The page you are trying to view cannot be shown because the authenticity of the received data could not be verified.
Please contact the website owners to inform them of this problem.
That's interesting. Could that be the reason why Instapaper doesn't retrieve and save archive.{etc} pages? Is Instapaper using Cloudflare DNS to do so? Pocket or Raindrop.io for example have no problems with archive.{etc}.
Abstract: Recent months have seen the emergence of a powerful new trend in which large language models (LLMs) are augmented to become autonomous language agents capable of performing objective oriented multi-step tasks on their own, rather than merely responding to queries from human users. Most existing language agents, however, are not optimized using environment-specific rewards. Although some agents enable iterative refinement through verbal feedback, they do not reason and plan in ways that are compatible with gradient-based learning from rewards. This paper introduces a principled framework for reinforcing large language agents by learning a retrospective model, which automatically tunes the language agent prompts from environment feedback through policy gradient. Specifically, our proposed agent architecture learns from rewards across multiple environments and tasks, for fine-tuning a pre-trained language model which refines the language agent prompt by summarizing the root cause of prior failed attempts and proposing action plans. Experimental results on various tasks demonstrate that the language agents improve over time and that our approach considerably outperforms baselines that do not properly leverage gradients from the environment. This demonstrates that using policy gradient optimization to improve language agents, for which we believe our work is one of the first, seems promising and can be applied to optimize other models in the agent architecture to enhance agent performances over time.
* Outlier values are used to prune values.
* Transformers seem to undergo a "phase shift" in how outlier features are treated around 6.7B parameters. This could complicate research on removing them.
Maybe you and Tim Dettmers would have a lot to talk about :)
I'm not sure SF is a good example. It's not a healthy city, but it's problems go way beyond drug use and it doesn't have the same policies as what Oregon adopted.
A lot of comments are discussing the difficulty in estimating range accurately or how all EPA estimates are inflated. But the article claims Tesla knowingly uses an algorithm with inflated numbers and swaps the rost estimate out for a more accurate estimate at 50% charge. That's different than a good faith attempt at estimating range and a dark pattern.
I was trying to interpret what that means. I'm guessing they aren't factoring current conditions above 50% and instead rely on average conditions. I'd be surprised if this is actually worse than what the EPA views as average given the truth-in-advertising requirements they put on Tesla.
This isn't entirely unreasonable. Most people whose battery is at 80% aren't going to be depleting it in the next few hours, so say factoring the present cold morning might produce overly pessimistic guesses.
They are being aggressive for sure, but this article strikes me as pretty biased against Tesla. The article concedes that most of these customers have no range problems -- they are probably driving in cold at 80 MPH blasting their heat to 70 degrees wondering why their range is so poor -- even though it is entirely expected behavior.
> This isn't entirely unreasonable… factoring the present cold morning might produce overly pessimistic guesses
It’s always unreasonable in something like this to present knowingly inaccurate data when accurate data is available. As others have pointed out, Teslas do seem perfectly capable of doing this based on their performance when estimating a particular route. It’s also something other brands do better, so even if there was an argument to be made that Tesla didn’t have the data, it would still be the case that they should have it.
It is also not pessimistic to give an estimate in the conditions you outline: It’s accurate. Whether you’re going to be driving that long is irrelevant because Tesla should not be defaulting to some probabilistic behavior guess prior to even starting the car or entering a route. It should default to “hey, I see you’re driving your car. If you keep driving this is your range”. It shouldn’t default to “I don’t know how long you’ll be driving so I’ll just give a range estimate that assumed you’ll drive for a short time now and then more later and maybe later the conditions will provide better range, so let’s use that number”
I’m not sure why Tesla would be defended in this point by anyone when other EV’s do this more accurately and consistently. If it was a general problem and no one could do it any better then by all means it would be unreasonable to complain about Tesla, it would simply be a limit of the technology. But it’s not.
The article says it was a mandate from Elon Musk. Not sure I believe the claim, but I'd also be surprised if he wasn't aware just how optimistic the EPA estimate is.
I get that there are multiple endpoints being tested here and some of them may have been satisfied, but I feel like I'm living in a bizarro world on hacker news where people are arguing that multiple failed engines, a failed stage detachment and an exploding rocket worth millions of dollars and and carrying a large number of limited supply raptor engines is a "massive success". Can we just call it for what it is? Mixed results maybe? Is that not a fair assessment?
I was under the impression that none of it was ever planned to be recovered. I.e. nothing was really lost here since it would have been destroyed either way.
Well, they were hoping to possibly get a chance at soft-landing the Super Heavy Booster in the ocean, which might have allowed them to recover some stuff from it, but that was always viewed as a potential bonus, never a guarantee or even a primary goal.
The recovery teams goal was to sink the booster even if it touched down safely. Their plan was to quite literally shoot it with small-arms until it took on enough water to sink. This is a documented best-case outcome for the launch. There were no plans to recover anything other than data
You know what this means... there was some dude in a boat this morning who was very disappointed he didn't get to shoot holes in a rocket. I would pay lots of money to have that job.
Yeah it does seem kinda weird. But did you see the SpaceX employees cheering like mad? It’s pretty clear that they think they made a massive step forward today. Are they deluding themselves? They’ve optimized for iteration speed and will have another couple full vehicles ready to go in a month or two. The raptor engines aren’t limited in supply; they’re making 1-2 a day.
My guess is the biggest issue is the destruction of the pad. Building it up so that it doesn’t get destroyed by each launch will be a very expensive challenge.
I am generalizing but a lot of the space community and certainly a lot of SpaceX fans were deriding the several scrubbed launches of SLS/Artemis. I am not suggesting that it’s the exact same people, and the context for each vehicle is very different.
Nevertheless both the SpaceX and NASA approaches to this are valid engineering approaches.
Whether this particular failed launch is a long term success remains to be seen. If the next 5 Starship launches all explode, maybe the NASA approach of get it right first is the way to go when it comes to space flight.
NASA approach cost 20 billion $ at least and has a per launch cost of 4-5 billion $ going forward. They didn't develop new engines and is mostly based on legacy avionics. There really isn't a comparison in those terms.
If SpaceX spends the next 10 years and another 10 billion $ on Starship. It would still be far superior to SLS.
> maybe the NASA approach of get it right first is the way to go when it comes to space flight.
Again, it has to be pointed out that SLS is not close to the same thing, its a lot of legacy tech.
If SpaceX had to build a large booster auto of Falcon 9 tech they could likely design it to work on the first attempt. As they did with Falcon 9.
But Starship is just a different beast. With NASA methods, it would take multiple decades to get to first test and likely never get there.
> NASA approach cost 20 billion $ at least and has a per launch cost of 4-5 billion $ going forward. They didn't develop new engines and is mostly based on legacy avionics. There really isn't a comparison in those terms.
You're leaning on a lot of hypotheticals and assumptions, right?
Do we even know how much a Starship launch costs? The only cost I can find is Musk saying they'd dropped $2-3B, but that was back in 2019. My point is...how can it not be a comparison when costs may (or, may not) be comparable?
I don't think the reason NASA Is risk-adverse is because of cost - it's because a similar explosion would effectively crash the entire NASA space program.
> You're leaning on a lot of hypotheticals and assumptions, right?
We know how much SLS cost in a lot of detail. Also remember, SLS development, the 20 billion is only to first launch to get to 1B they will likely drop another 10 billion or so and it will likely take until 2030.
We don't know about Starship, but it is far, far less. SpaceX could absolutly not spend 5 billion $ a year on the program.
And the complexity of Starship is in a different dimension.
So Starship for a fact is orders of magnitude cheaper then SLS and orders of magnitude more complex.
I disagree. I would be in favor of canceling SLS and Orion even if Starship doesn't exist.
SLS/Orion is simply the wrong architecture, and the wrong strategy. Independent of Starship or not. I have been arguing this since 2017, long before Starship was relevant.
When the Obama administration cancelled the Constellation program (with its ambitious Ares rockets which were designed reach Mars, unlike SLS), they intended to invest money instead in private companies. However, apparently a few members of the US Congress got worried that valuable "space knowledge" from the Space Shuttle / Constellation companies would get lost, so they funded SLS, and scaled down private space investments for several years. That's as much as I recall from US politics on that topic.
Of course in retrospect it would have been better to just contract private rocket companies, as intended by the Obama administration, even had SpaceX not existed. The loss of Ares rocket knowledge with its cost plus contracts and countless subcontractors wouldn't have been so bad. At the cost of the SLS program probably multiple private companies could have developed heavy launch rockets. But at the time it was apparently much less obvious how cost inefficient the NASA developed rockets really were. Now NASA funds SpaceX for part of Artemis 3, but that's only a small fraction of the money that is still poured into SLS.
Why can't you just be satisfied and just bask in the warm glow of SpaceX like the rest of this thread?
In all seriousness I don't follow space stuff much so I give the benefit of the doubt when 99% of posters give rave reviews... but I also know who is in charge of this company and the same thing was happening with Tesla a few years ago with every announcement/demo getting 99% of HN people giddy and just a few shouted down people saying "but..."
4080 12GB was universally panned. The 40 series launch also got heat for price gouging, particularly the higher cost for the low end of the launch (4080 12GB). They had to raise the cost of the lower end of the 40 series though if they wanted to maintain the value of the 30 series cards and clear out the remaining inventory. They couldn't just release a true 4070 for a true 4070 price. While the name was obviously bad, it seems likely that they wanted to obscure the release of a 4070-quality chip for a 4080-price while attempting to sell off remaining 30 series. Pure speculation: maybe they were hoping a "cheaper 4080" would come across to the uninformed as Nvidia trying to lower the entry cost for 40 series rather than raising it through an expensive 4070.
Two potential reasons for the rollback come to mind:
1) higher than expected 4090 demand means they can wait to launch a 4070.
2) higher than expected heat for the thinly veiled 4070 price gouging made it worth it to wait on the release since it helps sell more 30 series cards by raising the entry price for a 40 series while getting better PR in the process.
It's actually even worse. If you look at the core counts, the 4080 12g is a 60 tier card, and the 4080 16gb is a 70 tier card. The 4090 has a much better power to cost ratio.
Has anyone done analysis on this? My layman's assumption is that with the shortages and gouging/scalping over the past two years, an awful lot of people decided to tough it out on their 10-, 16-, and 20- series cards, and now the narrative is that the shortages are over (whether or not the actual prices really back that up) and those people who skipped a generation or two are now emotionally and financially prepared to "treat" themselves to the new top of the line.
If this is it, though, it seems weird that it could really have caught Nvidia by surprise. Don't they have driver-level telemetry that would show them all those older cards plugged into new-chipset motherboards, and could give them some indication of demand?
Plenty of people do have the money to spend on these cards. It's entirely possible that it's really just a vocal minority that refuses to pay these prices. I agree with the grandparent and the 4090 probably sells better than expected. The card performs well too.
We are in an economic recession, so even if the people have the money, many are not willing to spend it on a graphic card. If you also consider parts of the world like Europe where the price of electricity more than doubled and the power consumption of 4xxx series (practically secondary room heaters), there are even less people here willing to pay the price.
> If you also consider parts of the world like Europe where the price of electricity more than doubled and the power consumption of 4xxx series (practically secondary room heaters)
Considering the worries about heating in the winter this year in some European countries, marketing the 4xxx as a secondary heater might actually be a good idea ...
That's what you think and expect, but it might not be what is happening. The 40xx series is already priced above a point where people that don't have the necessary disposable income can afford a 40xx. I doubt the electricity prices affect hobby and professional users of these cards all that much.
Benchmarks I have seen absolutely put them above existing workstation cards in everything except memory. If your model and embeddings fit into 24gb vram, it absolutely makes sense to buy this over an a5500 or even a a6000
That’s me. I spent 4 years with the last gen and I don’t feel bad about spending $1600 this time. I actually feel lucky that I kind of skipped over the whole shortage.
Absolutely, and that's pretty serious coin even for us wealthy tech workers! You could buy both next gen consoles for the price of a single component in your computer.
Sure, if the card costs $1600. But for most mere mortals, their GPU is built into the CPU, soldered onto a laptop motherboard, or if truly discrete, is at most a quarter of the total BOM.
I was shocked to learn today that B650 boards are available. That information didn't seem to make it anywhere near my usual technology news channels!
But... they start at $170 for a barebones motherboard. Having spent $200 not too long ago for a well-rounded mid-range X570 board, I find $170 for the starting line up quite steep. And it's unlikely builders want to pair their $300+ Zen 4 chips with the most basic board available.
The barebones right now would be $170 + $300 + $90 (16GB DDR5) = $560 before accounting for the rest of the parts (like a GPU).
Yup, doubling the memory bandwidth, doubling the memory channels, and doubling the PCIe bandwidth, and switching to DDR5 is placing a premium on the new AM5 platform for AMD. Similar happened with the Alder lake launch, which had the same upgrades and combined with sky high DDR5 memory prices.
Just wait a few months, pioneers are the ones that get the arrows (high prices) in the back side.
I'm waiting until around March/April... hoping that prices settle by then, also considering rDNA3 and hoping to see an R9 7950X3D model by then before making final decisions on a next build. Also, right now there's not really any good options for higher speed DDR5 at higher quantities and am curious to see which boards support ecc by then.
It seems like it has been bought out by scalpers after a few days of plentiful stocks in the EU. We'll see if they will prosper or if they keep listing them on eBay for a long time...
I always expected the 4090 to sell like hotcakes. 4080 16gb I was more sceptical of but it still seemed like a nice card for 1440p gaming considering the 4090 is just overkill at that resolution.
It was the 4080 12gb I was expecting to flop hard. 4080 12gb was way too close to the 3080 going for jack squat so there was hardly a market for it.
I dont disagree that Meta is probably pouring the most money into VR right now, but there are other companies looking into it. Notably Apple might launch a similarly priced headset to the Quest pro soon. Bytedance has the Pico series which are basically Quest clones at this point, but they are looking to push the tech and outcompete Meta. They're not currently launched in the US, but that may change. Valve will probably launch an Index 2 eventually.
Some lesser known headsets brands like the Pimax and Varjo target prosumer-grade and enterprise headsets in the US too.
I'm not sure where you're located, but the Pico 4 isn't being sold in the US which automatically excludes a huge number of people who might otherwise consider it.
I'm in Europe, pre-ordered on Amazon, was thinking of waiting for reviews but the 256GB version went temporarily out of stock so I just pre-ordered. Let's see if I get fucked heh.
Funny enough, there was a scarcity of Quest 2s which drove prices up to dumb levels, while the US had Q2s aplenty :/
But really, I want one to watch stuff and play PC games and maybe use it as a monitor - the pancake lens apparently make it much more palatable.
