Snowman Research

37 posts

Snowman Research banner
Snowman Research

Snowman Research

@ResearchSnowman

Come for the stock picks. Stay for the AI Research insights. 📍NYC | SF

New York, USA Se unió Mayıs 2026
150 Siguiendo13 Seguidores
Tweet fijado
Snowman Research
Snowman Research@ResearchSnowman·
🧵1/ Why I invested in Cerebras pre-IPO Colleagues and I benchmarked them against competitors, and Cerebras was the fastest by a wide margin. Getting an answer in 1 second vs. 10 completely changes a product experience, particularly in the cutthroat AI world.
English
1
0
5
369
Snowman Research
Snowman Research@ResearchSnowman·
As Dwarkesh points out, domain specific rubrics have driven a lot of recent AI progress. Let’s say a Lab spends $10M on 10,000 of these rubrics for challenges an accountant might face, and then spends millions more training the model. If this makes accountants 10% more efficient, the ROI on that investment is fantastic. In fact, even if the data required is 3-5x more expensive than I’ve estimated it’s still likely a great investment (and why labs are only accelerating their data spend). The challenge is the long tail. As you move into increasingly specialized professions with fewer practitioners and lower aggregate economic value, the economics of building bespoke rubrics become less attractive. That’s where recursive self-improvement becomes important. If models can reliably generate and evaluate increasingly difficult tasks for themselves, they can become more data efficient.
Dwarkesh Patel@dwarkesh_sp

Narration: the data efficiency black hole. 00:00:00 – What is really driving AI progress? 00:03:11 – Comparing human vs AI sample efficiency 00:08:46 – Does sample efficiency matter? Also on pod and YouTube feed.

English
0
0
0
56
Snowman Research
Snowman Research@ResearchSnowman·
@_shahaf_ Very well timed. 60B is a 2.5% change in SpaceX stock at the current valuation
English
0
0
1
19
Shahaf
Shahaf@_shahaf_·
@ResearchSnowman I find the cursor acquisition well timed on musks side. It’s all stock and the stock is pumped right now. Definitely going down a lot once investors can start selling en masse.
English
1
0
1
41
Snowman Research
Snowman Research@ResearchSnowman·
$SPCX is now becoming vertically integrated in a way OpenAI and Anthropic are not: they own the data centers to the application layer Today, Cursor announced a new 1.5T model, likely pretrained on top of the upcoming Grok v9 medium model. This means Cursor is leveraging SpaceXAI compute in two ways: 1. Larger base model - previous Cursor model was based on the 1T Open Source Kimi model 2. 10-20x more post training training compute Next question, when will we get a Cursor model trained on Grok large? And when will SpaceXAI buy Tesla's custom AI chips? Or maybe SpaceX just buys Tesla 😉
Max Weinbach@mweinbach

New Cursor model trained from scratch! 1.5T+ and all seemingly Blackwell chips

English
1
0
1
336
Snowman Research
Snowman Research@ResearchSnowman·
9/ Routing will likely only become more of a thing Fable is not just a model. It is a capability-gated system. For cyber, bio, and frontier LLM development, classifiers ensure the models behaves as it should. That is a departure from the idea of giving the model itself the ability to know when to decline answering questions in certain areas. This routing approach likely simplifies the model training because you can focus on training the most capable model possible and trust that its abilities can be selectively gated at deployment time.
English
0
0
0
58
Snowman Research
Snowman Research@ResearchSnowman·
8/ Just cause it can hack doesn't mean it can manage a vending machine Fable 5 slightly underperformed Opus 4.8 on Vending-Bench 2, a task where it must manage a vendoing machine. This task takes common sense business knowledge and negotiation skills.
English
1
0
0
81
Snowman Research retuiteado
Ravid Shwartz Ziv
Ravid Shwartz Ziv@ziv_ravid·
🏀🤖 There was lovely weather this weekend, so I let my agents run without babysitting. I combined the two things all my feeds are obsessed with right now: the New York Knicks and autonomous research agents. So, who will win the game today? 🧵
Ravid Shwartz Ziv tweet media
English
2
1
8
6K
Joe Holder
Joe Holder@JoeHolder_·
This is so NYC specific it’s insane 😭😭😭
English
51
858
12.6K
695.6K
Snowman Research
Snowman Research@ResearchSnowman·
@citrini We are still compute bound. If there was greater chip supply capability would be improving even faster.
English
0
0
0
412
Snowman Research
Snowman Research@ResearchSnowman·
Surprised he isn’t using voice mode to take the calls for him
Matt Turck@mattturck

A day in the life of a VC in 2026: 9am: Board meeting. My main value-add is aggressively pushing for Anthropic/OpenAI usage in non-engineering functions. Briefly ponder how I became an SDR for foundation model companies. 1pm: Lunch with another VC. We discuss how startups can find "blue ocean" away Anthropic/OpenAI. We conclude we should probably just invest the rest of our funds directly into Anthropic/OpenAI. 3pm: Pitch meeting. Me: "Do you run on Anthropic or OpenAI?" Founder: "Both." Debating internally whether a company reselling Anthropic/OpenAI with a 10% gross margin is a good investment but hey, at least they're in the "token flow". 4:30pm: Deep due diligence. I ask Claude if it plans to build this exact startup natively in its next release. Same to ChatGPT. They both say yes. I pass on the deal. 6pm: Urgent call from a portco CTO: "We need an intro to upgrade our Anthropic tier!" I immediately agree to help them spend more of the venture dollars we just invested in them, on Anthropic. 8pm: Brainstorming next guests for the podcast. Thinking I should probably just try to get some folks from Anthropic and OpenAI.

English
0
0
1
176
Snowman Research
Snowman Research@ResearchSnowman·
Love the concept of idea-driven vs. goal-driven research from this classic post by @johnschulman2. The best AI researchers are obsessed with a long-term objective (e.g. making robots walk), not a particular method joschu.net/blog/opinionat…
English
0
0
0
85
Elad Gil
Elad Gil@eladgil·
The events of the last 6 months in technology are arguable amongst the most important in human history The tools now increasingly exist for recursive self improvement of models & agents We are likely in very early lift off & exponential Largely unnoticed outside of tech
English
267
424
4.7K
583.4K

Descubrir

@_shahaf_ @JoeHolder_ @wolfejosh @citrini @johnschulman2 @eladgil @elonmusk @BarackObama