unnat sharma

language & agents @deepsource

[unnat sharma @unnxt30]

@dawkrish @dunkedtoast Huh

[unnat sharma @unnxt30]

@dawkrish @dunkedtoast Youre balding too much if that counts 👩‍🦲😮‍💨

[unnat sharma @unnxt30]

@dawkrish Skill issue

[Internet Freedom Foundation (IFF) @internetfreedom]

Statement : The blocking of BitChat's code on GitHub is unconstitutional and authoritarian.

New Delhi, 24 July 2026

The Internet Freedom Foundation (IFF) condemns the order issued by the Indian Cyber Crime Coordination Centre (I4C), Ministry of Home Affairs, directing GitHub to remove the code repositories of BitChat.

The order, Notice No. 11072601011432, was issued at 11:16 pm on 23 July 2026 under Section 79(3)(b) of the Information Technology Act, 2000 read with Rule 3(1)(d) of the IT Rules, 2021. The order directs GitHub to disable access to three repositories, including the Android application and its release files, within three hours. It threatens the platform with loss of safe harbour and criminal prosecution. No copy was published by the Government of India. The public learnt of it from a post by @jack, whose team develops BitChat. Censorship in India now comes to light through disclosure by the censored.

Since 17 July 2026, the Ministry of Home Affairs has suspended mobile internet around Jantar Mantar as per public reports about five times, most recently within a 1.5 kilometre radius from 4 pm until midnight on 23 July. That radius takes in Janpath and parts of Connaught Place. Reports describe signal jammers at the protest site and people walking two kilometres before their phones work. Inside that zone a student separated from her group during a detention drive cannot send a message to say where she is. Thousands of students and young people have camped at Jantar Mantar since June, seeking accountability for examination irregularities. Permission for their march to Parliament was refused. Metro stations were shut and also internet connectivity has been blocked.

BitChat is an open source application built for exactly this situation. It passes messages from phone to phone over Bluetooth, without mobile networks or a central server. It is striking that the order does not identify a single unlawful message. It objects to what BitChat is. In its own words, the application is dangerous because it enables communication "even during network restrictions" and can "circumvent lawful restrictions" during "internet shutdowns". Hence, the government's objection is that citizens can speak to one another while it has switched the internet off.

The order is illegal on at least four grounds.

1. Section 79(3)(b) is not a blocking power. In Shreya Singhal v. Union of India (2015) 5 SCC 1, the Supreme Court read down the provision. Intermediaries may be required to act only on a court order, or a government notification confined to the grounds under Article 19(2) of the Constitution. Blocking is governed exclusively by Section 69A and the Blocking Rules, 2009, which require a hearing and reasons recorded in writing, subject to review. Directions issued under Section 79(3)(b), Rule 3(1)(d) and the Sahyog Portal evade these safeguards, and constitutional challenges to this parallel regime are pending before High Courts.

2. The reasons in the order are circular. The order asserts that the repositories contain "information which is prohibited under any law" without naming any such information, and rests on what the application is "capable of" enabling. Anticipated misuse of a communications tool is not a lawful basis to prohibit the tool. By this logic a telephone exchange could be sealed.

3. The order cites Section 43 of the IT Act, a civil compensation provision, alongside conspiracy and abetment offences under the Bharatiya Nyaya Sanhita, 2023, against a platform that hosts code.

4. A three hour deadline issued close to midnight forecloses legal assessment and recourse, and fails the proportionality standard in Anuradha Bhasin v. Union of India (2020) 3 SCC 637.

The order also fails on its own terms as deleting a repository does not delete the application from any phone that carries it, and the mesh keeps functioning without servers. What the takedown actually prevents is scrutiny of the underlying code.

IFF demands that the Government of India:

1. Withdraw Notice No. 11072601011432 dated 23 July 2026 issued to GitHub.
2. Publish every takedown direction issued under Section 79(3)(b), Rule 3(1)(d) and the Sahyog Portal, with the reasons recorded for each.
3. Restore full connectivity around Jantar Mantar, publish all suspension orders, and disclose the legal authority for the deployment of jammers.

We stand with the developers and the young protesters whose speech this order seeks to silence.

[unonumero56 @unonumero_56]

It's just not a good week for gear 5 😭

[Jai Pradeesh @dolftax]

https://t.co/hLrzp5Pe30

[Kylian Mbappé @KMbappe]

Nous reviendrons. 🇫🇷🙏🏽

[Google Research @GoogleResearch]

Introducing SensorFM, a large-scale Sensor Foundation Model that learns from 1 trillion-minutes of unlabeled wearable data drawn from five million consented participants.

SensorFM learns a single, reusable representation of sensed human physiology that transfers across cardiovascular, metabolic, sleep, and mental health, as well as lifestyle and demographic factors.

More →https://t.co/lbi1DG0zAW

[unnat sharma @unnxt30]

Totally agreed, Im slowly working my way through it, my favourite part is that they make you think, I did the BPE naively, then did a profiling only to find out my top candidate scanning was naive, and I get to experiment with heaps thanks to it, and see the gains. Loving it.

[Stanford NLP Group @stanfordnlp]

The “problem” with CS336 is not the ~22 hours of videos but the larger number of hours it takes to do the assignments.

But that is where most of the real learning occurs.

We’re reminded of @karpathy’s seminal tweet:

https://t.co/fvSeE2bDkE

2026 site: https://t.co/E1pzUSC6Tr

[Harrison Kinsley @Sentdex]

Dario's args:

"Opensource you can see the source, here you cannot see inside the model"
- yes you can that's literally the open weights part btw.
- I cannot see the weights inside Claude, but I can GLM 5.2
- Models like Nemotron3 Ultra go further, all the data, training scripts, and model is opensource.

"Alot of the benefits like many people working on it, being additive doesn't work in same way"
- yes it does. We have seen endless fine tunes of various open source models for real improvements.

"Ultimately you have to host it on the cloud"
- no you dont. Dario is seemingly totally unaware of smaller moes and even dense models like qwen 27B.

Not only does dario not take part in social media, I am beginning to think he's never tried open source models at all and has no idea wtf hes on about

[Derya Unutmaz, MD @DeryaTR_]

DeepSeek should ban Anthropic from implementing Dspark into their models! In fact, Anthropic should be banned from using any AI research from China and required to remove any non-US data they used to pre-train their models! Cry, Dario, cry!😅

[OpenAI @OpenAI]

We’ve designed and built our first AI chip: Jalapeño.

Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

Chips are foundational to the AI economy. Building our own expands our full-stack platform from products to models to infrastructure, and will help us scale intelligence, serve more people, and expand access to AI.

[Diego Rodriguez @asciidiego]

remember the guy who used to write tiny c files and train gpt-2 without a gpu?

he's shilling slack integrations now

[unnat sharma @unnxt30]

have been working my way through the @stanfordnlp CS336 assignments. Just finished the BPE implementation from scratch. Breath of fresh air in a world where agents do all the work for you.
Looking forward to implementing the transformer next.
Will be blogging my way through 😁

[Alisa Liu @alisawuffles]

I'm joining OpenAI next week!🥹 The job search turned out to be really challenging but also super rewarding, so I wrote a small blog to share what I learned along the way and hopefully make the process a little less mysterious for the next person. https://t.co/6FigSBdenD

[unnat sharma @unnxt30]

The great lock-in is here.
Looking forward to reading this hit work by .@chipro

[shirish @shiri_shh]

They are putting DATA CENTERS in the ocean now.

Panthalassa, a startup from Portland, just raised $140 MILLION.

what they do: build floating platforms that sit out at sea and run AI. no power grid needed. the ocean waves make all the electricity. the seawater keeps the chips cool.

How it works: big floating balls bob up and down with the waves. that motion makes power. the power runs the AI chips inside.

backed by PETER THIEL. company now worth almost $1 BILLION.

land is running out of room and power for AI. so the next move is simple. go to the sea.

[Deedy @deedydas]

Pretty neat that with one URL change, you can now replicate and iterate on AI papers without having to even provision your own GPUs

[Exa @ExaAILabs]

Introducing Exa Agent: frontier web research at less than half the cost of GPT 5.5 and Opus.

/agent orchestrates a mixture of cost-effective models to complete any web research task, from simple data enrichments to building gigantic lists.

[Maxime Rivest 🧙‍♂️🦙🐧 @MaximeRivest]

https://t.co/7CEdeQqEpK

[FIFA World Cup @FIFAWorldCup]

Stars shining on the world's biggest stage ✨

#FIFAWorldCup