Today, I'm excited to announce that we've raised a $23M Series A led by the OpenAI Startup Fund. One more small step towards building the world's first self-organizing workspace.
"You do not create dynamic equality just by raising the level of those at the bottom, but rather by making the rich rotate."
@nntaleb
's article on inequality is the most important and least well understood idea of the present socio-economic discourse.
The worst advice in startups is ship a half baked product and then iterate. The reality is products don’t change very much the moment you have actual users. Your initial launch is gravity.
Some personal news: Earlier this year, I decided to step away from Mem. It’s been an amazing 4.5 years since founding the company (and nearly 9 since
@kevinfmoody
and I first started talking about the idea!).
1/
@kevinfmoody
and I are thrilled to partner with
@davidu
and
@a16z
to accelerate the Mem vision. A thread about why we embarked on this journey to begin with.
The next month is going to be huge
@memdotai
. We're launching:
- the biggest update to the core experience, probably ever
- v1 of our API
- sneak previews of our mobile app to members of the community
What this means is rich parents will still be able to afford to send their kids to private classes, while poor kids who previously saw education as a means to break free from poverty will be left behind.
I'm opening a role for a prototype/full-stack engineer
@memdotai
to build novel capture and search experiences uniquely enabled by LLMs. You will report to and work directly with me, in-person in SF. We will push the bounds of what is possible.
DMs open.
If you’re thinking of starting a company one day, don’t make the mistake of doing the “big company tour of duty” for the purposes of “learning”. This is the number one regret I hear from friends who are trying to start companies now.
@Suhail
We’re working on this right now! Lightning fast input w/ lightweight organization. Released our private alpha recently - would love to have you try it if you’re game!
Claude 3.5 is so much better at codegen than all the other models. The fact this doesn't show up in a dramatic way in any of the eval leaderboards tells me people are measuring the wrong things.
text is the universal interface, always has been, read all about it! the compute fabric of the world is falling under the dominion of text once again
new blog post just dropped, guest authored
@scale_AI
16 months ago, we started with a simple question: "How might we make the world's information more useful?"
Tomorrow, we unveil act 0.
Mem is entering beta.
The problem with the information we capture today is that it too often becomes useless, lost, and forgotten. Beginning tomorrow, this will change.
Mem is entering beta.
Join the waitlist here:
The products we use today are built in modern day factories that we call software companies. Even the best products we use suck.
We enable anyone in the world to turn an idea into software, instantly.
Come join us.
Last month I got early access to Adaptive Computer. My reaction to the founder:
“This is magical”
They’re hiring a design engineer. Unique opportunity. Apply at .
Yes, we invested. :)
I've been waiting for this product release ever since
@kevinfmoody
and I started this company in 2019. We knew then what we're unveiling today:
Mem is a search company.
What early Silicon Valley intuitively understood is that great products are mostly built on vibes and taste, and that if you force someone to justify why something should be done too early, you get to sound smart at the expense of real value being created.
Biggest realization since moving from FANG to startup is a lot of my "product" skills were actually narrow set of shibboleths to make product thinking legible across wide group, deal w/ reviews, mitigate risks. Some of the best product thinking is illegible, difficult to defend.
What works:
- launch early to learn and then launch a new product based on those learnings (while retiring the old)
- wait until the product is truly ready
I'm gonna be honest — the whole "writing a lot of code doesn't make you a good engineer" narrative sounds reasonable but is simply not true. There are exceptions, yes, but the best engineers I've worked with are all prolific committers.
Knowledge work hasn't changed much in the past few decades. The most meaningful recent transformation was when the internet went mainstream — and people could start to access global wisdom at the tip of their fingers.
For as long as I can remember, I’ve wanted to build the software equivalent of a super-powered human being whose primary purpose is to help me remember less, create more, and dream bigger. 2023 is the year.
An LLM is inherently a compression engine (from information to a finite set of parameters). Compression is unavoidably lossy. In the long run, their true purpose is not recall (which require pinpoint precision), but reasoning.
1/ We don't build software that holds your attention — this is why we don't track "time spent on app" as a KPI. The world already has too many of these. Our goal is to help you move information out of Mem as quickly as you can move something into it.
The skills you learn at a large company are mostly people, politics, and process. At a startup, nothing matters except your ability to build something people want. There are no stakeholders to manage except for your customers (and they only care about your product).
Some predictions on AI over the next few years and beyond:
1/ almost every task that involves manually transforming and moving information from point A -> B will be done by AI by end of 2025. There will be a lot of corporate "restructuring".
Over on Hacker News, people are criticizing Marissa Mayer's contact management startup and wondering why she doesn't retire. (What a lame reaction to someone taking an entrepreneurial swing!)
One of the responses to these comments was excellent. I ❤️ free markets and capitalism.
But that's not all that we're announcing. We're also bringing Mem X out of private beta. Starting today, anyone can begin building their very own self-organizing workspace.
1/ Ok this is one of my coolest AI experiences to date...
I've been trying out the different "AI add-ons" lately, and a month ago upgraded to
@memdotai
's "Mem X" AI feature
It's been fascinating to see all the creative use cases that people are finding for Smart Write & Edit since we released it last week — like those in this video from
@UnmistakableCEO
Whenever a marketing video for a product focuses more on how it’s made than what it can do for you, it’s because they’ve built something that no one wants.
One of our most requested features of the past two months. We've always said that using Mem would allow you to do less and accomplish more. Starting with flows, now with tasks — we're delivering on this promise.
Free idea for someone building LLM dev ops. If an API call fails because one endpoint is down, automatically re-route to a different provider. Handle the rerouting + prompt transformation logic automatically.
The folks who are shitting on Zuck re Oculus should shut up. He is trying to invent something new. Just give him some time to try. Why tear down someone who is actually trying hard?
People who think that the “hype” around this new wave of AI (which includes generative AI) is even *remotely* close to the hype that surrounded crypto is in for a rude awakening. The demos today barely scratch the surface.
@jevakallio
Hey Jani, one of the founders here. I'll clarify because this is not what we said. Our mission is the opposite of what you're suggesting — we want to place your data back into your hands.
I grew up in an era where this was okay to say — simply because it's the truth. Bull market of the past decade has created a distorted view of what is *actually* required to do great things.
Now this is taboo, but no less true.
Life hack: rewire yourself to enjoy grinding.
MIT did this for me & most classmates. The school embrace the grind because there is no other way.
The harder you work early in your career, the faster you’ll compound. The payoff later in life can be astronomical (>10x).
Hint: if you’re having trouble with gpt4-o it’s probably because you overfit your prompt to make it work around the limitations of gpt4-turbo. Try to remove previous prompt hacks and you’ll see model performance improve dramatically.
Working wonders for us.
“Mem X is basically an artificial intelligence technology that augments your notes and knowledge…”
Thanks for the shoutout,
@FrancescoD_Ales
and glad you’re enjoying the early access!
We’re just getting started.
The goal is not to remove the existence of billionaires - it’s to remove the permanence of any individual billionaire. That is what
@nntaleb
means by ergodic equality. This is what
@SenSanders
and
@SenWarren
don’t understand.
In February, I left Google after seven (7!) years to join a tiny four person startup. I was drawn by the team
@memdotai
's mission to transform how we manage information. Here's why 👇
1/ We spend a lot of time creating and capturing information, but it often gets lost in folders or information silos.
Mem is building a world in which every person has access to the information they need when they need it.
With recent advances in AI, language models in particular, we sit at the precipice of a new era — one where machines cannot only improve our accessibility to knowledge, but actually help us think and reason.
Smart Write integrates existing information from your Mem knowledge base to produce text (on any topic) that is uniquely personalized to you. The more knowledge you have stored inside of Mem, the more powerful this becomes.
For those too who didn’t bother to read the actual article: this does not mean disincentivizing becoming rich (aka wealth taxes, socialism), it means removing the mechanisms by which rich people stay rich when they no longer contribute value.
This Weds evening is the 3rd installment of Chat8VC, our San Francisco event series celebrating builders in generative AI, featuring
@startupandrew
of
@Shortwave
and
@DennisHXu
of
@memdotai
.
A few spots remain - RSVP here:
With our latest update, you can now forward important emails you want to remember to Mem.
We'll automatically index and surface relevant information (like addresses, confirmation numbers, etc.) that you can search for.
At its core, all of knowledge work revolves around a set of common activities. We seek out information and inspiration, which we synthesize, and then store. Over time, we draw on the knowledge we've accumulated in order to generate new insights and ideas. Then, the cycle repeats.
I almost forgot how gatekeepy research/academic domains were until I started going deep into AI over the past year. Esoteric buzzwords for simple concepts (e.g. “program synthesis” - aka software that can generate code).
Always hated this about biology, still hate it.
We are building towards a Mem that can model the world in its entirety — to power experiences like these. Could not be more excited to unveil v0 of the Mem API to the world.
Incredible work from the entire team, especially
@MacroMackie
for leading the charge and...
The Mem knowledge graph powers the rest of your life. Imagine if you could ask it for "all the links related to project [x]" or for "restaurants that my friends recommend in Chicago".
I've loved getting to use from
@DennisHXu
and
@kevinfmoody
.
Beautiful note-taking, contextual task management, and truly insane speed and usability.
I wrote more about what they're building in the newest edition of In Flight.
As for what’s next.
@mikesoylu
and I started a new company in January. We are soon entering a world where all of the day-to-day products that we use will be malleable. Not just by the engineers building them, but by the people using them.
From early empirical testing, this new gpt-4 turbo model is severely undersold.
It’s below the embeddings updates on the blog but seems like a significant upgrade for code generation and instruction following relative to the 1106 preview model.
I told our designer today that I hate the new trendy illustration style with the massive hands so she made this to troll me 😂
Have to say, it definitely looks better when the
@memdotai
logo is involved...
cc
@priscillamok
Is there a better human out there than
@JuliaLipton
? I doubt it. If you're a founder and aren't already subscribed to her newsletter, do yourself a favor and subscribe
Each week she shares one awesome person looking for a new gig. Also, stuff like this 👇
I’ve been thinking about this quote from Steve Jobs a lot lately:
“Everything around you that you call life was made up by people that were no smarter than you. And you can change it, you can influence it… Once you learn that, you'll never be the same again”
Capture anything, anywhere, anytime. This was probably the most requested feature in our backlog and is now broadly available on the Mac Desktop app (Windows coming this summer!).
With all of the delightful, magical experiences being built in AI right now, novel mobile UX continues to be one of the most under-explored areas. What will a personalized knowledge assistant built on new-wave language models look like on iOS/Android?
Yes but*
The real end game is not disposable, but evolving, software. The interfaces and application logic will change as your needs do but the underlying data will follow each iteration.
This is the hardest technical/product challenge that no one is talking about.
I believe there’s going to be four main categories of software moving forward:
- Commercial software
- Boutique software
- Personal software
- Disposable software
Disposable is a sustainable way of executing a use case without the need to maintain it.
People often ask if we're building a "digital brain", or a "second brain". We're not. The future will hold no distinction between the digital software that we run and physical brain we were born with. They are one and the same. That's what we're building
@memdotai
14/ In a world where we can find publicly accessible knowledge in the blink of an instant, it's still nearly impossible to access information that is uniquely relevant to ourselves.
When you realize that you can build and adapt your own software, you start to realize that everything we use today - even the “good” products are horrible experiences.
Once you see it, you can’t unsee it.
For the longest time, science fiction has taught us to fear truly AI. The primary narratives depict conflict between man and machine.
But science fiction is exactly that — fiction. If the past year has been any indicator, the future will be more inspiring than it is dystopic.
Google employees explain why we haven’t seen ChatGPT like functionality in their products; the cost to serve an AI result is 10x to 100x as high as a regular web search today plus they’re too slow relative to how quick search results must be returned.
Unsurprising that the same group of people who support the inane policies that have led to San Francisco’s downward spiral are openly supportive of a literal terrorist organization who murdered hundreds of innocent civilians just 48 hrs ago.
Deranged.
What if you could hire someone whose only job was to process all of the information that you ever come across, organize it automatically, and resurface relevant knowledge at the exact right times? 👇
Here's a brief demo from
@isafulf
showcasing an early version of Mem X, and what it'll be capable of from the get-go.
We're rolling out our first wave of invites next week. Join the waitlist at
LLM inference providers, I am *begging* you, please increase the output token limit. There's no reason for a 100k+ context model to only allow 4000 tokens.