Leonardo Scarfò

Building AI projects from Europe, also love the US. Notes from the gap between the two ecosystems. Startups, AI, and whatever I get wrong :)

[Leonardo Scarfò @scarfo_leo]

my way for a sustainable use of AI in entrepreneurship going forward, build slow fashion https://t.co/TUO54SrX4U

[Leonardo Scarfò @scarfo_leo]

@audrlo also “consumer hardware” nowadays is considered a $5k setup of DGX Spark or others

$5k in most countries is like a 3 month salary bro

[Leonardo Scarfò @scarfo_leo]

@olivieropinotti I’ve been realising that the semi-absence of bureaucracy is what makes the US the best place to build

not taxes, not investment opportunities, this is the single most good aspect of founding a company in the US

[Leonardo Scarfò @scarfo_leo]

I'm officially starting my first real vibe-coding project

will Dario come and baptize me or something?

[Leonardo Scarfò @scarfo_leo]

System justification theory, coined in 1994 by John Jost and Mahzarin Banaji. The theory describes people’s motivational tendency (often unconscious) to defend, legitimize, and rationalize the systems they depend on.

The same Opus 5 thing I was talking about this morning

people are perceiving more and more that LLMs are just spitting out more gibberish than ever in the last few weeks but, since they use them everyday, they are far more prone to make themselves guilty instead of the LLMs

[Leonardo Scarfò @scarfo_leo]

It’s because modern LLMs are yapping way more than before, do you remember how much GPT 4.1 or o3 yapped?

then Opus came and anything sounded clear and human, just concise and useful answers to hard questions

Anthropic started to mess this up with 4.7 on honestly, this morning I switched to 4.6 because 5 way being very confusing for no apparent reasons even on brainstorming chats

so it’s not always about us users, since we use LLMs every day subdole small behaviour changes like these are very difficult to grab

[Leonardo Scarfò @scarfo_leo]

It’s because modern LLMs are yapping way more than before, do you remember how much GPT 4.1 or o3 yapped?

then Opus came and anything sounded clear and human, just concise and useful answers to hard questions

Anthropic started to mess this up with 4.7 on honestly, this morning I switched to 4.6 because 5 way being very confusing for no apparent reasons even on brainstorming chats

so it’s not always about us users, since we use LLMs every day subdole small behaviour changes like these are very difficult to grab

[Leonardo Scarfò @scarfo_leo]

@mfranz_on Yes please Marco! I’m interested in servers cuz I know near to nothing about them

domestic ones seem even more interesting

[aether @bluewdrift]

With books unread, muscles untrained, and thousands of skills untouched - if you're bored, you're not even trying.

[Leonardo Scarfò @scarfo_leo]

@EXM7777 Hermes is way too goated bro, people just keep coming up with useful new scenarios with it

Hermes is the real AI we expected 6 years ago

[Leonardo Scarfò @scarfo_leo]

@rashmigb they seems like those gurus tryna sell you anything here, sophists ahh LLMs

switched to 4.6 just this morning

[Leonardo Scarfò @scarfo_leo]

@thorstenball That's what I experienced today with Opus 5

I freaked out cuz it kept saying nonsense and getting confused on the most useless things, yapping no stop 😫

I switched to 4.6, wasn't gonna resist a second more

[Leonardo Scarfò @scarfo_leo]

@mfranz_on @ivanfioravanti @italianbldrs what about Rome? would be the half way point to many people!

[Leonardo Scarfò @scarfo_leo]

I usually always listen to a good success story cuz they usually help, everything is possible except for the founder being born from an elephant and things like that so why not listening

you should pay attention tho when people start asking you to follow their same path or listen to their advice in a very "artificial" manner (often asking money for it), that's a clear signal of lying 95% of the time

[Leonardo Scarfò @scarfo_leo]

@jonnilundy I usually spend some time chatting with Claude about the theory behind what I'm building so that I can give precise and helpful instructions to Claude Code AND learn new things

that's how you should approach every problem, the bare minimum

[Leonardo Scarfò @scarfo_leo]

@mfranz_on @ivanfioravanti @italianbldrs what about an IRL meeting!

[Leonardo Scarfò @scarfo_leo]

@thejustinwelsh That’s what I unfortunately realised a while ago and fully embraced like 2 days ago

better late than never

[Leonardo Scarfò @scarfo_leo]

You always need a plan.
Always.
If you think you don’t need a long term clear plan for your 6 years goal you’re wrong as hell, and I just realised this.

moving on weekly plans won’t take you to accomplish your 6 years goal, and i’ll say this how many times I’m required to

if you don’t have a clear plan, and for clear I intend a plan which comprehends having a vision for:

> the type of person you’ll be when the goal is accomplished,

> what your surroundings will look like precisely when you’re finished (your home, the exact people around you, the number of kids you have, how much money you earn ecc)

you’ll never gonna get anything near where you would like to be, nowhere

always giving your best and “seeing where I end up to” doesn’t always guarantee the success you wanted, simply for the fact that you’re always gonna succeed

do you understand what I’m trying to say? If you think “let’s just see how things go” you for sure are gonna accomplish your goal

but
you’re
never
gonna
learn
anything
from
it

just because you ain’t got a reference.

for this exact same reason you need a very very VERY clear vision, you need to constantly check everyday at what point you are based on what you expected you would be

if you don’t have a long term plan for your long term goal you’re just gonna feel like you’re improving whereas you ain’t doing shit

if you’re just gonna acquire a bunch of useful skills you may end up doing nothing extraordinary with them

instead, think about your future as a long term stock investment: in order to accomplish that future I think I’ll need to have skill n 1, skill n 2, skill n 3.. AND BET YOUR NEXT 6 YEARS ON THIS VISION

energy can be spent in the most foolish ways, don’t waste it

keep your mind clear and sharp and, most importantly, keep your eyes on the road you want to follow

I just wanted to share this here because I myself have just realised this

atm my long term goal is to become one of the greatest entrepreneurs out here and help the max amount of people I possibly can, but without a super-duper clear plan I’ll get nowhere near this goal of mine

you don’t gotta feel ok with yourself just because you’re putting effort in, you also gotta be present and constantly check where that effort is going 24/7 to see if you’re really gonna make a difference

be present, be hungry, be optimistic, be foolish, just don’t go off the trails

[Leonardo Scarfò @scarfo_leo]

I've been talking with Opus 5 for nearly a hour now and I couldn't stand it anymore so I just switched to 4.6

5 feels way more exaggerated, I think it misses the versatility of Opus models: it can only do very hard work, simple brainstorming chats aren't possible anymore cuz it just get very confused real fast

In 3 different questions about the same topic in the same chat (with plenty of markdown files as knowledge sources) it managed to give me 3 different answers, he couldn't stand consistency apparently

4.6 is SOTA for human-level brainstorming and chatting, 5 is way too much tistic

[Leonardo Scarfò @scarfo_leo]

@ddsboston24 thanks man! appreciate you spending time to read my work 🥹

[Leonardo Scarfò @scarfo_leo]

my way for a sustainable use of AI in entrepreneurship going forward, build slow fashion https://t.co/TUO54SrX4U

[Leonardo Scarfò @scarfo_leo]

dedicate separate time to separate things

don't mix things up, give your brain a bit of love

[Leonardo Scarfò @scarfo_leo]

I've been learning the hard way that stable and fact checked knowledge is fundamental to LLMs

I preferred giving AI just 7-10 md files per time scared that it couldn't handle much more context, instead I'm trying to load up more and more now and honestly I don't know why I haven't done It before

[Leonardo Scarfò @scarfo_leo]

firstly you should build in public because you enjoy doing so, because your personality genuinely feels the need to help others and make others aware of the helpful things you're building

Secondly, and only secondly, you should take building in public as a serious distribution channel

meta ads when done right are far more valuable than building in public, and also require way less effort

people feel it when you're forcing yourself to be online