QV

Vertical SaaS, Marketplace, Roll-up businesses

[The All-In Podcast @theallinpod]

🚨 BIG INTERVIEW: SPENCER PRATT JOINS ALL-IN

David Friedberg sits down with Spencer Pratt to discuss his fight to save Los Angeles.

(0:00) Spencer Pratt vs. the Machine

(3:01) Inside the Palisades Fire: Drained Reservoirs, No Sirens & Watching His House Burn on His Phone

(14:03) Why He's Running for Mayor: FireAid's $100M Scandal & the NGO Corruption Nobody Talks About

(28:10) Karen Bass at 20% & the Real State of LA: Crime, Homelessness & a City in Free Fall

(38:23) Spencer's Plan to Fix LA: Enforcing Laws, Auditing Everyone & the Billionaires Ready to Rebuild

(52:22) Hollywood, LAUSD & Small Business: What It Actually Takes to Make LA #1 Again

(56:25) The Permitting Nightmare Killing Small Business & How AI Fixes It Overnight

(1:04:22) His 8-Year Vision

@friedberg @spencerpratt

---------------------------------------

Thanks to our partner https://t.co/HFpRpqlQLi for making this possible

Most advertisers have never heard of the platform with an $11B annual run rate in ad spend.

https://t.co/HFpRpqlQLi by AppLovin — 1B+ daily active users, full-screen video ads watched for a median of 35 seconds, and businesses are profitably spending hundreds of thousands of dollars a day on it.

Advertiser access is in closed beta. The window is open at https://t.co/GDPKNzfNXy

@AxonAdsManager

[QV @will_be_fin]

There are two types of ppl, interpreter and translator. The former is an approximation to client's solutions and the later is a layer for execution. Neither should exist. I am excited to see we are finally able to achieve someting without convincing machines what I want.#AI

[QV @will_be_fin]

https://t.co/weXOlMf6R7

[QV @will_be_fin]

https://t.co/mixyYrsG9Z

[QV @will_be_fin]

For 10 years, TLS has been an expensive, buggy mess. No logic, constant crashes, and arrogant service. Someone please build a replacement.

[Andrej Karpathy @karpathy]

I had the same thought so I've been playing with it in nanochat. E.g. here's 8 agents (4 claude, 4 codex), with 1 GPU each running nanochat experiments (trying to delete logit softcap without regression). The TLDR is that it doesn't work and it's a mess... but it's still very pretty to look at :)

I tried a few setups: 8 independent solo researchers, 1 chief scientist giving work to 8 junior researchers, etc. Each research program is a git branch, each scientist forks it into a feature branch, git worktrees for isolation, simple files for comms, skip Docker/VMs for simplicity atm (I find that instructions are enough to prevent interference). Research org runs in tmux window grids of interactive sessions (like Teams) so that it's pretty to look at, see their individual work, and "take over" if needed, i.e. no -p.

But ok the reason it doesn't work so far is that the agents' ideas are just pretty bad out of the box, even at highest intelligence. They don't think carefully though experiment design, they run a bit non-sensical variations, they don't create strong baselines and ablate things properly, they don't carefully control for runtime or flops. (just as an example, an agent yesterday "discovered" that increasing the hidden size of the network improves the validation loss, which is a totally spurious result given that a bigger network will have a lower validation loss in the infinite data regime, but then it also trains for a lot longer, it's not clear why I had to come in to point that out). They are very good at implementing any given well-scoped and described idea but they don't creatively generate them.

But the goal is that you are now programming an organization (e.g. a "research org") and its individual agents, so the "source code" is the collection of prompts, skills, tools, etc. and processes that make it up. E.g. a daily standup in the morning is now part of the "org code". And optimizing nanochat pretraining is just one of the many tasks (almost like an eval). Then - given an arbitrary task, how quickly does your research org generate progress on it?

[QV @will_be_fin]

@karpathy Since you're already running this in tmux with 8 GPUs, have you considered adding a "Librarian" Agent?
Instead of just 8 researchers, have 1 agent monitor the logs of the other 7, find contradictions in their data, post a "Weekly Review" style summary to shared comms.

[QV @will_be_fin]

@0xhashlol thanks, I'll test it.

[QV @will_be_fin]

tried to replicate openclaw setting for cursor because I like IDE but failed because lack of coordinations between sub-agents - currently using a setup.md file to enable communications.

[QV @will_be_fin]

Currently testing replicating Openclaw's sub-agent settings in Cursor and it seems unusable. @cursor_ai any suggestions?

[QV @will_be_fin]

@GianniDalerta @johann_sath please share more

[Perplexity @perplexity_ai]

Introducing Perplexity Computer.

Computer unifies every current AI capability into one system.

It can research, design, code, deploy, and manage any project end-to-end.

[QV @will_be_fin]

@ArmanHezarkhani Will test this when I build financial models next time

[Elvis @elvissun]

https://t.co/DotZ3V6XhJ

[chiefofautism @chiefofautism]

i mapped the ENTIRE supply chain behind a single ChatGPT query

76 nodes in 13 countries with 10 layers, from a quartz mine in North Carolina to your chat window

so i built an interactive map where you can trace every path yourself

every time you type a prompt, you are touching brazilian sugarcane that turned into ABF varnish by Ajinomoto in Japan that used to package Nvidia GPUs in Taiwan

a single quartz mine in Spruce Pine NC that supplies the ENTIRE semiconductor industry with crucibles, no backup, one landslide and chip production stops globally

ASML in the Netherlands, the ONLY company on earth that makes EUV lithography machines, they need Zeiss mirrors polished to less than ONE ATOM of roughness, and TRUMPF lasers from Germany to power them

chinese germanium, ukrainian neon gas, chilean copper, australian iron ore

all flowing through TSMC fabs that print at 2 nanometers, thats 10 atoms wide

this is a PHYSICAL supply chain more fragile than most people realize

everyone debates which model is better

nobody talks about the quartz mine that all of them depend on

[Shubham Saboo @Saboo_Shubham_]

https://t.co/I00UtFrJdg

[QV @will_be_fin]

@thecryptoskanda @hosseeb @KyleSamani @cdixon @Hertzflow_xyz 再牛逼的项目到了后期都需要有人接盘，不管哪里的项目到了后期全世界只有那么十家左右的机构能接，所以亚洲的天花板比国外天生就低，过程中的容错率也低，更加注重生意本身

[QV @will_be_fin]

@thecryptoskanda @hosseeb @KyleSamani @cdixon @Hertzflow_xyz 我给几个数据支撑点，同样是美元Credit Fund，美国的target return 在北美10%很好，15%非常好，到了亚洲就要多加10%的risk premium，保底15%个点的cash interest，再加10%的超额收益，所以过去传统行业亚洲能支撑这种融资成本的一定是超高利润率，超高现金流行业，比如房地产。

[Bitcoin Archive @BitcoinArchive]

BLACKROCK CEO LARRY FINK: "There is a role for gold and there's a role for Bitcoin."

"You own these assets if you're frightened of currency debasement."

[QV @will_be_fin]

@wasifmoramen @chainlink @HotpotDAO212 @HotpotDAO888 @hotpot_dao try to DM you but failed. LMK. Thanks!

[QV @will_be_fin]

. @chainlink we want some hotpot #smartcon