropomorphize LLM intelligence because humans communicate through words and we infer the intelligence of human counterparties through comprehension of their language, but this leads them to wrong conclusions; for example if we observe that a new model proved some incredible mathematical theorem, some will say, “well, don’t we have AGI now, huh?” But to me, it’s actually more like, “well, given how hard it would have been for a human to do these mathematics, and given the limited economic effect of LLMs upon the world so far, isn’t it actually a negative datapoint vis-a-vis the generality of LLM intelligence?”
173
I still find myself going back to Opus 4.6 with SOPs, they did a good job on fine-tuning with RLHF
2
421
Ayush Saha retweeted
If you get paid in USDT/USDC as an Indian 🇮🇳
You are probably not settling in INR the right way
Don’t get hit with 30% tax or fall for P2P routes
We made the full compliance stack just for you
17
6
132
6,828
