if you've been thinking about agents and gen interfaces,
you need to about hear about this fluid database.
LLMs will be rendering interfaces but they need proper data grounding otherwise they won’t work.
Let me explain — GitHub included
Feynman helped invent the atomic bomb through his physics brilliance. But he also lightened the mood at Los Alamos by pranking the military and showing their security was no match for his safecracking skills!
In the Oppenheimer movie, did you notice Feynman playing the bongos?
Feynman kept exposing flaws in the security systems. He'd break into filing cabinets full of nuclear secrets just to prove a point: the "top secret" info needed better protection.
When Feynman arrived at the secret NM facility in 1943, he turbocharged bomb development with his mathematical genius. He organized human "computers" to process calculations in parallel which speed up the operation.
But Feynman was more than a number cruncher!
He loved solving puzzles. So he spent his free time picking locks and cracking safes around Los Alamos.
These playful pranks turned out to be beneficial...
The new movie about J. Robert Oppenheimer and the A-bomb leaves out one of Los Alamos' most interesting characters: Richard Feynman.
What was his contribution?
@absurdtrader
Nemyslim si, ze AI bude v dohledne dobe plne nahrazovat kvalifikovane a kreativni profese, ale spis umozni jejich praci odvadet rychleji a kvalitneji.
👉 We've privately launched
@SteerCode
with our friends.
Yesterday, we hit a milestone of 100 DAU 🎉
I'm excited because the most common feedback we receive is: 'It's much better than what I've seen before on the codebase QA.'"
built low latency voice chat with interruption support for
@themind_ai
but I feel today it will become obsolete.
nevermind i just can't wait to experience the one model with audio in and audio out.
let’s see if openai solved the end-pointing problem (figure out when ai should
What if
@SteerAI
had a constantly expanding and improving skill/command library?
I believe this is the future of AI agents: personalized code snippets running in a sandbox orchestrated by LLMs.
I present to you - Cloud Environments for AI agents.
Watch how our
@e2b_dev
plugin allows ChatGPT to write GitHub crawler in Nodejs, install NPM dependencies, and print today's trending repos
@alex_valaitis
@MosaicML
I believe that
@MosaicML
is greatly benefiting from releases of more powerfull open-source LLMs.
MosaicML helps companies to fine-tune their own models which are usually trained on top of existing base models. And LLama2 is a great base 🔥
I'm calling it, we'll soon have a new way of programming!
Just exploring these new patterns with
@themind_ai
.
Dynamic & probabilisitic programming.
I LOVE IT!
FluidDB is a database that can update its schema based on the ingested data.
FluidDB saves unstructured and semi-structured data in a structured format.
most importantly, you can query it using traditional query languages as well as natural language.
okey, but how is this
Based on early user feedback (thanks!), we're completely rethinking the MIND experience.
Building an AI companion is uncharted territory, requiring lots of experimentation.
@NickADobos
Great experiment!
I think you can speed up the process by streaming a JSON that describes the app/design layout, which renders pre-made components. Maybe wrapping shadcn-ui.
i'm an adhd founder, and I want my mind to be organized and clear.
this is the reason why we're building
@themind_ai
, it'll be the first product powered by the Fluid Database.
if you've been thinking about agents and gen interfaces,
you need to about hear about this fluid database.
LLMs will be rendering interfaces but they need proper data grounding otherwise they won’t work.
Let me explain — GitHub included
New: Google Gemini + Vercel AI SDK chatbot template ✨
Book a flight, select your seat, check out, and more — all with server-driven, interactive Generative UI components.
Open-source and hackable!
MIND is too general, built for AI power users.
We're rethinking the whole experience to be driven by something specific, let's call it "agents".
MIND + AGENTS
have you wanted your own personal site with a blog?now is a great time to make one — new year, new you!
what I'll be sharing:
- What I Like / Find Interesting
- Short Blogs to help me find clarity in my thoughts
this one is on GitHub, you can fork it and make it your own.
THIS IS WILD, people really love
@themind_ai
voice chat.
We're closing the voice chat after 16.67min. Sorry!
Let's increase it to 30min and see if that's enough.
Fluid Database — database fully operated by an agent
the todo list is clear for the next days...
i want to make sure you can try it out with the CLI and that it's ready to be integrated into the
@themind_ai
app.
In hindsight, successful product decisions seem totally obvious.
But when you're starting out, nothing is certain or clear.
At least that's been my experience.
// THANK YOU
To everyone who’s supported us since the days of
@NFTScoring
, we thank you! You're the ones who made Singular possible.
This was just the beginning of our long-term mission... Let’s shape the future together.
11/11
Crazy fact that everyone deploying LLMs should know—GPT-4 is "smarter" at temperature=1 than temperature=0, even on deterministic tasks.
I honestly didn't believe this myself until I tried it, but shows up clearly on our evals. ht to
@eugeneyan
for the tip!
Our waitlist asks users what they expect from an AI companion.
The answers are incredibly diverse, but I believe there's a product that can eventually meet all their needs.
UX experiments on conversational AI
@themind_ai
The dialog should should support:
- tiktok style transcripts
- low latency replay
- walkie-talkie flow
- memory
I really need access to
@GroqInc
APIs for the internal dialogue simulation.
iOS implementation by
@ondrejkorol
Together with
@ondrejkorol
, we are learning, iterating, and pushing forward human + AI interaction.
If you have access to MIND, you know how the product feels different every week (sometimes not in the right direction, though).
First digital presence for our AI startup.
Invested 1.5MD into branding and a waitlist page.
Here's what we have so far - please don't look at the mobile view!
Introducing 100K Context Windows! We’ve expanded Claude’s context window to 100,000 tokens of text, corresponding to around 75K words. Submit hundreds of pages of materials for Claude to digest and analyze. Conversations with Claude can go on for hours or days.
MIND needs to be simpler, guiding users to their goals.
AI is a general tool, but most people don't know how to use it.
Splitting my time between SF & Prague has made this clear. The tech bubble naturally uses AI, but regular people struggle.
That's why goal-oriented AI
@immad
AI model of yourself, working in the background to make your life easier.
All your personal data would be streamlined into your own GPU cluster at home to fine-tune your own hyperpersonalized AI.
(bet on the future)