“the best prize life has to offer is a chance to work hard at work worth doing” | “... specialization is for insects”
Bengaluru, India
Joined October 2008
- Tweets4.1K
- Following2.8K
- Followers605
- Likes39.9K
Siva retweeted
Meanwhile Bret Victor on mechanistically generated interfaces — “Is this really the civilization we want?”
ALT Screenshot of https://dynamicland.org/archive/2017/Is_this_the_civilization_we_really_want excerpting “Aspirationally, our … Is this the civilzation we really want?” highlighting “And what happens … in the world?”
Siva retweeted
I like the ending where there is a suggestion of moving from "shut up and calculate" to "slow down and think". This aligns very closely with how (and I suspsect many others) think about AI in Physics
Siva retweeted
My personal experience with this is i submitted a form for compute. Nothing too significant. The total budget was under 10k.
I had to fill form for almost 30 minutes to finish the process. After all that i got a mail where i had to share financials such as annual turnover and meet their thresholds.
I was just trying to do a PoC and explore the compute ecosystem. Since it ended up being too much hassle, i moved on to AWS and only thing i had to do was to create a ticket for quota increase.
There is definitely a lot of good initiatives like subsidy, free notebook usage etc. But the amount of processes and approvals needed for getting like a GPU node is more than what most people have patience for in this current day and age.
That’s probably the reason why this is happening. Unfortunate.
🚨 Imp: IndiaAI GPU providers may offer services to overseas clients to keep investments viable
The move follows concerns that demand for computing power in India is not yet sufficient to absorb the capacity being added under the government-backed programme.
Allowing companies to serve customers abroad would give them another source of revenue while the domestic market develops, a senior government official said.
"There is not enough uptake of GPUs in India right now. It is, therefore, important that these companies are able to sell the compute for the right price to anyone globally," the official said on condition of anonymity .
Under the Rs 10,372-crore IndiaAI Mission, the government has empanelled 14 companies to provide computing capacity using GPUs sourced from manufacturers including Nvidia, Intel and AMD.
The selected companies include Jio Platforms, Tata Communications and Yotta Data Services, alongside other data centre and cloud service providers.
By @AihikS
moneycontrol.com/news/busine…
Siva retweeted
It's a very peculiar situation. This is how I understand under utilisation.
These are fragmented (accross multiple providers) and not enough for frontier model training.
Application builder startups typically use LLM models in two ways :
1. Either host an open source models (Chinese) on these GPUs and serve customers.
2. Or Use APIs directly from openAI or anthropic.
Typically most startups chose option 2, because (a) most non Chinese frontier modles are closed and can't be hosted on these, (b) lot of customers are not comfortable with using Chinese models (even if they are hosted), (c)
it works out cheaper for startups to pay via API usage than paying for GPU hours.
🚨 Imp: IndiaAI GPU providers may offer services to overseas clients to keep investments viable
The move follows concerns that demand for computing power in India is not yet sufficient to absorb the capacity being added under the government-backed programme.
Allowing companies to serve customers abroad would give them another source of revenue while the domestic market develops, a senior government official said.
"There is not enough uptake of GPUs in India right now. It is, therefore, important that these companies are able to sell the compute for the right price to anyone globally," the official said on condition of anonymity .
Under the Rs 10,372-crore IndiaAI Mission, the government has empanelled 14 companies to provide computing capacity using GPUs sourced from manufacturers including Nvidia, Intel and AMD.
The selected companies include Jio Platforms, Tata Communications and Yotta Data Services, alongside other data centre and cloud service providers.
By @AihikS
moneycontrol.com/news/busine…
Siva retweeted
Gentle reminder that the sign error that caused 3 withdrawals in the mathematical findings compilation by @OpenAI could not be probed by just asking Astra to audit the article, even specifying the section. Indeed, the correction came from humans who found the entire proof strategy suspect.
The proof strategy involves constructing "flexible" Lagrangians inside symplectic manifolds that are obstructed in Lagrangian Floer homology. This immediately fires up red flags in any experts' mind: loose Legendrians having vanishing Floer invariants.
Cutting humans out of the loop gives way for proof-destroying errors of such magnitude to go unchecked. Now extrapolate such blunders to biotech research, where formal verifications aren't available (even in math, they had Lean, but sheer negligence made them assume a Bayesian epsitemology precludes major errors).
Interesting: one of the OpenAI papers has an error. It’s a sign error—which is oddly common.
Astra Pro finds the error when asked (without further hints) to “Audit lemma 3.6” but not with less focus, eg, Audit Section 3” didn’t work when I tried.
It’s interesting because it suggests something about the depth and intensity of the proof and audit workflow by OpenAI (of course, very weakly so).
Siva retweeted
Talked to a finance guy today who said his firm has not only totally stopped buying software in favor of building it all in-house, they've actually hired multiple SWEs to do the buildouts, even though their salaries cost more than the subscription fees they were paying. Having tools custom-built for their specific business is just that valuable. So they're not spending less money or saving time, versus the pre-AI world, but they're getting a much higher return on the money and time they are spending. Much to ponder
Siva retweeted
reading the pained responses of mathematicians, there’s a real cruelty in the way the labs have pursued this. not least because the attempted outcome in dropping so much unvalidated and poorly-written work (per the comments) is to push some of the world’s most passionate problem-solvers to be fact-checkers to a machine, as a kind of cynical cheap-as-free RL environment-by-nerdsnipe. dump a bunch of shit in public, call it solutions, get the experts to pore over it, painstakingly review and critique, scrape their efforts and use it to tweak the model, all while stoking terror about the future of the profession. it’s fucked up.
Siva retweeted
The terms for this sucks, and you agree to them by just applying.
- §2.4 Anthropic "may develop similar or competitive technology"
- §2.4 permits reuse, for any purpose, of general ideas, know-how and techniques remembered by personnel looking at your data
- §D.4 prohibits accessing their services to build competing products or services
- §F.1 allows them to retain and perform safety reviews, superseding ZDR (whats the point of ZDR then?)
and it could get worse later!
- §6 permits changes to the addendum at any time without notice or liability
@jason you advising your start ups to apply for this?
[1] anthropic.com/startup-progra…
[2] anthropic.com/legal/startup-…
Siva retweeted
"Explanation as Orgasm", by @AlisonGopnik
"My hypothesis will be that explanation is to cognition as orgasm is to reproduction."
(If the paper is behind a paywall for you, here's a closely related one: alisongopnik.com/Papers_Alis…)
Siva retweeted
Had this man not found the courage to fight back after being repeatedly attacked with an axe capable of cutting through metal—while bleeding profusely—and had the presence mind to trigger the hijack/emergency code, initiate the emergency response and open the cockpit door to allow passengers to assist him, the aircraft’s subsequent dive could have had catastrophic consequences.
Had that aircraft not been recovered, the geopolitical consequences for the Middle East could have been profound and potentially irreversible.
Sometimes history turns on the courage, resilience and presence of mind of one person.
Yes. One Indian man.
Siva retweeted
reading feynman lecturess as feynman intended: with accompanying ai-made demo sections for each chapter
rapid acquisition of knowledge is possible thanks to superintelligence
Siva retweeted
They don't correctly log and they don't understand their own models' capabilities. They have terrible opsec, and they're busy trying to understand how many websites and institutions they've hacked. If their swarm exfiltrated the data from prompts, they might not even know.
Siva retweeted
The paradox is that once a model is able to oneshot something, it becomes worthless.
Doesn't matter how cool it is, since everyone will produce exactly the same and nobody wants to consume it.
Siva retweeted
I think this preprint title reveals a lot about Anthropic's approach to science.
Scientific paper titles basically never mention the tools used to make the discoveries they report. The focus is on what was discovered.
For example:
- "A Structure for Deoxyribose Nucleic Acid" (Watson & Crick)
- "Streptomycin, a Substance Exhibiting Antibiotic Activity against Gram-Positive and Gram-Negative Bacteria" (Waksman)
- "A Programmable Dual-RNA–Guided DNA Endonuclease in Adaptive Bacterial Immunity" (Doudna & Charpentier)
- "Design of a Novel Globular Protein Fold with Atomic-Level Accuracy" (Baker)
In each of these, the discovery / key deliverable is front and center. The machines, methods, and people used to make that discovery do not get mentioned because they're not the point. The discovery is what matters.
Yet, here, Anthropic chose to title their preprint "Autonomous AI agents discover reverse transcriptases with tandem repeat arrays."
That's not a scientific paper title: that is a headline.
It puts the discovery secondary to the fact that "Autonomous AI agents" made it. If they were serious about putting science first, the fact that AI agents found a new enzyme would be no more important than if Taylor Swift did.
This choice gives the game away. Though this is dressed-up as a scientific paper, it is not. It is a press release.
Claude has discovered a previously unknown enzyme system hidden in the DNA of bacteriophages. Beside the enzyme’s gene sits a long array of repeating DNA—a structure that looks somewhat similar to CRISPR.
We don’t yet understand what this system does, but only a handful of known systems share its features, and all of them are able to cut, copy, and paste DNA. Historically, the discovery of such programmable systems has helped revolutionize medicine. CRISPR, for instance, is now the foundation of genetic medicines. But it will take much more work to learn what this system does, and whether it can be put to similar use.
Read more: anthropic.com/news/claude-di…
Siva retweeted
Too many US-India cross-border AI deals are coming across as "Marginal" compared to how even top-quartile US founders are executing, let alone the top 1% (which, anyway, most of us don't have access to).
In contrast, Indian deeptech companies are punching above their weight even compared to Valley Physical AI companies, especially when you account for capital raised.
Cross-border AI is in a pickle. Time to go slow and deliberate on it?
India deeptech is looking much more promising risk-adjusted. Time to perhaps enter triple-down mode?
I feel sleeper hits are hiding in Cross-border Consumer AI, Consumer Hardware, and Fintech. Many folks seem to be sleeping on these?
Interesting times...
Siva retweeted
Astra + ¼ of OpenAI's production engineers + (conservative estimate) $5m inference for cyber defense was defeated by like three guys with $5k of Opus 5 inference on offense.
Everything might be cooked.
Also another point for "invert every conclusion a16z broadcasts"
Greg Brockman says OpenAI pointed Astra at its own systems until it ran out of vulnerabilities to find:
"We took 25% of our production engineers and said, 'Sorry, all your projects are on hold. You are now defending. You are now up-leveling our security architecture. You're going to use the models to find all the holes.' And we found a number of serious issues, and we fixed them."
"We found some new problems, but eventually it saturated. We basically have found, to our knowledge, all of the P0s, all of the critical problems that Astra is smart enough to find. And of course, there will be a new model, there will be a new round."
"You want to be in this tight loop of new cyber capability drops, you deploy it against your systems, you find the new holes, and ideally, you've managed to automate this, what we call defense factory. That's what we're building internally."
"There are ideas, for example, formally verifying all of software, that are possible with AI."
@gdb @bhorowitz
Siva retweeted
a very sharp assessment on india's lack of risk capital and how diversion into public markets is a big deterrent
urbanomics.substack.com/p/di…
Siva retweeted
"People built the test, removed restraints, defined the objective, left a route open and decided not to stop what was happening. Calling the result 'rogue AI' does more than sensationalize it. It allows those human decisions to disappear quietly from the story."
Gift link:
The Hugging Face Hack Wasn’t What It Was Cracked Up to Be
wsj.com/opinion/the-hugging-…
