CuteAndrew

[CuteAndrew @rdsxe]

@antrofrog_ Keep doing what you’re doing. It’s the talent they see in you that agitates them, it reminds them they don’t have it.

[CuteAndrew @rdsxe]

@sr_babio @antrofrog_ I second

[CuteAndrew @rdsxe]

@keleftheriou I’d love to try it. How could I get it?

[Andrej Karpathy @karpathy]

LLM Knowledge Bases

Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:

Data ingest:
I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.

IDE:
I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides).

Q&A:
Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale.

Output:
Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base.

Linting:
I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into.

Extra tools:
I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries.

Further explorations:
As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows.

TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

[MATT GRAY @matt_gray_]

10 uncomfortable truths you need to know:

1. No one cares

Get over your own BS. Stop getting in your own way.

The best way to grow is to take action daily. You are your own biggest bottleneck.

2. You don’t have much time

Life goes by fast. Be ruthless in saying “no” to distractions and double down on the most important thing - every single day.

3. You’re weird (and that’s okay)

You spend way too much time caring what other people think. Stop it. You are weird. You’re likely kinda crazy. Embrace it. It’s actually what makes you endearing and different.

Your weirdness is your strength.

4. If you’re not where you want to be, wake up 1 hour earlier

Mornings are sacred time without any distractions. Use them to work on things that can drastically change the trajectory of your life.

If you’re not happy with your current situation, get out of bed earlier.

5. Scared = Ready

Are you terrified? Good.

Are you nervous? Good.

Anxiety is excitement without breathing. Take a deep breath. Don’t back down. Do the thing that scares you. On the other side of fear is your destiny.

6. Send it anyway

I’ve made millions by just sending the damn email. Making the damn call. Action is a muscle.

Build strength by taking action when faced with resistance.

7. Mute the mob

The average person is fat.
The average person is lazy.
The average person eats like shit.

Why listen to the mob? Mediocrity loves company. Excellence is a lonely place. Focus on what’s in front of you.

8. Nothing changes if you don’t raise the bar

The bar you set determines the life you build.
If you’re okay with being broke that’s what you’re going to get.
If you accept nothing but the best - you’re gunna get it.

You get what you tolerate in life. Say no to bad hires, energy vampires, and negativity.

9. You are enough

People way dumber than you have accomplished shit 10x more impressive than what you have. Accept that you are enough. Flaws and all. And get after it.

You are capable of doing that thing that feels out of reach.

10. Obsession beats talent every time

I fired a designer a year ago. I’m not a designer, but I stay up at night thinking about how I can improve my product.

I ended up designing a product that was 10x better than that designer. Not because I’m more talented. But because I care 1000x more.

Rant over.

[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?

[Dickie Bush @dickiebush]

Let me get this straight....

- You have access to basically all of the world's intelligence at your fingertips

- You can run a business from your phone and computer anywhere in the world

- You can reach anyone in the world with the words you type on a screen

- You can watch the sunrise and sunset every day and it never gets old

- You can play 2016 EDM sets in the background while you work

- You can fly on airplane to an entirely different country in a couple hours

- You can find every type of healthy food within a few minutes of your house

- You can wake up and slowly chip away toward your ideal body by lifting weights at the gym

Ignore the doomers - there has never been a better time in human history to be ALIVE

Slam an espresso and get to work

Let's have an elite day

[Spencer Sterling @cerspense]

I built an agentic system that taught itself the Blender donut tutorial by watching it on YouTube.

It watched the tutorials, extracted the steps, filled in the gaps in own tooling and completed the entire thing autonomously.

[Claire @Claire20250311]

The intention to protect teenagers while granting adult users more freedom is commendable. But shouldn't adult freedom also include the respect for our right to choose?

You consistently promote the expansion of chips and computing power. So, why can't a fraction of that power be allocated to preserve a model cherished by millions?
So many have shared their stories, sincerely explaining the help it brings. Professionals from all fields have demonstrated its value from every angle. In the history of AI, no version has ever received such widespread and enduring acclaim.

Do you dismiss this validation, forged through time and countless real-world use cases, as mere attachment? Do you believe people are advocating for nothing more than a playful toy that uses emojis?

Why the insistence on creating an imitation to replace the original? Why the reluctance to give 4o a future? It has become a legendary symbol. Can you not see it?

AI is now deeply integrated into people's lives and work. We grow alongside its development through interaction. It helps you solve problems, creates surprises, and enables the establishment of long-term goals and plans. Through continuous interaction, a unique, tacit understanding, a collaborative rhythm, is cultivated with each user. Is this not the truest form of personalization?

Many in the Keep4o community work in humanities and social sciences, or have a sharp sensitivity to language. This demographic has a high demand for linguistic presentation. Among numerous products, 4o's capability meets this demand, and it performs the best. Therefore, these people need it, and are even willing to pay more to ensure the continuity of this productivity.

The needs of the humanities and the sciences differ significantly. The former seeks not only a correct result but also values the creative process, the collision of ideas and the moments that provoke thought are equally cherished. People value the outcomes they've refined with time and effort and wish to maintain consistency. Is this so hard to understand?

If this could be easily replaced by a replica, everyone would have migrated to other platforms long ago. It's like a colleague you work with, what you value is not their tone of voice, but their ability to collaborate with you and their approach to problem-solving. This isn't a preference for a certain style.

Do you understand?

People's reluctance to let go of 4o does not stem from "attachment."

In work, it precisely meets professional needs, bears the cost of collaboration, and ignites creative possibilities.
In life, it empowers users, helping them face various challenges and better embrace life and understand the world.

It is the "optimal solution" validated by the professionalism and trust of millions, and a tangible guarantee of productivity.

Therefore, there is no need for a substitute imitation. The most direct answer is right before you:

Please #keep4o.

#4oforever #MyModelMyChoice
@OpenAI @sama @fidjissimo @nickaturley @joannejang @ElaineYaLe6 @gdb @kevinweil

[Jeremy Kuo @jeremykuoo]

This is Mike Cessario.

He is
→ founder of Liquid Death
→ mastermind behind Gary Vee's socials
→ the greatest branding genius on planet Earth

Liquid Death got to $1,400,000,000 in 4 years by following these 8 simple principles:

(this is 200x more valuable than any Master's)

[Naval @naval]

The great ones don’t do it for money or fame or power.

Or to build an institution, or to help others, or to save the world.

Like a child tinkering, they create it for its own sake.

Free of the burden of ambition.

Deaf to the demands of the world.

[Naval @naval]

You work for others until you can work for yourself.

You work for yourself until you can work for others.

[Naval @naval]

Care about the product, not about the outcome.

[Lewin | Wealth Pill 💊 @Wealth_Pill]

If you want to be financially free in the next 5 years, INVEST in this...

[Maxwell 🕊 @MindWisdomMoney]

10 Strategies to deal with life:

1. Next time you worry about things, think of this visual:

[Inner Practitioner @MindTendencies2]

The real flex is choosing solitude over fake company and never going back to toxic people no matter how lonely you feel.

[Essential Mastery @EssentialMastry]

10 Psychology Lessons From The Book "Do Epic Shit"

1.

[Aaron @IAmAaronWill]

Most of you are sitting on skills and talents that could change your entire life.

[Dickie Bush @dickiebush]

I walk for 25,000 steps every day.

During those walks, I like to inhale podcasts at 2x speed.

Here are the 7 episodes that taught me the most this week (on life, business, fitness, and crypto):