Rakshit

Tweets are notes to self | Interested in tech startups, culture & @Arsenal | Opinions are personal

[Rakshit @rakshran]

@volklub @Trendulkar What about xp95?

[Nitin Sinha @NsNitinsinha]

Europeans travel more because:

• Their countries are tiny and connected like Indian states
• They get stronger passports, cheaper flights, student hostels, work permits and social security
• Average salaries are much higher relative to travel costs

An Indian middle-class kid often carries family responsibilities much earlier. Supporting parents, saving for home, education loans etc.

And calling temple visits or family trips “not travel” is peak elitism. If a family saves for 2 years to visit Kedarnath, Goa or Kerala, that experience matters just as much as a backpacker smoking weed in Bali “finding himself”.

Also funny how people romanticize European gap years but ignore:

• Indians already migrate, struggle and adapt more than most populations globally
• Millions leave hometowns at 18 itself for coaching, jobs and survival
• Indians work in every continent on earth

Travel doesn’t automatically make someone deep, cultured or intelligent.
Some people return from 12 countries with only fake accents.

And honestly, there’s nothing wrong with wanting stability either. For many Indians, that stability is the result of one generation’s sacrifice.

Travel should expand your mind, not your superiority complex.

[Arsenal @Arsenal]

The Arsenal. Your Premier League champions.

[Harshil Mathur @harshilmathur]

It’s the best time to build all the ideas you’ve ever had - the app you wished existed, the design you thought was obvious, the library you always wanted.

And finally realise why nobody built them :)

[Rakshit @rakshran]

https://t.co/8uC3UsFNGt

[Rakshit @rakshran]

I am redesigning my website. My last update goes back more than 3 years. This was still the pre-AI era, where I had to read up a lot of documentation, scour through GitHub to find inspiration (and reusable code) and spend hours on Stack Overflow, debugging plumbing and syntax issues.

But now we have Claude Code, which does most of these tasks for me in the terminal.

I am on the $20 Pro plan. And the limits on it are not reasonable enough to work on any serious long-form project.

I also pay $20 for Gemini and couple that with other SaaS subscriptions, my monthly software bills easily touch $80-$100.

When I asked my wife for permission to spend $200 on a Claude Max plan, I received a vehement veto.

So I went down a rabbit hole to find a workaround.

Fundamentally, Claude Code is an agentic coding tool with two distinct components (they’re easy to conflate):

1. The harness. The local program that runs in your terminal. It reads files, edits them, executes shell commands, manages git, asks for permissions, holds conversation state, injects memory files like CLAUDE dot md, and runs the loop: think → call tool → observe result → think again.

2. The brain. The LLM the harness calls out to whenever it needs to reason. The brain produces no side effects on its own. It reads a prompt and emits text, including tool calls. The harness is what actually executes those tool calls on your machine.

The harness is the durable surface. The brain is, in principle, a commodity slot.

Conceptually, swapping one model for another within the same harness should work.

Claude, in fact, offers users a way to slot other models in by setting environment variables.

I write about how to use free open weight models via OpenRouter and never run out of limits on Claude Code.

[Amend and Pretend @amendandpretend]

@ecommerceshares Data centers lower your ESG score

[Rakshit @rakshran]

@jack This is a run

[Rakshit @rakshran]

@Rajiv1841 Grew up watching cricket in the 2000s and interest faded sharply post the retirement of the legends of that era. Have been tracking RR’s matches and make it a point to watch Suryavanshi. He is generational! Hopefully stays the course and becomes a legend himself.

[Rakshit @rakshran]

https://t.co/WanLSn65nM

[Rakshit @rakshran]

I believe that we are in an interim, transient state in the development of AI as a technology.

AI is most likely to dissolve the boundary between "using an application" and "getting a thing done". Models should be able to take inputs implicitly and perform tasks independently.

Within this paradigm, Google owns the surfaces where intent arrives (Search, Chrome, Android, Gmail, Maps, YouTube) and the surfaces where action happens (calendar, inbox, documents, drive), resolved to a single account.

In software, where distribution and transaction costs are zero, @benthompson's Aggregation Theory is the seminal thesis on who wins: "The most important factor determining success is the user experience: the best distributors/aggregators/market-makers win by providing the best experience, which earns them the most consumers/users, which attracts the most suppliers, which enhances the user experience in a virtuous cycle." Here's the final post on distribution in the Why Google might win series.

[Rakshit @rakshran]

I am the husben

[Tamanna @tamvaani]

>be me
>wants to buy white pants
>husben suggests me to not buy white pants because it’ll get dirty
>buys white pants to prove him wrong
>spills coffee on white pants the first day i wear them to office

[Tamanna @tamvaani]

only way to visit natural beauty is to have ambition

[Rakshit @rakshran]

https://t.co/cCCakMWlRg

[Rakshit @rakshran]

Remember Google Bard?

OpenAI shipped ChatGPT in November 2022, hit 100 million users in two months, and Google declared code red. Larry Page and Sergey Brin came back for emergency meetings.

Google's response was Bard. In its very first public demo, Bard claimed that the James Webb Space Telescope took the first pictures of an exoplanet. It hadn't. That honour belonged to the European Southern Observatory's Very Large Telescope in 2004.

You could have verified this by Googling it.

$100 billion in market cap evaporated in a single day.

And then, Sundar Pichai, acting as a wartime CEO, did something decisive. He merged Google Brain and Google DeepMind: two labs that had been kept apart for a decade under DeepMind's acquisition terms.

Three years later, Gemini is among the leading models across three varied benchmarks that matter:

1. LMArena (user preference) - real users vote on two anonymous models across whatever tasks they actually care about

2. On hard problems with canonical right answers - Olympiads, competitive programming and Humanity's Last Exam

3. General intelligence (ARC-AGI-2) - a benchmark specifically designed to be unsolvable by scaling, brute force and chain of thought reasoning alone

The model layer is table stakes. Other frontier labs are consistently pushing the boundaries of competition.

Talent is fungible, techniques are being replicated, and access to capital is also largely democratised. Hence, the only thing that matters at this layer is sustained competitiveness. And Google has done exactly that.

[Wise @trikcode]

the generation that refused to accept cookies.

is now giving AI access to their desktops, files, and bank accounts.

[Rakshit @rakshran]

https://t.co/wQRD2kBzMH

[Rakshit @rakshran]

When Elon Musk agrees with your thesis, it is motivation enough to release a new essay.

The last two posts in the "Why Google might win?" series were about things Google deliberately built: custom chips that cut compute costs and private infrastructure that no competitor has fully replicated.

Part 3 is about something that accumulated. Data.

Most publicly available AI training data captures a fragment of human cognition but not behaviour.

Google can trace the full arc of a human decision through outcomes and map it accurately to individual users.

Search captures intent. Chrome and Analytics capture commercial behaviour across 85% of the web. Gmail captures the downstream evidence: receipts, bookings, confirmations. Maps and Waymo record where people end up physically. Fitbit captures biological response.

And that is before you get to multimodality and compounding through billions of YouTube videos where audio, visual, and text are locked together in the same file. Or millions of miles of Street View imagery. Or Waymo's real-world driving data.

I write about why this matters structurally as AI shifts from language models toward agents that act in the world.

[Elon Musk @elonmusk]

@bindureddy Google will win the AI race in the West, China on Earth and SpaceX in space

[Rakshit @rakshran]

@amix3k You can just buy woice ~ https://t.co/pgxmMBSrjU

Works amazingly well - without hiccups

[Thomas Frank @TomFrankly]

Currently 892 hours into automating a 30-second task I do 4 times a year

It's gonna be so worth it once I get everything working