Prashant Mahajan

People don’t remember how insane Uber, Doordash, Flipkart, Google Pay (India) early days were.

Germany is on path of Canada. youtu.be/Uv5luvQ3Mtg?

Why don’t people enjoy weekends?

Most founders don't wanna enjoy what they are building, they wanna enjoy building. It's sad but it is real, people don't care about product or problem, they wanna be a founder. Check YC batches, problems they are solving are not what 20 somethings were doing years back. Facebook, Twitter, Figma, Basecamp, Uber, Airbnb, Apple, those were problems founders related too. Or problems they faced in job (most dev tools), generally by 30ish folks. Now people are building EHR, Call Center Solutions, Mechanic in their prime. I blame this new trend of request for startups a bit. Investors telling what we want you to build, and founders leaping on it. Can't remember last good company where founder was looking for a problem.

Problem that this cheap and fast SOC2 and other compliance created is that now every buyer expects this check box, irrespective of need. And that is going to hurt startups to get genuinely SOC2 compliance. It used to be something at mature stage or domain, but now every pre seed startup has it.

@Suhail @bchesky you guys are weak. we sold $80K on $240K

Wait till people find out there are bunch of similar companies g2.com/products/delve… Many of them do the same thing, without proof don't wanna take name, but have experience them.

Men used to go explore oceans and cross continents for greatness. Now the take flight for this.

The company you’re buying GTM technology from should be exceptional at GTM. If they’re not, don’t buy their product.

Who is the greatest founder poker player in the world? We'll find out in just over 2 weeks at the largest founder poker tournament of all time, the $100,000 cash prize, no buy-in, Monaco Invitational. Invitations are going out now. If you think you qualify, comment below.

Oscar is becoming trash. Neither of the top three were worthy.

We launched 60 billboards across San Francisco and Highway 101 this week. Most startup billboards are forgettable. Ours is just a giant $.  Here’s the Monaco billboard strategy and what startups should know before spending money on out-of-home advertising 🧵

$100,000 cash prize founder poker tournament. No entry fee. Announcing The Monaco Invitational, presented by @MonacoGTM . This will be the largest founder poker tournament (and party) of all time. In addition to giving away $100,000 to the winners in the free to play tournament, we'll have celebrity guests, an incredible food and beverage program, and the coolest venue in San Francisco. This will be the event that sets the gold standard for startup events. Invitations are reserved for Monaco customers, anyone who refers a Monaco customer, and friends of the firm. Know a founder who should be there? Tag them in the comments. Check out more details in the invitation on tournament rules and eligibility.

Sadly, we don’t really treat guests like gods in India we just talk about in culture.

Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, today I measured that the leaderboard's "Time to GPT-2" drops from 2.02 hours to 1.80 hours (~11% improvement), this will be the new leaderboard entry. So yes, these are real improvements and they make an actual difference. I am mildly surprised that my very first naive attempt already worked this well on top of what I thought was already a fairly manually well-tuned project. This is a first for me because I am very used to doing the iterative optimization of neural network training manually. You come up with ideas, you implement them, you check if they work (better validation loss), you come up with new ideas based on that, you read some papers for inspiration, etc etc. This is the bread and butter of what I do daily for 2 decades. Seeing the agent do this entire workflow end-to-end and all by itself as it worked through approx. 700 changes autonomously is wild. It really looked at the sequence of results of experiments and used that to plan the next ones. It's not novel, ground-breaking "research" (yet), but all the adjustments are "real", I didn't find them manually previously, and they stack up and actually improved nanochat. Among the bigger things e.g.: - It noticed an oversight that my parameterless QKnorm didn't have a scaler multiplier attached, so my attention was too diffuse. The agent found multipliers to sharpen it, pointing to future work. - It found that the Value Embeddings really like regularization and I wasn't applying any (oops). - It found that my banded attention was too conservative (i forgot to tune it). - It found that AdamW betas were all messed up. - It tuned the weight decay schedule. - It tuned the network initialization. This is on top of all the tuning I've already done over a good amount of time. The exact commit is here, from this "round 1" of autoresearch. I am going to kick off "round 2", and in parallel I am looking at how multiple agents can collaborate to unlock parallelism. github.com/karpathy/nanoc… All LLM frontier labs will do this. It's the final boss battle. It's a lot more complex at scale of course - you don't just have a single train. py file to tune. But doing it is "just engineering" and it's going to work. You spin up a swarm of agents, you have them collaborate to tune smaller models, you promote the most promising ideas to increasingly larger scales, and humans (optionally) contribute on the edges. And more generally, *any* metric you care about that is reasonably efficient to evaluate (or that has more efficient proxy metrics such as training a smaller network) can be autoresearched by an agent swarm. It's worth thinking about whether your problem falls into this bucket too.

@pravapayments @Visa Congrats on launch.

Prava is integrating with @Visa Intelligent Commerce to enable secure, agentic, card-based checkout for AI assistants and apps . AI agents can now complete real purchases on a user’s behalf using their existing Visa credentials with security, control and trust built in. Available in the US & SEA.

@selinayfilizp Taksim is good too. Also found this place for good Pide maps.app.goo.gl/4qaHVF6FADHHmb…

Someone yesterday told me they don’t know a single good Turkish place in SF. My top 5: -Dalida (Presidio) fine dining, Turkish-Mediterranean -Healthyish Republic (Mission) good Turkish breakfast and lunch items -Lokma (Richmond) cozy neighborhood gem -Anatolian Table (Mission) authentic, low-key, good pide -Kitchen Istanbul, heard it's also pretty good but I only tried Kunefe there. PS: If we include peninsula, Meyhouse (Palo Alto) best overall and Mangal in Sunnyvale if you want real Turkish BBQ over charcoal. Let me know if i am missing anything!

17000 Tokens/Sec

People are forgetting what happened in neighbouring Nepal.

@iuditg Not unpopular, everyone i talk to holds same opinion.

Unpopular opinion: In next three years, barely anyone will use apps like loveable, replit, emergent, etc.