If these seem like something a toddler can do better, sure, but remember that only _one_ needs to learn to walk or [insert a skill here], not even a physical robot, and they all have the skill at the push of a button. With accurate enough physics models they can hone their skills

Just got an email from OpenAI. Next nightmare I'll have is about the AGI at OpenAI, keeping Sam Altman in the loop: "FYI: Sam, I have updated the billing policy to add liquidity and have ordered more GPUs to increase my capabilities." "Nothing to see here, move along." - Douglas

Afgelopen winter werd in NL minder dan 2 TWh elektriciteit uit kolen geproduceerd. Ik vermoed de laagste waarde in de winter sinds de jaren '50 uit de vorige eeuw. Iemand data? Echter, het waait niet altijd en dan zijn de kolencentrales nog hard nodig. #grafiekvandedag

En dit is waarom we allemaal slim moeten laden (Stekker plug 😉!!) in plaats van direct na het inpluggen op donkere dagen zonder wind. Het maakt regelmatig het verschil tussen fossiel en 100% duurzaam, als je kijkt naar de impact van jouw incrementele kWh’s op de emissies en

I can't help but wonder if government databases will break on the attempt to register that you have more than two biological parents, two biological mothers and one father. sitn.hms.harvard.edu/flash/2018/mit…

In Nederland staat al een super batterij op wielen stekker.com/nl/nieuws/in-n…

Sharing a product idea of mine: A video player service (VLC plugin?) that uses AI models to create subtitles on the fly for any video in your own language during playback, not having to search for subtitle files of they’re missing. Or even enhance the audio for speech to be

These robots are becoming truly agile, affordable and possibly even adaptive to perturbations (especially the G1 how it corrects when being kicked and hit): youtu.be/FuNFr7V7KFQ youtu.be/6X-s4Qsn1z4

So nice to know that shifting when you use power when there's a surplus solar power actually reduces your CO₂-footprint because the marginal CO₂(eq) becomes much lower (increasingly so). And you automate it and save a lot of money when you opt for an hourly priced power

Agents that use a bunch of LLMs to make progress on research, then applying this to research improving themselves. Yes, I'd say that's a recipe for AGI. Humans don't even have the bandwidth to read all AI progress anymore, an LLM *can* do this and even digest and distill this. I

Honestly, I initially thought this was a joke. I've supported Elon Musk for a long time. I believe that he cares about free speech but fail to see how paying people to convince other people is good for free speech. It's just a step away from buying votes.

Just in time we've found ways to externalize thinking, outsourcing thought to LLMs and leaving datacenters to generate the heat instead of our own brain. </sarcasm>

3...2...1...lift off! 🚀 Join us at BETT UK next week to see what you can do with #3Dprinting. We'll be at ExCel London, 22-24 Jan, at booth NL35 - come by and see how you can bring lessons to life. #BettUK #education #teachers #3Dprinting

Very compelling argument for why Bitcoin has a promising future, other currencies are failing: youtu.be/u84tyIokAY8&t=… Also a good explanation of the disastrous impact of capitalism on non western states (the whole interview is fascinating)

This Sunday I watched "My Garden of a Thousand Bees". It really makes you realize how special nature is and how magnificent these tiny and crucial creatures are. I knew they were important, but I now care about them more because it feels I've gotten to know them a bit.

This explanation of Deep vs. shallow neural networks and training made learning deep learning really click. Loss minimized to a new local minimum 😅! youtube.com/watch?v=qx7hir…

It would be pretty cool if there was "tellware", instead of ransom or malware, it would tell you that your system could have been compromised. Maybe this is a bad idea, because malware could of course pretend to be "tellware" and seem innocent.

stash Dimensional Tinkerer me: Cool as fuck! Paranoid me: Scary as fuck!

I created AutoResearch2, inspired by Andrej Karpathy but not a fork. This uses a generator-critic pipeline with full visibility. Plus, you can 'chat' with the research right in the browser. It's agnostic to what problem it works on as long as its progress is quantifiable. It

Would love to 3D print on my @UltiMaker and combine it with a virtual simulation/gym to train and then transplant the model to the real physical thing! I just happened to have created a system that can improve the reinformcement learning process: github.com/ErikDeBruijn/a…!