
ai @cloudflare, prev @localhosthq @lossfunk
bengaluru, india
Joined July 2020
- Tweets1.1K
- Following408
- Followers424
- Likes6.2K
kavin retweeted
Soundcheck: BaseCamp Quiz Night
📍THE HUMMING TREE
@MotwaniSuhas @sonumbonweb @siddharth17 @ashimhta @xan_ps @bolna_dev @TheProductfolks
kavin retweeted
first night at our delhi campus!
a little pocket of silicon valley in 🇮🇳
20 founders, 50 days to build.
across robotics, film, ai, space, healthtech, etc.
20 builders in india's biggest hackerhouse.
50 days.
Each on their own journey.
@localhostIND for the books.
kavin retweeted
absolute banger article explaining how nginx works the hood and why it's event-driven architecture outcompetes (destroys) apache
kavin retweeted
One failure mode that I’ve repeatedly fallen into is thinking “surely someone smarter than me must have already figured this out” when in fact no one has.
Replying to @KelseyTuoc
If you are smart, competent, a fast learner and willing to really throw yourself into something, you can answer a question to which our civilization does not have an answer with weeks to months of work. You can become an expert in months to years.
kavin retweeted
This week was arguably one of the biggest in 2025 for India's AI industry.
I summarized everything from Sarvam, Maya Research, Lossfunk, CognitiveLab, and more.
Here's everything that happened:
kavin retweeted
Replying to @zephyr_z9
no one owns that domain lol, some random dude keeps on changing it to whoever owns the SOTA and it is the funniest thing ever
kavin retweeted
turns out the founder of anthropic used to be a geophysicist, a phd student in computational neuroscience at princeton, a consultant, and then literally invented gpt-3 at openai. insane arc
Replying to @daltonc
Was just thinking about a guy I know who decided to start his career over and devote his life to AI research like 10 years ago, and is now one of the founders of Anthropic.
kavin retweeted
This whole "no moat" situation in AI is fascinating, and it's probably the best thing that could have happened for humanity. Here's the deal: typically in tech, companies build these incredible fortresses around their innovations - patents, trade secrets, specialized hardware, you name it. It's how they keep competitors at bay and maintain their market position.
But AI? It's completely different. Every time OpenAI drops something groundbreaking, within months you've got DeepSeek in China, Google, Microsoft, and a dozen other players replicating their results. The secret sauce isn't so secret anymore. There's no special hardware you need, no magical dataset that only one company has access to. The algorithms? They're largely based on public research. The data? It's mostly from the internet, which belongs to everyone.
And this is incredible for consumers. Think about it - when you've got this level of competition, with everyone able to replicate everyone else's advances, what happens? Prices plummet. Innovation accelerates. No single company can get away with charging premium prices or restricting access because someone else will just come along and offer it cheaper or even free.
The only real advantage anyone has in this space is speed - being first to market. But even that advantage is shrinking. What used to be a 12-24 month lead for OpenAI has shrunk to just a couple of months in some cases. Sure, they've got that first-mover advantage with ChatGPT being one of the most visited websites globally, but that's more about marketing and brand recognition than technological superiority.
This is exactly how transformative technologies should work. Look at historical parallels - the printing press, radio, the internet. Once these technologies were understood, they spread everywhere because they were fundamentally democratic technologies. AI is following the same pattern, and it's beautiful because it means no single entity - whether that's a corporation or a government - can monopolize this power.
We're watching the democratization of intelligence in real-time. And sure, some companies might not love this situation, but for humanity as a whole? This is exactly what we needed. When everyone has access to these tools, when anyone can innovate and build on top of them, that's when we'll see the real revolution begin. No gatekeepers, no artificial scarcity, just pure technological progress driven by global competition and collaboration.
kavin retweeted
DeepSeek discovered their new model having an "aha" moment where it developed an advanced reasoning technique, entirely on its own
kavin retweeted
every man has a Stack
for some it is creatine + magnesium
for others it is arch linux + vim
some just answer perplexity + claude
others have vercel and cloudflare, or nginx and redis
some guys will even say bitcoin and ether
the essence of being a man: you must have a Stack.
kavin retweeted
can’t believe I missed out on one of the greatest money-printing opportunities of all time. FML.
my existence on this platform is a disgrace.
$7B market cap is unfuckingreal. I’m exiling myself from X temporarily. I need to cry and rethink my entire life.
My NEW Official Trump Meme is HERE! It’s time to celebrate everything we stand for: WINNING! Join my very special Trump Community. GET YOUR $TRUMP NOW. Go to gettrumpmemes.com/ — Have Fun!
kavin retweeted
We are now down to 160 GPU hours for speed running SOTA evals in the 100-200M param smol model class (~31x less compute!) 🚀
Thanks to @KoszarskyB, @leloykun, @Grad62304977, @YouJiacheng et al. for their work that helped to make this possible, and to @HotAisle for their great support :)
There are some great tricks to extract from the Deepseek paper, looking forward to improving the training efficiency much much more!
Andrej Karpathy@karpathy26 Dec 2024United States
United StatesConnected via United States App StoreAccount-level information, not a live location or per-post device.
DeepSeek (Chinese AI co) making it look easy today with an open weights release of a frontier-grade LLM trained on a joke of a budget (2048 GPUs for 2 months, $6M).
For reference, this level of capability is supposed to require clusters of closer to 16K GPUs, the ones being brought up today are more around 100K GPUs. E.g. Llama 3 405B used 30.8M GPU-hours, while DeepSeek-V3 looks to be a stronger model at only 2.8M GPU-hours (~11X less compute). If the model also passes vibe checks (e.g. LLM arena rankings are ongoing, my few quick tests went well so far) it will be a highly impressive display of research and engineering under resource constraints.
Does this mean you don't need large GPU clusters for frontier LLMs? No but you have to ensure that you're not wasteful with what you have, and this looks like a nice demonstration that there's still a lot to get through with both data and algorithms.
Very nice & detailed tech report too, reading through.
kavin retweeted
we are truly living in the (ai)ge of acceleration
in the past month alone:
- openai o3: passing ARC AGI benchmark
- google veo: most advanced video model
- genesis: best generative physics engine
- gemini flash 2.0: cheap powerful dev model
- openai sora: released and launched
- google quantum chips
- openai o1 pro (unlimited use)
- llama 3.3 70B (best open source model)
- nvidia jetson nano super + blackwell chips
- xai grok new image generator
- midjourney new world building feature
- cognition’s devin AI engineer released
- elevenlabs super fast flash model
- google notebook LM and whisk updates
- tencent deepseek model
- humane cosmOS
- deepmind genie 2
- amazon nova
+ can’t wait till anthropic joins the party





































