Fabien Devos
@Fabien_Devos
Followers
2K
Following
3K
Media
109
Statuses
3K
CTO and co-founder @vybe_build (YC X25) . Co-founder @wolfia_app (YC S22). Ex @wealthfront and @facebook. Co-creator of @hacked_app.
San Francisco
Joined March 2010
A year ago, I started working on a tool to build Web apps. But nowadays you can just chat with an AI and get the exact app you want. So I teamed up with @qhoang09 and we're super excited to announce @vybe_build, an AI to build internal apps, backed by @ycombinator !
5
1
34
We need at least 2 more breakthroughs: 1) Reasoning in latent space and 2) long term memory stored in weights. This could be the beginning of breakthrough 1
Impressive work.
0
0
1
Would sign 😅
I don't know what labs are doing to these poor LLMs during RL but they are mortally terrified of exceptions, in any infinitesimally likely case. Exceptions are a normal part of life and healthy dev process. Sign my LLM welfare petition for improved rewards in cases of exceptions.
0
0
0
ARC-AGI is arguably the most useful benchmark out there. Also one of the few that isn't even close to being saturated.
Absolute jokers - new Deepseek paper removed ARC-AGI benchmark because they "didn't know" what it was measuring... Allow me to spell it out, it (somewhat) distinguishes systems which are able to *synthesise* new understanding in response to novelty - which ... drumroll ... is
0
0
1
Intelligence isn't a collection of skills. It's the efficiency with which you acquire and deploy new skills. It's an efficiency ratio. And that's why benchmark scores can be very misleading about the actual intelligence of AI systems.
120
330
3K
Training on random strings -- interesting methodology here, and neat quantified findings
How much information do LLMs really memorize? Now we know, thanks to Paper by @AIatMeta, @GoogleDeepMind , @NVIDIAAIDev and Cornell Univ. This paper will be absolutely crucial for ongoing lawsuits between AI Companies and data creators/rights owners/copyright-holders.
7
45
270
I used Vybe to vibe code an internal app for Orange Collective in 5 minutes - here’s a screenshot. It’s a company tracker for the latest YC batch (using mock data, but I’m connecting it to our own internal data on Supabase, APIs, apps we use like Notion/GSuite, etc.)
Vybe (@vybe_build) is building Lovable for internal apps. Vybe comes with pre-built building blocks: integrations, SSO, and AI tools, so anyone in a company can build and launch internal apps. https://t.co/Ze4ZQLRayo Congrats on the launch, @qhoang09 and @fabien_devos!
2
3
13
Vybe (@vybe_build) is building Lovable for internal apps. Vybe comes with pre-built building blocks: integrations, SSO, and AI tools, so anyone in a company can build and launch internal apps. https://t.co/Ze4ZQLRayo Congrats on the launch, @qhoang09 and @fabien_devos!
10
11
163
(1/5) 💥🚀 Big launch day! @Fabien_Devos and I are excited to share what we've been cooking for the past few months! We're buiding "Lovable for internal apps". We come with pre-built building blocks: Integrations, SSO and AI tools so anyone can build / launch internal, demo:
13
13
67
@qhoang09 @vybe_build @ycombinator You can learn more and join the waitlist at https://t.co/ACUCGKOl26 - RT if you want to be bumped to the top :)
vybe.build
Enterprise-ready custom apps. Built by AI. Powered by your data. Ready in seconds.
1
0
2
Some people today are discouraging others from learning programming on the grounds AI will automate it. This advice will be seen as some of the worst career advice ever given. I disagree with the Turing Award and Nobel prize winner who wrote, “It is far more likely that the
deeplearning.ai
The Batch AI News and Insights: Some people today are discouraging others from learning programming on the grounds AI will automate it.
523
3K
12K
"A calculator app? Anyone could make that." Not true. A calculator should show you the result of the mathematical expression you entered. That's much, much harder than it sounds. What I'm about to tell you is the greatest calculator app development story ever told.
583
4K
34K
Vibe coding rocks. I've been doing it too and just seeing a project evolve at 10 times the speed it used to is incredible. Doesn't mean you should deploy the result in prod without checking what it does ofc but it's incredible to get started or build a prototype.
There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It's possible because the LLMs (e.g. Cursor Composer w Sonnet) are getting too good. Also I just talk to Composer with SuperWhisper
0
1
5
A potential mathematical explanation for polarisation, and with a hopeful conclusion as an added bonus! Of course, my prior is to want to believe this to be true so... 😅
Ever wondered why presenting more facts can sometimes *worsen* disagreements, even among rational people? 🤔 It turns out, Bayesian reasoning has some surprising answers - no cognitive biases needed! Let's explore this fascinating paradox quickly ☺️
0
2
1
The very minor news of the day (while oNs are whooshing past us): prompt engineering is dead. Good. 😅
to anyone wondering if the high ARC-AGI score is due to how we prompt the model: nah. I wrote down a prompt format that I thought looked clean and then we used it...that's the full story
0
0
0
A pattern recognition / system 1 model, called in a loop to guide a search process, was always our best shot at creating systems that can reason. It's the whole idea behind deep learning guided program synthesis -- use system 1 intuition to drive a system 2 search routine.
8
33
441
Dan is correct! Many arguments or experiments which purport to show that spacetime is quantum, construct strawmen theories which no one believes could be fundamental. 1/
This paper shows nothing. The masses are never in meaningful superposition: their state would decohere *long* before the gravitational field (semiclassical or otherwise) could measure them. They verbatim assume this doesn't happen, which is insane.
6
7
46
I know it's flogging a dead horse, but I thought this recurring debate would make a good launching point for a series of posts on why it's reasonable to question the quantum nature of gravity. So I started a Substack - follow along for more!
No it's not https://t.co/w5Mayztr7X
2
1
10
Basically brain surgery on Neural Nets 🤯 Also another example that shows that interpretability is possible. I'm tired of the "black box" rethoric, when the box is quite literally transparent. Sure it's not easy, but definitely possible.
so why don't we try direct neuron interventions? we can directly stop the model from interpreting these numbers as dates or biblical verses, by setting those neuron activations to 0. and that works! zeroing out september 11th attack neurons also works.
0
0
0