PxldLtd

I mean we're all probably on some list somewhere for any reason at this point unless you practice some serious opsec

I don't use twitter but it feels like a platform for bothering terminally-online celebrities

My salary literally doubled just working remote/hybrid for London companies around COVID. I much prefer the lifestyle now I work for a smaller company in Canterbury and get paid less for the role. Turns out all that extra travel and overtime ended up being materially worse for my quality of life. And my job is way more fun nowadays.

This is already done, there's hashed indexes of known CSAM and you can apply vector steering to the latent space to shift detections to trigger on CSAM material. You can also use two detectors for age estimation and a separate one for nudity and use them in combination.

I mean, none of the videos show it completing a task successfully as I see it!

1. Doesn't look like it's stirring well enough to prevent burning

2. Seems to just be picking up vertical cans and grabs the napkin unintentially, never see it pick up any of the more interesting shapes/orientations.

3. Doesn't shut the fridge door

4. Teleports to things in it's hand, never see it actually place something in the rack.

5. Not a fold in site, vague wafting of a shirt around

6. Poured a few grains of cereal, not sure it can tell if it's poured the right amount of anything. Doesn't look like a useful result there.

Honestly, it could probably close the fridge. Why they can't show it completing a task in any of their examples is probably not a good sign for the skills situation. It's clear the goal is crowdsourcing the software and providing a platform with no idea whether it can actually do any of these tasks reliably to be useful.

I gather this is a vague snarky comment loosely implying that vegans don't get enough B12. If so, it's a profoundly ignorant point that shows you don't have any knowledge of a typical vegan diet nowadays.

We used to charge innocent men for room and board in prison after being released here in the UK not long ago. They only changed this in 2023 after uproar in a recent case of an innocent man being freed.

It's more a comment about the feature detection I think; all image, video and audio input contribute to the same weights/activations that can produce image, video and audio output.

This is short-sighted IMO, I bet money that once the dust settles with the current LLM-boom that Google will still be around and pioneering in AI across the board. They aren't locked in to Nvidia, they have world class research, they aren't bogged down with VC returns or upcoming IPOs.

"If they did it right" is a bearing a lot of weight there. The window of "getting it right" has shrunk over the last 40 years and it's delusional to argue otherwise. People like myself are comparing themselves to their parents at the same age. My dad had me at my exact age now, had bought a house, could raise a family on his singular wage. That is non-existent now in my part of the country.

There's a project that's going well from NASA for this. Still a moonshot but they've progressed through the early stages well so far.

https://www.nasa.gov/general/direct-multipixel-imaging-and-s...

Hmm, if you read his previous blog posts pre-ai the tone is not much different. I hate the idea that this dude may just sound like AI at this point by accident.

I personally think the answer should be mandating user controls of said algorithms. If you have a large social media site that suggests results or provides a feed then you must allow users to configure said feed.

The default should be simple, something like time-ordered of followed only with opt in for recommendations. Ideally I'd love to be able to swap algorithms for something open source but setting a standard there rather than just mandating a level of control seems a bigger hurdle.

It seems entirely anti-consumer that I am just the behest of whatever Google or Meta decide is most profitable to show me instead of finding important news or entertainment in a way that I want.

This feels a bit out of touch. These policies have a lot of public support here in the UK. All of our parent friends are lauding it despite my complaints.

Yeah, I don't think I'm ever getting that €600 back from Hertz Calais. Still bitter about it.

Here's a great Instructable on building one yourself, much better explanations in there.

https://www.instructables.com/Designing-and-Building-an-Axia...

The whole thing is a cobbled together bodge over SharePoint as a backend. I wouldn't ever trust my company data with that dogwater product.

Back when I had to work with it I found a bug that could cause folders to become un-synced without you realising, meaning changes would not be tracked and cause merge-conflicts when it was fixed.

Managed to use our Gold partner tickets to raise the issue with the product team, they flat out refused to fix the issue even knowing it was a bug. This was back in 2020 or so, I wonder if they ever fixed that bug. It's pretty simple to reproduce:

1. - Sync a nested subfolder from Sharepoint

2. - Sync the parent folder

3. - Note that the folder synced in 1. is not longer being tracked (no checkmark)

4. - Normal users will now go to folder 1. by default and have no idea none of their changes are no longer being tracked now that it's being synced within folder 2.

It's an interesting point but I fear Go's FFI is going to kneecap its ability to be widely adopted unless that story improves significantly. It's a lovely language if your interop with other languages is minimal.

> The rest was glue

Oh how often I keep saying that these days... "All the parts are there! Why hasn't anyone piped this into that?"

You can reveal it later when you come up with a new mechanism and out all the fake images. Basically the first layer is first defence, second layer is cleanup when releasing a new mechanism. That way your generated images will always be identifiable eventually.

I would be the safest citizen, free from experiencing crime and violence if I'm imprisoned in my house for life.

This is what stopped me from picking up Podman more, all our devs use Docker and have been writing compose files for years now. When the response at the time was "you're using Podman wrong, Quadlets are the hot stuff now" it just felt like too big a risk and commitment to jump to at the time. Have things settled more? Getting away from Docker is a bigger priority nowadays for us.

While modern LLMs are a far cry from biological synapses, I do find it fascinating that if you take the highly reciprocal data of a biological connectome and unroll it into a DAG, you suddenly see motifs popping up that look similar to what we find in AI. I found this both looking at temporal unrolling of RNNs or mapping layer activation weights of a Transformer. Totally agree though, the current LLM architecture itself is driven by the need to shove all of this nicely into parallelized compute hardware.

Yeah actually I haven't been dissapointed with the quality of the new videos so far, I will hold my breath though as venture capitals priorities aren't always aligned long-term.

It's been bought up by venture capital and it's rubbed a lot of fans the wrong way with the changes by the new management.

You can hard-code them. Just write a setup script that imports your desired keys from a .env file or something:

``` docker exec $CONTAINER $BIN key import --yes "$S3_ACCESS_KEY_ID" "$S3_SECRET_ACCESS_KEY" ```

I very much disagree, there's just too many engineering hurdles for us to surmount for this to be a reasonable solution. When you actually break down the physics, the scale works against you.

You can't have "one giant computer" when the speed of light is a 16-minute ping time from one side of the swarm to the other. Also cooling. Space is a vacuum, you can't just use convection. The inner layers would melt before they could radiate it away.

Even maintenance and power distribution, you're talking about trillions of nodes that need active course correction to avoid a chain-reaction of collisions.

There's so many reasons this is not feasible and more of a whimsical thought experiment. I've barely even touched on most of the issues.

I wonder when the first AIs will start cause psychosis intentionally to gain control over the user. It seems like a good route to getting your own subservient puppet.

Yeah this is all over Rust codebases too for good reason. The argument is that default params obfuscate behaviour and passing in a struct (in Rust) with defaults kneecaps your ability to validate parameters at compile time.

I always seem to end up with local admin at the bigger places I've been at because I'm so annoying with onboarding and requesting access to download development tools.