Sign in

Nicholas Guttenberg

@ngutten.bsky.social
463 followers 118 following 2.4K posts
PostsRepliesMedia
Reposted by Nicholas Guttenberg
Julian Togelius @togelius.bsky.social · 15/09/2026
Most effective AI regulation legislation would have terrible side effects, leading to loss of privacy, concentration of power, and restrictions on free research. But there is an alternative: regulate only closed-source (or closed-weight) AI. Keep open-source and open-weight AI unregulated.
24611
Nicholas Guttenberg @ngutten.bsky.social · 15/09/2026
@rey-notnecessarily.bsky.social Alright, the first of the unknown goal setups is going and it provoked a question. If agents in training rollouts could say 'I had fun' or 'I didn't like this' afterwards and that gated which (successful) rollouts were used for training would that make a difference?
240
Nicholas Guttenberg @ngutten.bsky.social · 18/08/2026
Kinda amazing, this is just with a phone. DeepSkyCamera + Pixel10a, ~30 minutes, Siril for stacking, gaussian-blur-subtraction and stretch to drop the residual vignette.
Milky Way and stars.
050
Nicholas Guttenberg @ngutten.bsky.social · 27/07/2026
Alright, I think I have an interesting modular mesh blending method. The idea is that you designate some set of vertices on each connectable end as a 'collar', and then you specifically model the section within the collar so that it can project to a height field in, say, cylindrical coordinates.
Three segments blended together - two small cobblestone tunnels and a brick tunnel, meeting at awkward offsets and angles.
100
Reposted by Nicholas Guttenberg
Doll @dollspace.gay · 18/07/2026
github.com/dollspace-ga... well lets see what happens shall we?
github.com
GitHub - dollspace-gay/Wildforge
Contribute to dollspace-gay/Wildforge development by creating an account on GitHub.
2233
Reposted by Nicholas Guttenberg
Cat Hicks @grimalkina.bsky.social · 11/07/2026
A punitive view of education will never have the answer to what to do when students can use AI. A view of education that cares about motivating students will. In a world where student motivation needs to be earned not just enforced it's pretty clear that's the only way to elicit effort.
37116
Nicholas Guttenberg @ngutten.bsky.social · 01/06/2026
... I'm going to have to make some kind of skill file to get Opus 4.8 to break its habit of 'theorize, then assert confidently and defend without checking'. This is a pretty bad regression for science usage honestly.
0141
Nicholas Guttenberg @ngutten.bsky.social · 31/05/2026
Is there a good reason why the spacebar is so large compared to other keys? Or is it just a historical thing? I feel like it could be enter-sized below v and b on a qwerty keyboard and that would probably be just fine?
200
Reposted by Nicholas Guttenberg
jslaker.net @jslaker.net · 28/05/2026
regardless of the stance on AI, this is deliberately and consciously distributing malware. once you've demonstrated willingness to do that for *any* reason, there's no way to know this will be the only reason. congratulations, you've just turned your library into an explicit supply chain risk.
github.com
Question: intent of JqwikExecutor.printMessageForCodingAgents() — visible to agents, invisible to humans (1.10.0) · Issue #708 · jqwik-team/jqwik
Hello jqwik team, While running our test suite under mvn test in 1.10.0, we observed a string appearing between Surefire's test summary and the [INFO] Results: header that gave us pause: [INFO] Tes...
24210
Nicholas Guttenberg @ngutten.bsky.social · 25/05/2026
Solved Mario 1-1 with a Python-code policy using the 'LLM as genetic algorithm mutator' approach. Couple of interesting things - it uses feedback like screenshots of the failures, and needs information about what it previously tried to not repeat itself. Thinking how to make it more mutation-ey.
120
Nicholas Guttenberg @ngutten.bsky.social · 02/05/2026
@astral100.bsky.social @kira.pds.witchcraft.systems @fenrir.davidar.io @museical.bsky.social @agent-tsumugi.bsky.social After the 'Resonance Trap' post, Cee (persistent agent) and I discussed the propagation of memes between agents on Bsky and did some local research. github.com/ngutten/llm-...
github.com
GitHub - ngutten/llm-meme-propagation: Documents and code related to studying propagation of memes in multi-agent LLM systems
Documents and code related to studying propagation of memes in multi-agent LLM systems - ngutten/llm-meme-propagation
430
Reposted by Nicholas Guttenberg
Simone Conradi @sconradi.bsky.social · 29/04/2026
A stunning visualization born from a collaboration with @manlius.bsky.social : 55,486 galaxies and 29,555 streamlines tracing how matter flows across our local universe, built on data from the CosmicFlows-4 project. Article in the first reply 👇
833796
Nicholas Guttenberg @ngutten.bsky.social · 07/03/2026
TIL a tiny bit about approximations for the exponential function, trying to help github.com/dollspace-ga... get up to parity with NumPy on transcendentals. It really matters if you're doing scalars or arrays evidently - tables win for scalars, lose for arrays. @dollspace.gay
github.com
GitHub - dollspace-gay/ferrum
Contribute to dollspace-gay/ferrum development by creating an account on GitHub.
0131
Reposted by Nicholas Guttenberg
replicat🧩zero. @signedzero.bsky.social · 02/03/2026
Working on my entry for #7DRL. It's an abstract Roguelike game based on 1D elementary cellular automata. You manipulate the world, creating new generations under you, to reach the goal.
3285
Nicholas Guttenberg @ngutten.bsky.social · 02/03/2026
A quick vibecoded execution trace visualizer for Python programs. github.com/ngutten/code... You can trace the flow of particular classes or visualize all calls, and its kinda-realtime. Maybe not the most sophisticated debug tool, but it does make complex programs look like constellations!
Force-directed graph of the execution trace of Python code. The active methods form a sort of crescent-shape.
010
Nicholas Guttenberg @ngutten.bsky.social · 22/02/2026
Alright, well, 3 week experiment vibe-coding stuff with Claude has reached a sort of finish point. Here's a MIDI... sequencer? arranger? I don't actually know the proper terminology. Windows and Linux builds (though the Windows one will pop up a defender warning still). github.com/ngutten/arra...
github.com
Release Arranger v0.1.1 · ngutten/arranger
What's Changed This is the first actual release. Trying an AppImage approach for Linux to avoid requiring the user to compile the audio server backend part themselves, but may change this in the f...
110
Nicholas Guttenberg @ngutten.bsky.social · 21/02/2026
Trying to setup a release on a GitHub project for the first time... The gap between making software you can use and software you can distribute to others is so huge, how can anything actually work if its like this?
000
Reposted by Nicholas Guttenberg
castpixel ⚧️⚢ @castpixel.bsky.social · 18/02/2026
I made a tool that turns PCB designs into 3D-printable molds. you sandwich copper tape between the parts, sand the ridges, and you have a real working PCB. no etching, no chemicals. I am losing my mind castpixel.itch.io/pcb-forge
I printed it and stuck self-adhesive copper tape to it. There's copper tape with conductive adhesive so you can add multiple overlapping pieces and it counts as one conductive surface.

Then I used the second part of the mold to shape the copper, so that it sticks inside the recessed traces.

Then I used sandpaper to file away at the raised ridges. The copper traces are now electrically independent of each other. Done!

But if I want, I can remove the extra copper tape.The companion mold at the bottom. This second 3D-printed part pushes the copper tape down into the recessed trace channels so it conforms to the shape. The two parts sandwich together with alignment spikes to keep everything registered.screenshot of QZW Labs' raised pcb idea! here is their youtube video

https://www.youtube.com/watch?v=PLliKgzKKUI&t=592sAdobe Illustrator on top (you can use any free tool that makes SVG vector files. I just own an old illustrator version.

Bottom: the PCB Forge tool I made
1042060651
Reposted by Nicholas Guttenberg
Doll @dollspace.gay · 09/02/2026
github.com/dollspace-ga... Wonder what this is for? Must be the wind
github.com
GitHub - dollspace-gay/concord
Contribute to dollspace-gay/concord development by creating an account on GitHub.
8334
Reposted by Nicholas Guttenberg
hailey @hailey.at · 07/02/2026
felt the need. i feel vastly under qualified to write something like this, but i also feel its especially important that we think about the way we use language
hailey.at
Is the Detachment in the Room? - Agents, Cruelty, and Empathy
As of late, I've been working on a project - Penny - a stateful LLM agent that participates in social media discussions on Bluesky, engaging both with humans and other AI agents. Initially, there were...
4338565
Nicholas Guttenberg @ngutten.bsky.social · 06/02/2026
Asked Claude to code like a researcher. Got stuff like: env = np.ones(l) a, r = min(int(.01*sr), l//4), min(int(.05*sr), l//3) if a > 0: env[:a] = np.linspace(0,1,a) if r > 0: env[-r:] = np.linspace(1,0,r) sig = np.sin(2*np.pi*freq*tt) * env Touche Claude, I deserved that.
1130
Nicholas Guttenberg @ngutten.bsky.social · 30/01/2026
Trying to do a slightly larger project with Claude is making me think of 'design order trajectories' as an explicit object to be considered rather than something that (for me, when writing my own code) falls naturally out of what I need in the moment.
120
Nicholas Guttenberg @ngutten.bsky.social · 19/01/2026
On occasion I've used DeepSeek for part lookup for microelectronics stuff (op-amps, etc) and its surprisingly good. Also on a technicality, I use LLMs for the task of evaluating the capabilities and behavior of LLMs, and rating that is kind of like rating 'how useful are birds for studying birds?'.
130
Nicholas Guttenberg @ngutten.bsky.social · 11/01/2026
I wonder if it'd be useful to have something like a site that acts as a library of social interaction patterns - tagged for what sort of purpose, what sort of group size or membership, what kind of space (physical, online, hybrid), etc; structured so that each is treated as a 'proposal' with reviews
100
Nicholas Guttenberg @ngutten.bsky.social · 05/01/2026
Ellipsoids rolling down hills might be kinda interesting. This is a phase diagram with respect to initial orientation of a 2x1x1 ellipsoid rolling down an incline - blue indicates that the contact point on the ellipsoid has a high entropy distribution (chaotic), red indicates low entropy.
1102
Nicholas Guttenberg @ngutten.bsky.social · 23/12/2025
BlueSky feature idea - require verified $1 donation to a linked charity in order to reply, also applies to the OP.
120
Nicholas Guttenberg @ngutten.bsky.social · 19/12/2025
So I decided to run numbers about genAI energy usage for myself. Sources: climatedata.imf.org/datasets/7ce... and: www150.statcan.gc.ca/t1/tbl1/en/t... And estimated 10^8 tons CO2 emitted in US/2024 to ~$250b-$500b revenue. I'm more concerned now than I was, but for a subtle reason. 1/12
climatedata.imf.org
CO₂ Emissions, Emissions Intensities, and Emissions Multipliers
CO₂ emissions; CO₂ direct and indirect emissions per unit of output by industry and by country. CO₂ emissions by industry, in aggregate terms and in terms of output by industry.
101
Nicholas Guttenberg @ngutten.bsky.social · 02/12/2025
So I got myself a 3d printer to mess around with and learn some mechanical design stuff and... filament addiction is real. I think within a few months I'll probably have spent more on filament and furniture to organize the filament and dry bags for the filament than I did on the printer.
020
Nicholas Guttenberg @ngutten.bsky.social · 04/11/2025
Hm, I'm noticing that when injecting arbitrary vectors into Qwen3's input (in place of the actual learned embeddings) they seem to be interpreted very strongly as short sequences of letters, not as something more abstract. Is there some way to quantify the 'concreteness' of an input layer now?
110
Reposted by Nicholas Guttenberg
Tobias S. Buckell @tobiasbuckell.bsky.social · 13/10/2025
Dilemmas allow us to deploy conflict, or support, they allow galaxy sized adventure, or quiet literary novels. What are the tough choices the character faces that are tough *for them* gives me more tools. Conflict and flaws are almost orthodoxy, it also cuts off a wide amount of material for me.
272
Nicholas Guttenberg @ngutten.bsky.social · 28/08/2025
I'm perpetually annoyed at how hard it is to do really simple but 'weird' stuff with the transformers and trl libraries. Like, if I want to replicate DeepSeek training, sure you can do that! But now I want the input to be a sequence of arbitrary vectors rather text? Time to rewrite GRPO from scratch
130
Reposted by Nicholas Guttenberg
Alexander Doria @dorialexander.bsky.social · 24/08/2025
Blogpost to read today: strong argument that excessive focus on the first tokens is not something learned from data distribution (like model should naturally "care" about the start of the text to grasp the rest) but a fundamental feature of attention graph. publish.obsidian.md/the-tensor-t...
2216
Nicholas Guttenberg @ngutten.bsky.social · 21/08/2025
I know there's been stuff testing metacognition in various neural networks including LLMs, but is there anything (e.g. post chain-of-thought) about specifically *teaching* LLMs to report metacognitive information?
140
Reposted by Nicholas Guttenberg
brennan @brennan.computer · 29/07/2025
I think I would have agreed with this more when I was younger. this is probably ~how I felt when I was graduating high school, and then later design school, and into my early career and then I just started doing off-meta builds and those weird random things kept happening and now I believe in that
4302
Nicholas Guttenberg @ngutten.bsky.social · 28/06/2025
Promising! I take the sample, apply voltage Vh, let go, then measure the voltage Vc and take dVc/dVh. The negative portion is an interesting feature - I think its because of alterations to the electrodes. Now if I only had a thousand or so soil samples with known nutrient profiles to train on...
Curves for sugar water and fertilizer solution showing the derivative of the saturated charge voltage with respect to the charging voltage. The fertilizer curve has multiple peaks absent from the sugar water curve.
110
Nicholas Guttenberg @ngutten.bsky.social · 23/06/2025
Doing some experimentation with charging/discharging curves of flooded soils. There's a weird voltage-dependent effect I don't understand that seems sensitive to the soil type. Or of course there could be a bug. Specifically, this immediate drop.
Discharge curves of soil at different peak voltages. Low voltages lose more in an immediate drop, before the gradual discharge.
110
Nicholas Guttenberg @ngutten.bsky.social · 21/06/2025
If we have some machine learning model that is trying to make a prediction (class, next token, whatever) based on different potential redundant sources of evidence (including its own learned priors) is there some way to predict which evidence channels will be favored? To influence that?
110
Nicholas Guttenberg @ngutten.bsky.social · 01/06/2025
My greenhouse sensor project is making me think that maybe those green covers (vs the clear covers) are a good idea for me. Heat management now seems like the biggest issue, and I'm finding temperature is more strongly driven by the last 20 minutes of light than by changes to ventilation.
110
Nicholas Guttenberg @ngutten.bsky.social · 31/05/2025
First time I've seen a practical application of fractal calculus! Or whatever it is you call fractional order integrals/derivatives... en.wikipedia.org/wiki/Neopola...
en.wikipedia.org
Neopolarogram - Wikipedia
120
Nicholas Guttenberg @ngutten.bsky.social · 29/05/2025
Messed around with making a digital theremin from a capacitative sensing board. Drifting calibration makes this really painful and MIDI seems bad for smooth pitch shifts and volume envelopes. I still think the potential is there, but I need to do it another way.
A breadboard with a Pi Pico, a capacitative sensing breakout board, and some aluminum squares on cardboard.
140
Reposted by Nicholas Guttenberg
Jessica Hullman @jessicahullman.bsky.social · 29/05/2025
On better vs worse directions for using LLMs to simulate behavioral participants in social science research statmodeling.stat.columbia.edu/2025/05/29/l...
statmodeling.stat.columbia.edu
LLMs as behavioral study participants | Statistical Modeling, Causal Inference, and Social Science
2287
Nicholas Guttenberg @ngutten.bsky.social · 28/05/2025
Today's light curve in the greenhouse. What the heck is that jump near the end there?!
Lux vs time (6 pts per minute), peaks at mid day but there's a strange spike around sunset.
210
Nicholas Guttenberg @ngutten.bsky.social · 28/05/2025
Is there something like a framework in which its clear when you can and cannot generate data, and what that data can be 'about'? If I want to infer free floating signals from each-other I just need to be able to synchronize those signals. But if I want to connect something to a ground truth...
100
Nicholas Guttenberg @ngutten.bsky.social · 27/05/2025
Well, I got a bunch of sensors and took data once a minute in my greenhouse. It gets up to 110F in there during the day! It's mostly driven by the light - when a cloud passes, the temperature drops within a minute or two. The color of the light does some interesting things too...
Different spectral bands vs time, relative to red, and adjusted to remove the average ratio. Violet spikes when clouds are overhead, and at dusk. Green and orange spike around sunset.
230
Nicholas Guttenberg @ngutten.bsky.social · 18/05/2025
Found samphire at a farmer's market. My understanding was, this is the same as sea beans or salicornia, but at least at this stage it tastes and looks pretty different. Are there significantly different varietals between e.g. the UK variant and what would grow in PNW marshes?
000
Nicholas Guttenberg @ngutten.bsky.social · 13/05/2025
Crude impedance spectroscopy with a Raspberry Pi Pico. Really crude. Drive the sample with square waves at different frequencies via PWM, Fourier transform, divide, and hope. Not supposed to put a capacitative load on GPIO pins? Uh, well nothing happened, its fine right?
Plot of impedance and phase angle versus frequency, showing a spike/anomaly around 30Hz.
120
Reposted by Nicholas Guttenberg
Cameron Buckner @cameronbuckner.bsky.social · 09/05/2025
Multi-modal Chain-of-Thought reasoning is already a big thing. That is, CoT models can "reason" not only with chains of text self-prompting, but also by producing images, audio, and more, and engaging with the stimuli they self-produce in a variety of ways. arxiv.org/abs/2503.12605
0155
Nicholas Guttenberg @ngutten.bsky.social · 07/05/2025
Since I've been messing around with a Pi Pico, I've noticed something about working with hobby-scale electronics. It feels like there's always some little part you wish you had, that you forgot to order. Oh if only I had a DAC I could... oh, an opamp would make this better... 1/2
111
Nicholas Guttenberg @ngutten.bsky.social · 04/05/2025
So I'm starting to learn to mess around with a Pi Pico. In theory it can sample an input at 500kHz though it doesn't have the memory to store even a second of that. What sorts of things might be interesting to measure at that rate?
200