Thanks for sharing. Have you prodded the model with various inputs and written an article that show various output examples? I'd love to get an idea of what sort of "end product" 4x4090s is capable of producing.
You might find more information here helpful https://sabareesh.com/posts/llm-intro/
But i am still in process of evaluating post training process with RL. RLHF is almost a mirage that shows what is possible but not the full capability of what model can do
I'm not sure what it's called (abstraction vs. indirection) but I dislike when everything needs a class/object with some odd combination of curried functions. Some programming languages force this on you more than others I think? As a contrived example "StringManager.SlicingManager.sliceStringMaker(0)(24)(myStr)", I've seen code that reminds me of this and wonder why anyone uses a language where this not only an acceptable idiom, but a preferred one.
Using EEG to predict thought is like looking at the clouds in Mumbai to predict the clouds in Austin. The electrical signal from individual neurons are lost in a sea of large-scale oscillations, which are further blurred by the layers of bone, muscle, and tissue that separate the device from the brain. Bitrate is like 1 bit per second, completely insufficient for most use-cases.
Very excited to see what Sam & Greg are up to in the coming months! Guys like this don't just run away with their tails between their legs. They will be back.
Agree, moving the weights inefficiently will end up working the muscles you don't necessarily target in your main workout. It's less efficient, but probably gives a (very slightly) more well-rounded workout. If you're pressed for time, it could make sense to have a basic understanding of how to avoid dilly-dallying during your workout. If you go to a public gym, there will be other factors that affect your total time much more, like having to share equipment, which introduces a lot of uncertainty. Maybe these micro-optimizations are worthwhile if you have a private gym.
From a well-being/philosophical standpoint, maybe it's better to live life relaxed, and not one where you have to micro-manage every minute of your day to squeeze out every inch and penny of efficiency you can. That sounds like a horrible lifestyle, but I guess to each their own :)
Deep fakes were always a huge concern for me in AI. That's just one way AI can be weaponized, and this example is very clear on the damage caused. Thankfully Twitter quickly corrects the issue, but sadly only because it became viral. Sadly, Twitter won't be able to correct all of the little cases of AI misuse that don't go viral.
Neat and probably works on most dishwasher models, but not enough temperature precision for my comfort level. I wouldn't trust it. Rather just pan fry or bake myself a salmon. I don't have a sous vide, but I imagine this is not a replacement for one since the whole point is precision cooking.
> Neat and probably works on most dishwasher models, but not enough temperature precision for my comfort level.
I'm not so sure. IIRC, about a decade or two ago, dishwasher energy efficiency regulations forced design changes that really neutered dishwasher drying cycles. It does seem like they don't get as hot as they used to while drying. I wouldn't be surprised if this recipe worked in older dishwashers that were around when it was first popularized, but doesn't in more modern ones.
A sous vide dishwasher. Now there’s a project some crazy YouTube engineer ought to get on!
Dishwashers are already pretty much air tight. You just need a pump. How strong does a vessel have to be to resist an atmosphere of pressure? But I guess it’s sous vide rather than non vide.
I discovered this when one of the plastic conduits came loose inside my dishwasher in precisely the right way as to spray water directly at the inside bottom of the door. The top and sides have a rubber seal, but the bottom is protected pretty much entirely by the interior shape of the door and the bottom tray.
It makes sense, if you think about it, since you would absolutely want some way for the humidity to gradually equalize if you left it alone with the door closed for a while.
I wouldn't worry about it - you can eat it raw after all and the cook time isnt long enough for it to make you sick (due to food borne illness from being in the temperature danger zone.
Side note - sous vide steelhead is incredible, thinner pieces in my experience work out noticeably better
Raw fish served as sushi is generally frozen to -31 F for 15 hours to kill the bacteria before it is served raw. I would not eat grocery-bought salmon raw unless it is "sushi grade".
Unless you're buying locally caught Salmon at a fish market, it was almost certainly flash frozen. Even if it's not frozen anymore when presented in the store.
Sushi grade fish is flash frozen at -40°f - not all fish is frozen this way. Most [0] sushi grade fish in the US is frozen this way before distribution. This is not the case for ALL fish. It would almost certainly have been frozen, just not necessarily flash frozen with the intent to distribute as sushi grade.
Most fish you buy will have a warning specifically saying not to eat it raw if it wasn’t frozen safely in this manner.
Cooking salmon is mostly about texture. It's usually done around 120, but it takes 145 according to the USDA to actually kill off parasites. Deep freezing kills them instead.
According to a brand of dishwasher detergent, machines run within the food safe danger zone.
So, while it'll improve the texture, cooking salmon actually makes it less safe to eat.
USDA numbers oversimplify, but killing pathogens is a function of both temperature and time. 145 will pasteurize salmon instantly, but you can also hold at 130F for about 1 hour.
Totally. I personally cook salmon until 120F and then sear the skin side. But was addressing the parent poster about the temperature that kills pathogens.
Edit: I now see that parent was talking specifically about parasites. Indeed those are killed by freezing beforehand. But a long bath at >130F kills the bacteria.
Engineer who implements correct, comprehensible code but doesn't manage ticket statuses is more valuable than one who manages ticket statuses but scatters the codebase with technical debt and confusing abstractions/code. If "better communication" means spending an extra 10 hours with the latter dev to correct/re-teach them, then yes, communication is the problem. The most time I've lost at work is correcting/teaching engineers who eventually got let go due to low performance.
I don't really think the first part of your comment is true, at least not in my experience. All other stakeholders would prefer someone that updates their tickets and communicates what they are doing effectively, even if they produce absolute garbage, bug ridden software.
I've personally only ever seen someone removed from a team for having poor technical skills once and it was under pretty extreme circumstances. While I've seen many good (from a technical perspective) developers removed because they thought they were somehow above doing the everyday "busy work" like ticket management.
Unless you work alone, then refusing to do non-technical work is just saying you're going to let the rest of the team do it. You're much more likely to be removed or have your contract ended if the team doesn't want to work with you.
Without going into anything identifying, what exactly were those circumstances?
I'm imagining something like this happening, but at the same time I can't help but think if its a technical skill issue, handing them a book on the relevant technology and telling them studying this book is their job for the next two weeks would be better than trying to roll the dice again, put out a new job ad, interview, vet, rehire, and hope they aren't somehow just as technically incompetent as the original engineer.
Vindicated and excited. Gradient descent is likely not enough. I love it when we get closer to something but are still missing the answer. I would be very happy if "add more parameters and compute" isn't enough to get us to AGI. It means you need talent to get there, and money alone will not suffice. Bad news for OpenAI and other big firms, good news for science and the curious.
I imagine physicists got very excited with things like the ultraviolet catastrophe, and the irreconcilable nature of quantum mechanics and general relativity. It's these mysteries that keep the world exciting.
There's something ironic about implying that us not having a path to AGI is good news for the curious. If you're supervisiously curious then sure, we need to unlock another piece of a puzzle, more puzzle pieces means more puzzle solving.
But if you're able to actually take a step back, AGI would be the the ultimate source of new puzzles for the curious. We don't even all agree on how to define the "GI", approaching AGI wouldn't be unlike meeting extraterrestrial life sitting on a computer.
I think you misunderstood the parent, who was probably saying that the process for achieving AGI would be more interesting if it isn't just "more compute/training".
I think you misunderstood me since that's exactly my point.
From the trees, it's great for the curious that it might take more than compute and training.
From the forest, it'd be infinitely preferable if AGI were just a matter of more money. There are mysteries we can't even envision yet that would more than make up for any "lost curiosity"
Al right, all right, but to me the implication was that only those who have ludicrous amounts of money would be able to play with it. And I don't think that's the most desirable outcome, is it?
