The Spellbook team has spent the past 24 hours putting o1 through the paces for legal use cases. Here are our top 3 discoveries + a blog post: 🧵 1. It solves long document revision. Lawyers don't draft contracts much from scratch, they start with a precedent and modify it for their current deal.

It was really hard to get GPT4 to perform well at this. o1 knocks it out of the park. This is what I'm most excited about:

I’m very open to being wrong. Do you have any numbers or anything to share?

Aphex Twin + label probably makes $200k per month on Spotify. I don’t see a reason why you couldn’t grow your audience to 10% of that, you are one of the absolute best producers. $20k a month for 30 years is $7m, doesn’t seem too bad. And that doesn’t count selling merch, vinyl, or potentially touring. I don’t see a reason why you couldn’t grow to that audience

Burial + label is probably making around $20k-$25k/month on Spotify @sk33mask If you think I’m way off, happy to be corrected, just doing the back of napkin math. I think if you focused on growing your audience you could make a lot on Spotify in the long run

I love Arc and I hope they succeed, but....

Every software company that has tried this kind of hip & cool marketing has ultimately failed. Reminds me so much of Ello, the "cool" Facebook ().

Ello's unique character gave them instant buzz and they felt dated 6 months later.

It is really hard to create an enduring product with such strong flavor & character.

Enduring software companies are generally designed to feel like boring utilities. Facebook, Twitter, Slack, Superhuman: they sacrifice character so that the user can fill the vessel with their own soul!

If I pick up a phone, I want 100% of my focus to be on the person's voice, not on the "character" that the phone adds. Any unique character gets old and tiring if you use the product daily.

There have been so many flash-in-the-pan companies that have used this kind of design/marketing to get 12+ months of attention, but it burns out and feels dated fast.

Every time I get an email from Arc it now feels like: "damn I remember when I was so excited by this brand and now it's feeling old and dated"

I want to see The Browser Company succeed, so I cry:

You can build cool niche characterful products like Teenage Engineering, or you can build a boring utility that fundamentally changes the world, and that others fill with their own soul. Pick one! I love the character of companies like Teenage Engineering too, but they are comfortable not building a billion dollar company--they know who they are

Under Idealism, consciousness is generative, not interpretive

Under Idealism, Generative AI is a kind of “artificial consciousness”

I think this is a lot more significant than we currently realize

… What is idealism and why are smart people increasingly placing bets on it?

Physicalism = dominant worldview = reality is physical at the foundation

Idealism = reality is mind at the foundation

…

A mental model I've thought about in the age of 0% interest is "Performative Value". The ultimate result of zero interest is the cells in our organism (orgs, teams & individuals), learning how to do a Performative Dance to receive speculative dollars rather focusing on utility 🧵 twitter.com/i/web/status/1… 1. The damage of this is hard to understate. An generation of millennials has been trained in their early working years how to "perform" value rather than to deliver it. Performance looks like: "I did things the right way" (*applause!*) rather than "I achieved the right outcome"

There are many people in the world trained by video games to have a minute-by-minute value seeking loop running in the back of their brain at all times

There is a second group that doesn’t

These groups are very frustrated with eachother Group #1 is able to get a tonne of wealth and opportunity out of life, but they’ve eaten up the serendipitous alpha that Group #2 used to happen on by chance

Group #2 finds Group #1 cold and calculating

Group #1 find Group #2 lazy and ineffective, not doing their part

“Tech culture” is not one thing

Many successful people from the telecom industry struggled to make the leap to SaaS culture

Many successful people from the SaaS industry will struggle to make the leap to AI culture

A whole new set of values will emerge around this technology The biggest shift for many is going to be gaining comfort with non-determinism & ambiguity

The ”tinkering not training” mentality is at times closer to painting or songwriting than engineering
…

The “BigCo Delegator” startup failure pattern:

Masterful delegator from BigCo raises big first round $

Hires specialists for each seat in their checklist

*crowd cheers*

Outsources branding & web design to big agency

Launches to fanfare (wow! This is polished!)

And then… *thud*

There is zero PM-fit and a high burn team of specialists that is hard to manoeuvre

Delegator frantically tries to delegate PM-fit finding activities

They don’t know how to get in the trenches & chaos to do the gargantuan amount of shit work required

And eventually…

We're growing fast at @RallyLegal + spellbook.legal! Our customer base grew by 32% in November

We need more people ASAP to help us scale 🚀

Come build the ultimate AI-powered productivity tool for business legal work

Looking for:
- AEs
- Executive Assistant
- More TBA @RallyLegal Why do we care so much about streamlining business legal work?

Because legal transactions impose a "tax" on every one of us. Every hire, every sale, every purchase, every fundraise, every insurance policy, every incorporation of a new business.

...

As neural networks better at chess, they become more illegible: you can't explain why they make the moves they do in words

This will happen in the world at large as we approach singularity: you won't be able to explain in words why anyone does what they do Maybe you will for individuals. But for those that need to be competitive to survive, like businesses and investment funds, illegibility will become the new norm.

Legible stories will become the sign of a losing entity stuck in the past. While winners will look like RenTech

Yahoo lost because their homepage was ugly I remember, around 2002 when my teacher wrote GOOGLE.COM on the whiteboard in the computer lab

I had only used Yahoo before

As soon as I visited the site I never wanted to go back to Yahoo, and I was much too young to understand search quality

Is there an API that lets you seamlessly use different AI providers through a unified interface? Eg. GPT-3 models, Cohere, AWS, self-hosted models and everything else that is rapidly launching. Manifold for AI model providers would be cool @jevon

Seems like consolidation on a single provider is unlikely happen in high-performing AI products. Great AI products will be a mix of different models, different providers & self-hosted models

Humans have a magnetic attraction to structured pursuits, and each stage of life reveals that your last pursuits were not as dire as you thought they were Today this feeling strikes me at the gym, seeing myself in guys 10 years younger than me

I recognize the look: “Getting really fit will solve all my problems and give me superpowers”

Maintaining mental self-reliance can be lonely and gruelling. The world that lies beneath social narratives is so alien that you will, on some days, experience immense pressure to conform and question your own sanity. Here are some readings for those days ⬇️ (Ps. please add your own too)

@prestonjbyrne @Daan_Goossens Word is a great tool for writing unstructured English text. Stories, articles, cover letters, porno scripts. But it is not built for logic manipulation, testing/simulation, logic reuse. It does not have a real capacity to store data in a structured computer-interpretable format. @prestonjbyrne @Daan_Goossens Word files, or PDFs, once created, are inert. They are not stateful like real human agreements. We have to re-compile and execute legal programs in our brains, from scratch, ad nauseum.