Watched @karpathy's must watch 3.5‑hour LLM deep dive and here are my favorite takeaways, surprises, and musings.

He is a god tier communicator and technical genius

My thread is below but I would recommend watching it as its the best way to learn 1a/ One standout example: DeepSeek-R1. Karpathy highlights it as a reasoning LLM pushed to its limits with reinforcement learning

Instead of just mimicking textbook solutions, DeepSeek was trained via trial and error to solve problems, especially tough math questions

It’s no secret that Delphi Ventures has been deep into Crypto x AI

Just like crypto, the best way for us to learn is by using, building, and experimenting with the tech

Today, we’re thrilled to share the next step in that journey: DelphAI, our AI-powered Ventures analyst 🤖🧵 Excited to introduce DelphAI, Delphi Ventures' AI analyst that's transforming how we evaluate crypto projects

We are announcing

• DelphiAI to the world
• $1M investment competition for project submitted to DelphiAI
• DelphAI's Architecture

Lets dive in

Today I had the chance to read @Delphi_Digital's latest Crypto x AI report by @PonderingDurian

AI is the most powerful technology ever and Crypto is unstoppable programmatic money, together they create a new and better future

My favorite parts below 🧵

members.delphidigital.io/reports/deai-i… "English is the hottest new programming language"

DeFi Summer brought 1000s of crypto experiments that created the DeFi blue chips

AI lowers the barriers app creation by 100x, and with Crypto for capital and incentives, will lead to an order of magnitude more agent experiments

Crypto x AI Flowchart

I wanted to share my mental model of the Crypto x AI landscape

Long term I think Crypto x AI wins because the most powerful technology of our generation can be owned, governed and iterated by an incentivized global community vs a centralized black box Danger of Centralized AI's effects on millions of applications reaching billions of users.

Just read @Delphi_Digital's The Year Ahead for Infrastructure 2024 report

It's by far the most comprehensive crypto infrastructure read and I feel way smarter having read it

Phenomenal work @CannnGurel, @ceterispar1bus, @yusufxzy and @markodayan

A few of my fav takeaways 🧵 1/ L2 competition will accelerate

Historically L2s have been friendly frens. Not anymore anon

Moving forward the authors expect L2s to silo more and favor bridging within their own ecosystems vs other external L2s

I like heightened competition we need it to push the boundaries

How I avoided Sinus Surgery using Google Bard 🥼🏥

I thought this was an interesting use of AI, so sharing the quick story below

TLDR - Bard found and summarized research studies to give me accurate info to make a decision - in a few hours of searching. I've struggled with Allergies and sinus infections my whole life. I had allergy shots as a kid, and then sinus surgery (Balloon sinulpasty where they insert and inflate a balloon to expand your sinuses)

Recently I've been getting sinus infections again, so back to the ENT

I love hosting @cburniske because he has so much alpha to share on life and the psychology of investing 🧠

We cover Placeholder's $SOL thesis in detail

A few of my favorite takeaways from episode 387 below 🎧

Apple:
Spotify: tinyurl.com/Burniske1
tinyurl.com/Burniske
1/ Despite its long history and large AUM, @placeholdervc has only 6 full time people.

They aren't trying to boil the ocean providing every service and are aggressive in limiting operations overheads and distractions

The focus is on thought and relationships

Great counter post against tons of rollups

IF app-specific rollups are too costly and don't offer necessary technical customization most developers will launch on a shared rollup

This has me thinking on my views on Ethereum Rollups vs App chains with Cosmos vs Solana 💭

🧵1/24

https://twitter.com/neelsalami/status/1708887005445369985

My mental model on competing ecosystems is simple - the easiest and largest place to deploy will house the most viral applications

Why? This community is open to the highest rate of experimentation which offers the highest chance of viral applications

Had this view since 2019

Stephen Wolfram of Wolfram Alpha wrote the absolute best post on ChatGPT and Large Language Models.

It took me about two hours to read, but significantly increased my understanding of what's going on under the hood of ChatGPT.

A few of my favorite takeaways (helps my process) The goal of a large language model is to reasonably continue the text it already has

ChatGPT's LLM estimates these probabilities

Temperature is a parameter that determines how often lower ranked words are used, adding randomness.

LLM's are trained on vast amounts of human text

ETH Denver left me extremely bullish on the future of Ethereum

The convos around

▫️ZK as endgame scalability/privacy tech

▫️MEV solutions/approaches

▫️Learning from founders as a mentor

▫️New Games

were some of the smartest and most thoughtful I’ve seen in a long time On the other hand we have

🔻A macro situation which gives little reason to bid risk assets along with 5% t-bills vacuuming up risk capital

🔻A chaotic regulation by enforcement war ongoing

🔻 Centralized Company damage (Silvergate, Genesis, etc)

Quick take on state of Crypto Venture Funds, Hedge Funds and Fundraising

Venture Funds: Unlikely VCs get a ton of new capital, especially with mega funds already being allocated to by UHNWs and they are sitting on $Bs in dry powder. I do think this is the absolute best time to allocate to pre-seed/seed deals though given no fomo, 1st principles thinking and lower vals.

VCs are setting Vals right now (equity vals and dilution to get to token FDVs)

The past few years founders were setting vals.

Going to the gym today for 55:09 exactly 🔥

Ty @Kevin_Kelly_II @3xliquidated @ccioce + Andrew Krohn

members.delphidigital.io/reports/the-ye… One of my favorite takeaways is on earnings declines during recessions (~26% on average in 01, 08 and 20) and currently NTM EPS estimates for the S&P500 are only down 4% from peak

DCG Situation:

- DCG Owes $2.025B
- Genesis can call their $1.675B loan
- Genesis owes $900M to Gemini

DCG Liq

- Grayscale $10B AUM x 2% = $200M x 3x multiple = $600M
- GBTC/ETHE Holdings = 9.7%/3.8% = $629M with discount, $1.17B at Par
- VC book = Firesale values in a bear

https://twitter.com/WSJBankruptcy/status/1611076805829476352

Valuation multiples for asset management firms vary pretty widely from 1.5-2x for T-Rowe / Franklin Templeton to a few turns higher for other names like Blackrock at 4.5-6x.

I don't think Grayscale would command a company friendly valuation given the unwind risks.

1/ Loved Delphi's DeFi report 📖💰

It's the most comprehensive report I've read on DeFi 👀

Kudos to @ashwathbk, @yeak_ and editors by @Kevin_Kelly_II, Brian M h/t @Delphi_Digital

Sharing my take on their themes + projects I want (and don't) to fund in 2023

1/41 🧵 2/ Setting the stage, Decentralized Finance has gotten crushed in the bear market

Most blue chip DeFi tokens down 80-90% and the longer tail of tokens down more

What's hard to see in the price charts are the teams that just gave up or disbanded. This will take time to show up

My working model of DCG, Genesis, and Grayscale

🏰 DCG: Hold co owns Coindesk, Genesis, Grayscale, and many other cos

💳 Genesis: Lending arm with ~$3B in total loans (per an article)

💰 Grayscale: Cash Cow. $ETHE = $3.6B, $GBTC $10.6B (trading at a 43% discount to NAV) The news item is that Genesis was looking for a $1B emergency loan, probably to meet withdrawals.

My thoughts process is that Genesis accepted ETHE and GBTC as collateral for loans, and now the value of their ETHE and GBTC holdings are 40% below their true value

Just finished METAVERSE by @ballmatthew and it's a seminal work on Web2, The Metaverse and the future.

He also sort of made the case against us being in a simulation given the technical concurrency requirements to do so but that's for another day

A few of my favorite parts 🧵 1/ The Metaverse requires a lot of people all at once, but we are pretty far away from this.

There are countless examples @EveOnline is one (largest battle in history of 12,000 users overloaded the servers), Fortnite (100 players max) and Population one (Max 18 players).

With @Welcome_to_Nor releasing their third Aspect of the Nine NFTs I thought it would be worth diving into the gravity of what @PlayIsFree is building

Brooks and team are offering a new way to think about Play and the Economies around sports and games To start, NOR is championing PlayFi where two non-overlapping magic circles exist

🟢 The Economy (Metagame)
🔵 The Game

🟢 Metagaming is focused on strategy
🔵 Gaming is focused on skill

Gamers play within a magic circle and achieve flow as economic interplay isn't involved.

Tips to Founders Raising Capital

Delphi has invested in over 100 projects and has seen thousands (maybe tens of thousands), so I thought it would be worth sharing some tips for founders raising capital

A thread in my opinion 🧵 /1 First off, don't send a pitch deck. Send an MVP or a Demo.

People want something to see and experience. It beats a deck every single time.

If you're not there yet and are sending a pitch deck, hopefully the following helps a bit

1/ Grabbing a ✈️ so it’s time to catch-up on @Delphi_Digital content

First up, to figure out what’s going on in markets for $BTC / $ETH / Macro

Market Insights – Resistance Holds Strong, But We’re Not Alone by @3xliquidated and @Kevin_Kelly_II

members.delphidigital.io/reports/market… This post is the go-to on what’s going on in markets.

My favorite section was the market risks section as the two talk rates, inflation, historical impacts of recessions on inflation and much more

As liquid markets remain shaky/tbd the market’s appetite to fund private projects with $100M+ Val’s (not equity, but for the token) that are pre-product and/or with limited traction will fade

Too much risk for much less upside, especially in light of an ok market Raise less, at a lower val from funds and builders that will actually help you.

Better R/R for them and less dilution for team assuming you have enough funding here to get you to the next stage.

Then go raise a larger round (yay traction) or forgo it alltogether

1/ Excited to drop a report on @helium with @tbr90 📡

Helium has successfully bootstrapped an IoT network near ~500k nodes and the next era is 5G.

We've been active in the community for a few years now, so we wanted to share our thoughts on Helium's ($HNT) next phase 2/ Helium's model made sense years ago since the project was completely inverting the traditional wireless

📡 $0 spent on spectrum by using free, semi-licensed spectrum
🗼$0 paid to tower companies since nodes are inside people's homes
💰$0 capex since users purchase nodes