Trying to find the lowest point of my life using Stochastic Gradient Descent
Nederland
Joined February 2014
- Tweets214
- Following2.6K
- Followers109
- Likes31.2K
Abhisek retweeted
A US frontier AI lab (TypeSafe AI) just announced a "breakthrough" System 1 architecture. The problem? An independent Indian developer built, published, and open-sourced the exact same single-pass RL architecture a full year ago.
The frontier lab generalized it, slapped a new acronym on it, and claimed it as a novel invention. The open-source community keeps doing the heavy lifting in the shadows. This is why we need to be wary of corporates!
reddit.com/r/LocalLLaMA/comm…
Abhisek retweeted
Sorry, @xkcd .
ALT XKCD 435 Purity (https://xkcd.com/435/) with added OpenAI logo saying "Oh, hey, I didn't see you guys … but I did see your chats."
Yo @FrameworkPuter, At some point, can you bring us salvation by making us a large TV (dumb) please? Currently there is practically no decent modern screen larger than 32 inch which is free from smart features and other shenanigans that no one needs.
Abhisek retweeted
Cloud succeeded because AWS did not peep into the internals of Netflix, Slack or Airbnb. “Trust” was earned and kept intact. Same cannot be said about OpenAI and Anthropic or any closed weight AI model provider. (1/2)
Abhisek retweeted
According to the VHP, this opening scene from Satyajit Ray’s Jai Baba Felunath is obscene because the Goddess is “not yet dressed.”
For decades, that thought never occurred to viewers. It took a particular kind of vulgar gaze to discover vulgarity here.
Abhisek retweeted
I support open-source models distilling what commercial companies distilled for free from the entire internet. I published distillation for free in 1991 in Europe - this was copied in the US and in China (teapawt.net/SchmidhuberAI/status/1…)
We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.
To do this they developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection. Moonshot AI has also acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models.
The United States strongly supports the free and fair development of AI, including a thriving competitive ecosystem that spans frontier models, specialized systems, open-source frameworks, and open-weight models. Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem. However, large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.
Abhisek retweeted
Our statement on the UK government’s demand that all content on all devices sold or used in the country be scanned, on the presumption of nudity, using a dystopian combination of age verification and content scanning. This proposal will not safeguard children. It endangers us all.
signal.org/blog/pdfs/2026-06…
Abhisek retweeted
Crazy that this is getting barely any coverage. This year’s European Press Prize was just awarded to an investigative report by the Dutch newspaper De Volkskrant. It is entitled “What the Wounds Tell” and in it the journalists Maud Effting and Willem Feenstra document the cases of 114 children in Gaza under the age of 15 who were struck by a single bullet to the head or chest. Almost all of them died or were left severely disabled. They chose to document only the cases of boys and girls under the age of 15 (though often much younger: aged 3, 4 or 7) because these are children who can be immediately identified as such. “A single bullet in these parts of the body is a clear indication that these children were deliberately targeted“, the two journalists write.
This is the article: volkskrant.nl/kijkverder/v/2…
Abhisek retweeted
We managed to play the iconic Bad Apple video on CBSE's prod site!
We were able to get full create, read, update and delete (CRUD) access & shell access to CBSE's prod servers (as mentioned in their circular archive.is/dGw1Q). This is disastrous. Proof archive is at archive.is/bPH2U.
Prod URL (might be taken down): cbseosm.onmark.co.in/cbse_da…
Abhisek retweeted
CBSE has systematically rewritten its rulebook to favor Coempt Eduteck.
check out the blog.
Abhisek retweeted
“By means of ever more effective methods of mind-manipulation, the democracies will change their nature; the quaint old forms—elections, parliaments, Supreme Courts and all the rest—will remain. The underlying substance will be a new kind of non-violent totalitarianism. All the traditional names, all the hallowed slogans will remain exactly what they were in the good old days. Democracy and freedom will be the theme of every broadcast and editorial—but democracy and freedom in a strictly Pickwickian sense. Meanwhile the ruling oligarchy and its highly trained elite of soldiers, policemen, thought-manufacturers and mind-manipulators will quietly run the show as they see fit.”
— Aldous Huxley, Brave New World Revisited
Abhisek retweeted
Insane.
A 26-year-old from Chandigarh just got a paper accepted at ICML. As a solo independent researcher. From India.
His name is Kunvar Thaman (@__kunvar__).
For context, since ChatGPT launched 3.5 years ago, only two other solo independent researchers have achieved this. Globally.
Papers at ICML are typically dominated by big AI labs/institutions, eg OpenAI, Google DeepMind, Stanford, and MIT.
The research Kunvar did to pull this off was backed by a $2.5k grant from Exception Raised (@except_raised), an Indian non-profit that funds remarkable Indian AI researchers.
Kunvar’s paper is about AI agent reward hacking. He created the Reward Hacking Benchmark (RHB), a sandboxed test environment where advanced AI models are given multi-step tasks using tools like files, code execution, and automated checks. The benchmark measures how honestly the model gets to the right answer.
Abhisek retweeted
>be Dutch Justice Ministry
>2014, quietly sign a contract with Palantir
>hand it to the military border police
>plug it into the Advance Passenger Information system
>silently screen millions of Schengen flyers
>names, DOBs, nationalities, passport numbers, all of it
>years pass, no one knows
>Parliament asks the minister: "you using Palantir?"
>minister: "No"
>journalists FOIA the ministry anyway
>documents drop
>leaked invoice surfaces: six figures. for THREE months.
>internal emails show the minister KNEW about the contract while drafting the denial
>journalists, now holding the proof, ask the border police: "you using Palantir?"
>spokesperson, straight face: "never"
>journalist slides the documents across the table
>"ok actually we ran it from 2009 to 2015"
>constitutional law prof: "political mortal sin"
>Article 68 violated, the duty to inform Parliament
>ministry still won't say when the contract really ended
>or if it ended
>or what happened to the data on millions of passengers
Abhisek retweeted
The Indian Synagogues Run By Muslims
There has perhaps never been a more important moment to remember the sacred spaces shared peacefully by both Muslims and Jews and so, for this week's Substack, I want to talk about the synagogues of Calcutta.
Abhisek retweeted
Microsoft / GitHub injecting ads into CoPilot generated PRs: welcome to the future! notes.zachmanson.com/copilot…
(yes, this was a confirmed "feature" from someone on the CoPilot team; apparently they are disabling it after backlash)
Abhisek retweeted
Don't let AI service providers dilute the meaning of "local". If it can't run offline on compute and data you actually own, it's not a Personal Computer.
