harshad_geek's profile picture. Building AbleCredit http://ablecredit.com , Earlier, Founder of @Pleduhq . #ML, #AI #edtech. Previously Slackhq, Astro.  Substack: http://harshadss.substack.com

Harshad Saykhedkar

@harshad_geek

Building AbleCredit http://ablecredit.com , Earlier, Founder of @Pleduhq . #ML, #AI #edtech. Previously Slackhq, Astro. Substack: http://harshadss.substack.com

Закреплено

‘Nothing was yet his, nothing was original, and what he had appropriated through learning he thought stolen and himself a thief: a Goth looting Rome. He loved the loot, but despised the looter.’ Richard Rhodes on Oppenheimer, in making of the atomic bomb, pg 123


Dear @jk_rowling thank you for mesmerising generations of kids with your books. I was afraid my 7 year old might take long to finish them. To my surprise, she finished first one swiftly and even wrote a letter to you. Here's me with awkward act of tagging you with the letter 😃

harshad_geek's tweet image. Dear @jk_rowling thank you for mesmerising generations of kids with your books. I was afraid my 7 year old might take long to finish them. To my surprise, she finished first one swiftly and even wrote a letter to you. Here's me with awkward act of tagging you with the letter 😃

Harshad Saykhedkar сделал(а) репост

Two decades ago, Arduino put a microcontroller in every student’s hand, revolutionizing grassroots education. Today, as Arduino begins a new chapter in another domain, we’re opening the next chapter in education: putting FPGAs into every student’s hand. We’re delighted to…

Vicharak_In's tweet image. Two decades ago, Arduino put a microcontroller in every student’s hand, revolutionizing grassroots education.

Today, as Arduino begins a new chapter in another domain, we’re opening the next chapter in education: putting FPGAs into every student’s hand.

We’re delighted to…

Harshad Saykhedkar сделал(а) репост
trq212's tweet image.

Indian tpot , CS influencers , course sellers tell folks to focus on (imo) mid ideas 💡 Simple databases, trivial project in C or esoteric Linux syscalls. They are still living in 2010s. Young engineers following this advice are gonna miss the bus of AI hardware/software


Harshad Saykhedkar сделал(а) репост
MorlockP's tweet image. via @The_Petrichory

Now, wouldn’t it be fun if they could get the Italian air chief to join in the photo too ? Oh wait ..

German Air Force Commander Neumann and Japanese Air Force Commander Morita shake hands on the open tail ramp of a Luftwaffe Airbus A400M transport aircraft. Behind them: two Japanese F-15J Eagles and two German Eurofighters Don't know why, but it's a freaking awesome picture 💯

TallbarFIN's tweet image. German Air Force Commander Neumann and Japanese Air Force Commander Morita shake hands on the open tail ramp of a Luftwaffe Airbus A400M transport aircraft.

Behind them: two Japanese F-15J Eagles and two German Eurofighters

Don't know why, but it's a freaking awesome picture 💯


Codex be like When all you have is a Python hammer 🔨, everything is ..

umm is it normal for codex-cli to edit files using python code? it doesn't have an edit tool?

vikhyatk's tweet image. umm is it normal for codex-cli to edit files using python code? it doesn't have an edit tool?


On a relative scale, IMO @OpenAI GPT has higher IQ than Claude and Claude > DeepSeek. Claude appears smarter due to specialisation in coding tasks. But as a general co-pilot, GPT 5 is crazyyyyy good.


The interesting point in postmortem for me is they use Jax in the serving infrastructure. obvious in retrospect anthropic. But still great, outside google , a larger AI 🤖 lab use Jax/TPU stack. Secondly feels good that bug 🐛 in frontier labs code that I can understand 🙃

We've published a detailed postmortem on three infrastructure bugs that affected Claude between August and early September. In the post, we explain what happened, why it took time to fix, and what we’re changing:



Harshad Saykhedkar сделал(а) репост

In this context, GEPA works as a prompt optimizer, so the end result is a prompt (or multiple prompts for a multi-agent system, one for each component). However, one aspect that does not get highlighted enough is that GEPA is a text evolution engine: Given a target metric, GEPA…

LakshyAAAgrawal's tweet image. In this context, GEPA works as a prompt optimizer, so the end result is a prompt (or multiple prompts for a multi-agent system, one for each component).

However, one aspect that does not get highlighted enough is that GEPA is a text evolution engine: Given a target metric, GEPA…

आँधी की तरह उड़कर एक राह गुजरती है शरमाती हुई कोई क़दमों से उतरती है इन रेशमी राहों में, इक राह तो वो होगी तुम तक जो पहुँचती है, इस मोड़ से जाती है गुलज़ार 🤌


Harshad Saykhedkar сделал(а) репост

If you believe free speech is for you but not your political opponents, you're illiberal. If no contrary evidence could change your beliefs, you're a fundamentalist. If you believe the state should punish those with contrary views, you're a totalitarian. If you believe…


Harshad Saykhedkar сделал(а) репост

The Fourier transform is everywhere you look. It has helped scientists to study tides, detect gravitational waves, and develop radar and MRI technology. Here’s how it works. quantamagazine.org/what-is-the-fo…


Harshad Saykhedkar сделал(а) репост

How to be a terrible neighbour

TerribleMaps's tweet image. How to be a terrible neighbour

Nvidia understands that building AI themselves has smaller TAM. It’s like they want to be electricity, not the machine that runs on the electricity

NVIDIA has all the funds in the world, and literally makes the hardware required to train LLMs. They have co-built datacenters with xAI so they know how to do that. They also access model source openly as they work with LLM companies. Why do they not make their own?



This is one of the most comprehensive and useful dataset for Indic languages and voice models . Thank you for your work , team @ai4bharat

The scale of IndicVoices is powered by people. Nearly 1,900 coordinators, mobilizers, and experts worked alongside 16,000+ participants to capture speech in 22 Indian languages. Their role wasn’t limited to logistics. They built trust, helped participants navigate apps, and…

ai4bharat's tweet image. The scale of IndicVoices is powered by people. Nearly 1,900 coordinators, mobilizers, and experts worked alongside 16,000+ participants to capture speech in 22 Indian languages.
Their role wasn’t limited to logistics. They built trust, helped participants navigate apps, and…


🤣🤣

Someone in my building has learnt how to play a shankh. He plays it so often I feel like I'm living in Kurukshetra



Spot on analysis.. China focused on basic education first whereas India focused on select few higher education institutions. Only way to solve the problem is to build an ai assistant for every child to learn STEM better .

Bottom-up vs. top-down education. China took a bottom up approach, focusing on building up its base. Till the 1950s, it focused solely on mass primary education, trying to get everyone to basic literacy levels first.



Harshad Saykhedkar сделал(а) репост

"Before it is science and career, before it is livelihood, before even it is family or love , freedom is sound sleep and safety to notice the play of the morning sun." Richard Rhodes on the Jew exodus from hitlers Germany ..


Loading...

Something went wrong.


Something went wrong.