Founder | Ex‑Mindpeak | Solopreneur building in the AI space Deep Learning · Lifelong learner · 1 successful exit https://nitter.gq/t.co/0X1OU4Zrp6
Surfboard
Joined November 2008
- Tweets2.1K
- Following361
- Followers611
- Likes2.4K
Marc Päpper 🦆 retweeted
If your child becomes a reader, about 80% of the educational job is already done. That's our honest assessment, based on many of us working in education for over three decades. Everything else is secondary. Most parents think science education is important. Yes, it is. But if you can't read the biology textbook, you're not going to learn biology.
Reading is the meta-skill that enables all other skills. History requires reading. Science requires reading. Even math increasingly requires reading as it becomes more sophisticated. The child who reads voraciously will figure out everything else. The child who doesn't will struggle with everything.
Marc Päpper 🦆 retweeted
No analytics funnel for your SaaS?
You're blind to what actually makes your customers tick.
I spent a week figuring this stuff out, so you don't have to (yes, including how to debug it!) 🧵
First, a peek at my own funnel.
Warning: it’s brutal 😅:
Marc Päpper 🦆 retweeted
I've built an open-source version of the needed SuperWhisper for speech-to-text in Python.
It will run an AI model locally on your GPU (or your CPU if you wish, but it's slower).
At the press of a button (default: right Ctrl button) you can record your voice and it will print it as text.
This works in any app: your browser, your IDE, your terminal etc.
Github link and blog post below
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 so I barely even touch the keyboard. I ask for the dumbest things like "decrease the padding on the sidebar by half" because I'm too lazy to find it. I "Accept All" always, I don't read the diffs anymore. When I get error messages I just copy paste them in with no comment, usually that fixes it. The code grows beyond my usual comprehension, I'd have to really read through it for a while. Sometimes the LLMs can't fix a bug so I just work around it or ask for random changes until it goes away. It's not too bad for throwaway weekend projects, but still quite amusing. I'm building a project or webapp, but it's not really coding - I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.
Marc Päpper 🦆 retweeted
OpenAI's Deep Research is just a search+read+reasoning in a while-loop, right? unless i'm missing miss something, here is my replicate of it in nodejs, using gemini-flash and jina reader github.com/jina-ai/node-Deep…
Marc Päpper 🦆 retweeted
Excited to introduce R1-V!
We use RL with verifiable rewards to incentivize VLMs to learn general counting abilities.
2B model surpasses the 72B with only 100 training steps, costing less than $3.
The project will be fully open source.
Stay tuned!
github.com/Deep-Agent/R1-V
Marc Päpper 🦆 retweeted
🚨Join our team! We’re hiring a PhD student 🧠🎓 at @uni_wue & @UKW_Wuerzburg! Study mechanisms of decision-making in healthy people & Parkinson’s using new deep brain stimulation methods. German & English required. Apply by 20 Dec! 🌟🎄
Details: shorturl.at/IcNa0
Marc Päpper 🦆 retweeted
🤯 The level of sophistication of the XZ attack is very impressive! I tried to make sense of the analysis in a single page (which was quite complicated)!
I hope it helps to make sense of the information out there. Please treat the information "as is" while the analysis progresses! 🧐 #infosec #xz
Marc Päpper 🦆 retweeted
Party time! The SD3 paper made it to arxiv: arxiv.org/abs/2403.03206
Key takeaways:
- flow matching is very nice.
- back to work with @pess_r and a fantastic team ♥️
The paper is full of details on improved flow matching, scaling and engineering. Enjoy!
Marc Päpper 🦆 retweeted
We dug into a paper that’s been misinterpreted as saying GPT-4 has gotten worse. The paper shows behavior change, not capability decrease. And there's a problem with the evaluation—on 1 task, we think the authors mistook mimicry for reasoning.
w/ @sayashk
aisnakeoil.com/p/is-gpt-4-ge…
Marc Päpper 🦆 retweeted
I am hiring a PhD student in cognitive computational #neuroscience! 👩🏻🎓👨🏻🎓Come and help us discover how the human brain learns, decides and guides flexible behaviour in beautiful Würzburg 🧠🙌
Find out more & apply here: shorturl.at/twHL9
@bioDGPs_DGPA @FENSorg @SfNtweets
Marc Päpper 🦆 retweeted
"The truth is that your left brain has been interpreting reality for you your whole life, and if you are like most people, you have never understood the full implications of this...we mistake the story of who we think we are for who we truly are."
google.com/url?q=https://url…
Marc Päpper 🦆 retweeted
Congrats to github.com/mpaepper/llm_agen…, which has grown by 325 stars in the last 7 days and has reached 335 stars.
Thanks to the contributors: @mpaepper @ThingEngineer
ossinsight.io/analyze/mpaepp…
#Python
Marc Päpper 🦆 retweeted
Holy crap, the leap in new model v4 on @midjourney. This is thumbnails of a penguin in Venice, v3 on left, v4 on right --no specific style request. #AIart
Marc Päpper 🦆 retweeted
Implemented ByteDance's MagicMix StableDiffusion (arxiv.org/abs/2210.16056) paper.
You take an image and a new prompt and the method will tune the image in the direction of the prompt.
You can find the code here: github.com/mpaepper/stabledi…
Some examples:
Marc Päpper 🦆 retweeted
Lessons learned for the fastest path to success in building an #AI product 🚀.
We took these steps to cut down 80% of the time it takes to train a #machinelearning model from months to days.
8 steps:
Marc Päpper 🦆 retweeted
Great example of what's possible with the Dezgo API!
#stablediffusion #AiArt
