←
Dan Kaminsky128,692 posts
Dan Kaminsky avatar

Dan Kaminsky

@dakami

We can fix it. We have the technology. OK. We need to create the technology. Alright. The policy guys are mucking with the technology. Relax. WE'RE ON IT.

◷ Joined September 2007▧ 2,925 Photos and videos
Dan Kaminsky avatar
Dan Kaminsky@dakami·17 Jan 2018

I'm increasingly thinking that every functioning system has two forms: The abstraction that outsiders are led to believe, and the reality that insiders actually and carefully operate. You don't incrementally learn a system. You eventually unlearn its necessary lies.

Dan Kaminsky avatar
Dan Kaminsky@dakami·18 Apr 2021

Close. AI has plenty of doubt (most models can return probabilities for any prediction, if you configure them to). The problem is humans, not doubting the AI enough to notice when it doubts itself. It’s a tool, and it matters how you use it.

What separates humans from AI? It’s doubt on.ft.com/3dstHH9

Dan Kaminsky avatar
Dan Kaminsky@dakami·18 Apr 2021

hahahaha of course @ktemkin, QEMU TCG is basically chaining together small bits of compiled code for the host CPU anyway, no reason you can’t just intentionally generate that finite set at compile time. CLEVER

Here, Kate, made a script which generates a set of pre-compiled snippets of code (gadgets) with different semantics. This code will be compiled when the iOS app is built and will map to executable memory. Bypassing the need to create it at runtime. github.com/ktemkin/qemu/blob… (5/7)

Dan Kaminsky avatar
Dan Kaminsky@dakami·18 Apr 2021

it’s like all those corner cases in JavaScript but instead of being things you’d never actually do this is the most obvious way the API should work and the only thing worse than it being a shitshow is it’s v2 I am gonna fix this damn thing “Why has nobody tried this” “API sux”

Dan Kaminsky avatar
Dan Kaminsky@dakami·18 Apr 2021

developer.nvidia.com/blog/ex… “There are two ways to create CUDA graphs: construct the graph from scratch using graph APIs or use stream capture, which wraps a section of the application code and records the launched work from CUDA streams into a CUDA graph.” Uh, USEFUL

Dan Kaminsky avatar
Dan Kaminsky@dakami·18 Apr 2021

So I’m at this party in college, find a nice solved Rubik’s cube, and proceed to drunkenly unsolve the heck out of it. “Oh no! That took Jim hours!” Ok, so I’m smart. I can fix this, right? I’m terrible at Rubik’s Cubes, *sober*, which I was *not*. If only I had this.

Rubik’s Cube solved with #AugmentedReality and #ArtificialIntelligence {Video} #innovation #startups #tech #AI #AR

0:35
Dan Kaminsky avatar
Dan Kaminsky@dakami·18 Apr 2021

Rotating around some dimensions but not others in a time series representation ... that is a thing that we can experiment with artificially, and might already be doing. Interesting way to apply a sliding window, too, if there’s progressive dropout.

made me more curious on reading the abstract: «...was facilitated by a combination of ‘stable’ neurons, which maintained their selectivity over time, and ‘switching’ neurons, which inverted their selectivity over time. ....thereby protecting memories from sensory interference.»

Dan Kaminsky avatar
Dan Kaminsky@dakami·18 Apr 2021

I wonder if it’s annoying to neuroscientists constantly having to preface google searches with “NO NOT IN PYTORCH I’M TALKING ABOUT THE ACTUAL HUMAN BRAIN DAMNIT”

We are excited to announce the first In-Vitro 2D & 3D Neuronal Networks Summit on April 22 & 23! We prepared an exciting scientific program covering a wide range of topics, from iPSC-derived neurons to retinal and brain organoids. Register today: mxwbio.com/maxwell-digital-u…