iiinova — where your digital self keeps evolving

Matthias Plappert @matthiasplappert.com · 06/10/2026

📝 New blog post: Two types of AI agents Personal assistants and virtual employees will be two very different kinds of AI agents, and what separates them is who owns the work. matthiasplappert.com/blog/2026/two-…

030 0

Matthias Plappert @matthiasplappert.com · 14/07/2026

I agree but I also think it’s crazy and surprising that people don’t feel a moral obliteration to not be on a platform controlled by a virulent white supremacist. I can understand why people are like “propaganda won’t influence me” (even though it likely does) but the other seems like a no-brainer.

190 0

Matthias Plappert @matthiasplappert.com · 20/06/2026

I’d be most interested in the high level architecture and a concrete example of your typical way to build something. Show me how it actually works / looks. That’d be super interesting!

000 0

Reposted by Matthias Plappert

Johannes Schickling @schickling.dev · 20/06/2026

I’ve been building my own software factory over the past year. It again changed the way how I do software engineering fundamentally - similar delta compared to coding by hand vs using agents. I’m thinking of writing a blog post about it. What are your biggest questions?

8123 0

Matthias Plappert @matthiasplappert.com · 17/06/2026

I've started a new internal R&D project at @dfdxlabs.com: robotic manipulation research on real hardware! This is the first post in a series of research notes, covering the whole setup: robot arm, cameras, teleoperation, observability, data recording, and safety. dfdxlabs.com/research/202...

dfdxlabs.com

Building a desktop robotics research setup

The first note in an independent robotics research project, covering the physical setup, control stack, observability, data recording, and safety systems needed to run experiments on real hardware.

030 0

Matthias Plappert @matthiasplappert.com · 08/05/2026

My independent R&D company, @dfdxlabs.com, has a new website 🎉 dfdxlabs.com It explains what the company does: take on hard technical problems, figure out how to solve them, and build the first working prototype. I plan to use it as an outlet for the research I’m doing, so watch this space. 👀

dfdxlabs.com

dfdx labs

dfdx labs is an independent R&D company led by Matthias Plappert. We take on problems at the frontier, figure out how to solve them, and build the first working prototype.

000 0

Matthias Plappert @matthiasplappert.com · 05/05/2026

📝 New blog post: AI wants direct access to your data Why I switched my note taking app and built my own personal finance app: AI works best with direct, bidirectional access to your data. matthiasplappert.com/blog/2026/ai-w…

000 0

Matthias Plappert @matthiasplappert.com · 17/03/2026

📝 New blog post: You Have to Earn Your Calculator On the temptation of LLMs, and why sometimes you still need to do the work by hand. matthiasplappert.com/blog/2026/earn…

010 0

Matthias Plappert @matthiasplappert.com · 09/03/2026

📝 New blog post: Deepfakes for Code and the Asymmetric Internet AI hasn't just given us a new kind of spam—it has broken the economics of noise. The filter exists, but only if you can afford it. matthiasplappert.com/blog/2026/deep…

030 0

Matthias Plappert @matthiasplappert.com · 27/02/2026

📝 New blog post: Let's Talk About the Humanoid Robot in the Room General purpose humanoid robots are like self-driving, but actually much harder. matthiasplappert.com/blog/2026/huma…

000 0

Matthias Plappert @matthiasplappert.com · 12/02/2026

Is it just me or has Claude Code UI really gone downhill recently? Just got this beauty...

100 0

Matthias Plappert @matthiasplappert.com · 11/02/2026

I’ve decided to start blogging! The idea is to share short opinion pieces on topics in AI, machine learning and tech more broadly. Here’s my first post 🎉 matthiasplappert.com/blog/2026/la...

matthiasplappert.com

Embrace your Laziness in the Age of AI – Matthias Plappert

AI agents remove the natural friction that kept us from overbuilding. Now you have to be your own regularizer.

010 0

Matthias Plappert @matthiasplappert.com · 06/03/2025

o1 pro mode is pretty hilarious as well, especially step 2 (aka “optional wedge”)

010 0

Reposted by Matthias Plappert

Glen Berseth @glenberseth.bsky.social · 28/02/2025

How sure are we that the problem is only a software problem and not also a hardware limitation?

321 0

Matthias Plappert @matthiasplappert.com · 28/02/2025

This man is asking the right questions. It's very much a hardware problem still, especially wrt reliability, power efficiency, and cost. In that sense this humanoid robot hype is much worse than the self-driving cars one, because at least we knew how to build cars already.

020 0

Matthias Plappert @matthiasplappert.com · 28/02/2025

I'm actually very confused by the GPT-4.5 release

000 0

Matthias Plappert @matthiasplappert.com · 22/02/2025

We still don’t have broadly available self-driving cars but somehow humanoid robots for our homes are imminent. Yeah sure 🙄

130 0

Matthias Plappert @matthiasplappert.com · 06/02/2025

Also, it has an unusual amount of non-fungible experiences to offer that I think will remain (and will probably increase) in value because they cannot be automated.

100 0

Matthias Plappert @matthiasplappert.com · 06/02/2025

I’m actually quite bullish on Europe: I think it’s culturally well equipped to deal with the onset of increasing automation due to AI. We already have strong social security systems that can buffer the disruptions and peoples lives are less focused on work.

110 0

Matthias Plappert @matthiasplappert.com · 05/02/2025

This is cool, congrats! But the code benchmarks you use are non-standard and make it unnecessarily hard to compare these models to other one. Why do you not report HumanEval, Codeforces or SWE-bench performance?

000 0

Matthias Plappert @matthiasplappert.com · 31/01/2025

What made you a fan? Have heard multiple people switching recently but I’m not sure why

110 0

Matthias Plappert @matthiasplappert.com · 30/01/2025

So, even though Dario claims this to be expected, I think this changes the economics of the whole thing quite significantly and not in the favor of those who rely on massive outside investments (ie OpenAI and Anthropic). (6/6)

000 0

Matthias Plappert @matthiasplappert.com · 30/01/2025

So how valuable really is the second part? Surely you have to discount this now as well. (Side note: DeepSeek also intends to build AGI so the first point gets indirectly attacked as well because there’s more competition; Dario realizes this and wants more export control for this reason) (5/n)

100 0

Matthias Plappert @matthiasplappert.com · 30/01/2025

The second angle is what DeepSeek very directly attacked: they give you a recipe for how to train a frontier model for $5.5M and the weights themselves for free. They also became VERY popular with consumers VERY quickly: their app is still the most downloaded app on the App Store. (4/n)

100 0

Matthias Plappert @matthiasplappert.com · 30/01/2025

So you have two possible angles: we’ll build AGI and this will be very valuable and/or what we’re building on the way is also very valuable. The problem is that the first angle is very risky because you don’t know who will build AGI and when. So even though valuable, you need to discount it. (3/n)

100 0

Matthias Plappert @matthiasplappert.com · 30/01/2025

(This is a key difference from Google / Amazon / Microsoft: they have massive revenue and sizable balance sheets and can finance their AI projects this way.) But because they require massive, outside investments, investors reasonably expect a return on investment. (2/n)

100 0

Matthias Plappert @matthiasplappert.com · 30/01/2025

Essays like the one from Dario (darioamodei.com/on-deepseek-...) miss the point on why DeepSeek matters so much to investors. The key issue for OpenAI and Anthropic (and any other VC model company in this space) is that they absolutely require outside investments. (1/n)

darioamodei.com

Dario Amodei — On DeepSeek and Export Controls

On DeepSeek and Export Controls

110 0

Matthias Plappert @matthiasplappert.com · 29/01/2025

Yeah sure but the foreign phrasing is still misleading as it makes this sound like some foreign adversary somehow undermined the US govt, which is clearly not the case. I think the issue is that you can purchase that much influence regardless of whether or not the purchaser was born in the US

310 0

Matthias Plappert @matthiasplappert.com · 29/01/2025

I agree but the foreign part is strange; Musk is a US citizen and has been for a long time.

570 0

Matthias Plappert @matthiasplappert.com · 27/01/2025

I don’t know, NVIDIA revenue is mostly driven by a very small number of big spender customers and if this cools the risk appetite of those companies (because someone can come along and do what you’ve been doing but with orders of magnitude less capex), that’s a problem for NVIDIA.

fool.com

46% of Nvidia's Revenue Came From 4 Mystery Customers Last Quarter | The Motley Fool

Nvidia's incredible growth is increasingly reliant on just a handful of customers.

000 0

Matthias Plappert @matthiasplappert.com · 26/01/2025

OpenAI in particular still has an advantage of course because of its ChatGPT user base and brand. These open weight models are also still lacking behind in some features (lack of multi-modality and advanced voice mode, for example).

100 0

Matthias Plappert @matthiasplappert.com · 26/01/2025

It also increasingly looks to me like these models will get commoditized very rapidly. They also depreciate in value extremely quickly: Arguably the value of a frontier model is now only a few million dollars but it likely cost OpenAI et al many times more than that only a few months ago.

100 0

Matthias Plappert @matthiasplappert.com · 26/01/2025

I think these models likely caused a panic at places like OpenAI, Anthropic, Google and Meta: one of their main moats (having access to the biggest computers) seems weakened significantly. It is also proof that much less well-capitalized players can be competitive. Others will notice and compete.

100 0

Matthias Plappert @matthiasplappert.com · 26/01/2025

Finally got around to read the DeepSeek-v3 and r1 papers in detail and I’m very impressed. They keep things simple and pragmatic but go deep and optimize when worthwhile. I think access to compute remains key but they demonstrate how much more can be squeezed out when faced with constraints.

151 0

Matthias Plappert @matthiasplappert.com · 24/01/2025

It’s almost as if you build a toxic hell-hole of misinformation, conspiracy theories and crypto scams, people don’t want to be there. Who would’ve thought 🤷‍♂️

030 0

Reposted by Matthias Plappert

Casey Newton @caseynewton.bsky.social · 24/01/2025

Ladies and gentlemen: we did it

995241812656 0

Matthias Plappert @matthiasplappert.com · 20/01/2025

Oh I guess the KL is against the ref = base model not the previous iteration. I guess that makes more sense if you want to keep this model be interpretable / understandable to humans

120 0

Matthias Plappert @matthiasplappert.com · 20/01/2025

Bit weird that they have a KL penalty in there when the clipping part already ensures a trust region though (per the original PPO paper)

110 0

Matthias Plappert @matthiasplappert.com · 20/01/2025

(To be clear, this doesn’t change that the Trump coin thing is incredibly dystopian)

010 0

Matthias Plappert @matthiasplappert.com · 20/01/2025

Excellent (old but still very relevant) article by @molly.wiki on crypto market caps and the pitfalls with these numbers. So if your friends or the mainstream media say things like “Trump made $10B in a few days” you can be the person that goes “Well, actually …”

blog.mollywhite.net

Cryptocurrency 'market caps' and notional value

A naive equation works well when estimating the value of small quantities of popular crypto tokens, but it's also being used to estimate values of illiquid tokens and the market capitalization of all ...

110 0

Matthias Plappert @matthiasplappert.com · 16/01/2025

This 100%. Many stories from the early days at OpenAI (circa 2017) where it became painfully clear that the dude knows nothing about machine learning.

010 0

Matthias Plappert @matthiasplappert.com · 06/01/2025

Also, it seems clear that evolution doesn't "deeply understand intelligence", it just turned out to make us more likely to reproduce and thus it happened. So it seems pretty clear that there's an obvious counterfactual to this claim (unless you are a creationist)

020 0

Matthias Plappert @matthiasplappert.com · 06/01/2025

I also find this view extremely confusing since literally the main innovation in deep learning was to abandoned hand-crafted feature engineering (anyone remember SIFT?) and instead learn it from data. Turns out this approach works much better (but mostly sacrifices interpretability).

130 0

Matthias Plappert @matthiasplappert.com · 06/01/2025

Maybe this is what you meant but I think it’s likely executing the code and then doing majority voting over the produced output / return value (or some fuzzy similarity thereof). Much easier than doing it over the code itself

020 0

Matthias Plappert @matthiasplappert.com · 03/01/2025

That’d be quite an achievement! Do I understand correctly and you’re saying the only limiting factor is compute, not data availability anymore?

000 0

Matthias Plappert @matthiasplappert.com · 22/12/2024

Deleted (or I guess deactivated 🤷‍♂️) my Twitter / X account yesterday. I no longer wish to be associated with a platform that is this toxic. Quite happy that I finally pulled the plug.

050 0