Ashish Nagar

1.9K posts

Ashish Nagar banner
Ashish Nagar

Ashish Nagar

@ashishnagar

Entrepreneur, Founder @TheLevelAI, Product #Alexa, Climate Policy Buff, #Stanfordgsb Alumni, #IITdelhi alumni, Indian.

Stanford, CA Katılım Ekim 2009
476 Takip Edilen535 Takipçiler
Ashish Nagar
Ashish Nagar@ashishnagar·
We are literally paying Oil companies to abandon clean energy projects in America. Projects which would reduce electricity costs for everyone and help the climate. Could not believe my eyes. But at this point nothing should surprise you. At the same time China is running away with electrification. nytimes.com/2026/03/23/cli…
English
0
0
0
17
Ashish Nagar
Ashish Nagar@ashishnagar·
@kwharrison13 They might be good IRR generators but for the first 5-10 years are really winging it on the board. Learning from being in board meetings and pattern matching in startups is not ideal for the startup teams.
English
0
0
0
117
Ashish Nagar
Ashish Nagar@ashishnagar·
Totally agree. That is the way to go for most approaches except coding in my view. Because in coding the big labs are making products themselves, will continue to reinvent state of art and compete with Cursor or whoever else. Why would a developer use even a 1% lower performance model?
English
0
0
0
179
Yuchen Jin
Yuchen Jin@Yuchenj_UW·
People dunk on Cursor like: “it’s just Kimi K2.5,” “look inside, it’s a Chinese model.” There’s no shame in building on top of strong base models and doing your own post-training or RL (as long as you respect the license). In most cases you don’t need to pretrain from scratch. I think the whole industry will shift toward more post-training and RL on Chinese open-source models. That’s also part of why we’re seeing the biggest GPU shortage and H100 price spike right now.
Yuchen Jin@Yuchenj_UW

Cursor’s Composer 2 is likely built on Kimi K2.5. The model URL + tokenizer are strong signals. I love this direction: companies mid-train and post-train on top of OSS LLMs. Prediction: open-source model labs will monetize by taking a cut when others build on top of their models and scale to millions of real users. They will enforce this via licensing. That’s the flywheel. That’s how open-source AI thrives.

English
68
24
489
52.4K
Ashish Nagar
Ashish Nagar@ashishnagar·
Cursor had to do this to control ballooning costs. Fine tune a Chinese Open Source model. The challenge is advance coding is a specialized problem and the ONE area where smaller models will not work as well. If Anthropic models are even 1% better than an Open Source model then why would you not use Claude Code directly. And if both Anthropic and Cursor have to make 70% Gross Margins while competing with each other something needs to give. The bewildering question is why do Cursor investors not get this and keep on funding selling $2 for $1 company.
Lee Robinson@leerob

Yep, Composer 2 started from an open-source base! We will do full pretraining in the future. Only ~1/4 of the compute spent on the final model came from the base, the rest is from our training. This is why evals are very different. And yes, we are following the license through our inference partner terms.

English
0
0
2
61
Ashish Nagar retweetledi
clem 🤗
clem 🤗@ClementDelangue·
Looks like it’s confirmed Cursor’s new model is based on Kimi! It reinforces a couple of things: - open-source keeps being the greatest competition enabler - another validation for chinese open-source that is now the biggest force shaping the global AI stack - the frontier is no longer just about who trains from scratch, but who adapts, fine-tunes, and productizes fastest (seeing the same thing with OpenClaw for example).
Lee Robinson@leerob

Yep, Composer 2 started from an open-source base! We will do full pretraining in the future. Only ~1/4 of the compute spent on the final model came from the base, the rest is from our training. This is why evals are very different. And yes, we are following the license through our inference partner terms.

English
59
122
1.1K
149.7K
Ashish Nagar retweetledi
Gaurav Ahuja
Gaurav Ahuja@gauravahuja·
One of these two groups is mispriced Private AI labs: OpenAI valued around $840B, Anthropic north of $600B on secondaries. Both at 30x+ ARR. Public giants: Microsoft at ~$3T on 23x forward earnings. Amazon at ~$2.3T on 28x. Microsoft likely owns ~25% of OpenAI. Amazon likely owns ~15% of Anthropic and ~5% of OpenAI If private investors are pricing these labs for a $5T+ venture-style outcome then… Microsoft’s implied stake in a $5T OpenAI is $1.25T embedded inside a $3T company. Amazon’s combined stakes embed roughly $1T inside a $2.3T company. Publics too cheap on Al exposure? Or privates/secondaries in bubble territory? Which breaks first?
English
63
41
933
180.2K
Ashish Nagar
Ashish Nagar@ashishnagar·
Great point. The share of AI is right now being divided across many applications which will reduce some data center loads. BUT... the huge market of untapped applications where AI will transform should pull it up, no? In terms of AI Apps, feel like <5% penetration in potential use cases in the digital and physical world.
English
0
0
0
233
Josh Wolfe
Josh Wolfe@wolfejosh·
everyone is saying "there won't be enough compute." I think the opposite the amount of capex being deployed for multi-gigawatt data centers does not make sense why the narrative is going to shift from centralized mega-clusters to localized edge inference...
The Information@theinformation

.@wolfejosh, co-founder of Lux Capital, discusses why the massive build-out of AI data centers may be overextended: "The amount of spend, the amount of CapEx, the amount of build for these multi-gigawatt data centers, it to me does not make sense." “I'm just not that optimistic that all this compute is actually going to be needed."

English
47
22
215
80.5K
Ashish Nagar
Ashish Nagar@ashishnagar·
@cryptopunk7213 Meta already uses Gemini for a bunch of internal use cases. Did a $500M contract not that long ago. Hiring CEO of a data annotation company who is great in Davos is a hard way to make AI models.
English
1
0
2
420
Ejaaz
Ejaaz@cryptopunk7213·
holy shit Meta might ditch ai efforts and go with google gemini instead Meta to delay their new AI model launch and use gemini to power Meta AI - HUGE fucking win for google: - Meta's avocado model underperformed frontier models from openai, google and anthropic (shitty reasoning, coding etc) - this comes after Meta spent $20B hiring a new AI team thats produced... no ai models. - looking at licensing google gemini (google just licensed to Apple for $1B per year) Google is fast-becoming the preferred model for the largest companies in the world. Meta has 3.6 BILLION MAUs if this happens google will single-handedly have the largest AI distribution of any company.
Ejaaz tweet media
Andrew Curran@AndrewCurran_

META has delayed the release of Avocado until at least May after it underperformed on internal evals, according to reporting by the NYT. They are considering licensing Gemini from Google as a temporary solution.

English
122
159
1.9K
570K
Ashish Nagar
Ashish Nagar@ashishnagar·
@kapskom I listen to @ThePrintIndia a lot. But it is quite obvious how @ShekharGupta does not probe the government at all. Maybe its the cost of doing business at this point. No cut the clutter on Indore deaths, Epstein issue, Adani, Rupee free fall, AI summit fiasco.
English
3
12
163
8.4K
Kapil Komireddi
Kapil Komireddi@kapskom·
Shekhar Gupta sent me word last night that my contract with The Print—where I was a columnist—will not be renewed because I called Narendra Modi a “coward” and a “curse on India”. Gupta was perfectly fine with my calling the Gandhi family “a plague on India”, but Modi apparently is too sacred a cow. It is not clear to me whether Gupta took this call or was instructed to take it. The former would be worse because it would mean that he censors those with whom he disagrees. I obviously do not want anyone else to get into trouble for my posts—which is why my Twitter page carries the disclaimer that “I speak solely for myself here”—and I don’t wish Gupta ill. The Print gives opportunities in journalism to people who wouldn’t otherwise have any. Does sustaining such an enterprise require compromise? I don’t know. What I do know is that I value Gupta’s work as a chronicler of India’s republican vicissitudes. Which is what makes his pitiful conduct rather painful. I’m sufficiently blessed not to be affected by this in any material way. I’m vaguely worried that I’ll be hassled by the authorities, but as far as work is concerned I’ve other avenues in which to publish. I’m just sorry to see a man I mildly respected reduced to cutting ties out of fear. I don’t want anybody to become a martyr—but I’d be dishonest if I didn’t admit that I’m troubled by the knowledge that Gupta is a far smaller man than I had supposed. He didn’t have the courage—or even the decency—to speak directly to me. He delegated that duty to someone else. Sad.
English
981
2.6K
9.9K
949.4K
Ashish Nagar retweetledi
Josh Wolfe
Josh Wolfe@wolfejosh·
Sharing this publicly––hope it's not needed but helps you if it is. Lux team sent this memo to all Lux family founders yesterday "We send notes like this not because something is wrong, but because the COST of preparation is trivially LOW and the VALUE of being positioned well is asymmetrically HIGH. This isn’t a macro call. It’s a set of observations about correlated risks that are worth your attention. And a set of practical suggestions regardless of whatever happens next."
Josh Wolfe tweet mediaJosh Wolfe tweet media
English
31
79
1.1K
367.8K
Ashish Nagar retweetledi
Elad Gil
Elad Gil@eladgil·
In short run, many coming "we are doing layoffs due to AI" PR will be correcting for sins of 2020 era of over hiring versus anything to do w AI Many larger tech companies could slim down 50% w/o any AI changes Most AI productivity impact is still around bend
English
89
127
1.4K
138.8K
Ashish Nagar
Ashish Nagar@ashishnagar·
@chamath 100% @chamath this is the energy revolution which the AI world needs. You are in a unique position to do something about it with @DavidSacks and get Trump off the coal,oil,gas band wagon. At the minimum it is common sensical to have "all of the above" energy strategy.
English
0
0
8
566
Chamath Palihapitiya
Chamath Palihapitiya@chamath·
China added more solar capacity in 2025 than America has installed in its entire history. That's the most important energy chart you'll see today. And 2025 was also the first year when small-scale distributed solar pulled in more investments than utility-scale solar farms globally. Considering that the U.S. has hundreds of GW stuck waiting for grid connections, the conditions are aligned to start putting solar + storage on every American home. My research team put together a Deep Dive on solar, if you want the full breakdown. Here’s the link: chamath.substack.com/p/solar-deep-d…
Chamath Palihapitiya tweet media
English
627
883
4.1K
740.3K
Ashish Nagar
Ashish Nagar@ashishnagar·
I am huge AI optimist and can't believe everyday what AI agents are able to do. Still, some of the reasons why the AI CS Zendesk scenario is not likely in next 2 years: 1. Chatbots and Voicebots have been around for a long time. The current generation of voice agents make them 10X better in terms of quality, so CSAT is higher but they don't fundamentally make the hard problems easy. Most large companies which is buying a Sierra, LevelAI, Decagon voice agent is replacing an existing AI agent. They already had significant automation and the parts which could not be automated before are just much harder. 2. Hard problems in enterprise are connecting disparate systems to get automation done in a repeatable and reliable way. GPT based AI Agents are not repeatable and reliable to the extent enterprises want. We are making progress but a long way to go. 3. It's not the AI it's the API. We have an e-commerce customer who don't have basic APIs to automate the most simple order status query. Leave alone long tail queries. Thousands of customers would need IT transformation at unprecedented scale to make it happen. That is why folks like Infosys, Accenture become more important then less. IT transformation is a 10-20 year time frame thing. See how on-prem to cloud migration still is in mid-journey 20 years in. 4. A human service agent is not just an order taker or question answering system. They are an advisor, counselor, shopper, nurse practitioner. Tasks which are inherently human oriented for a lot of consumers.
English
1
0
4
302
Ashish Nagar
Ashish Nagar@ashishnagar·
I have spent the last 5 years building AI for customer service. Talk to 100s of customers, deploy and build AI for a living. Am a deep AI optimist. Your Zendesk imagination is far from reality as possible. I am not surprised that you wrote it, I am surprised people take it seriously.
English
4
0
13
5.9K
Citrini
Citrini@citrini·
I spent 100 hours over the past week researching, writing and editing the piece we just put out. It’s a scenario, not a prediction like most of our work. But it was rigorously constructed, dismissing it outright requires the kind of intellectual laziness that tends to get expensive. And we’ve released it for free. Hopefully you enjoy it. citriniresearch.com/p/2028gic
English
907
2.4K
14.7K
10.7M

Keşfet

@kwharrison13 @cryptopunk7213 @kapskom @ThePrintIndia @ShekharGupta @pitdesi @elonmusk @BarackObama