That's not how you should measure "worth". In that world, you'd have a P/E ratio of 1. Comparing to a bond, it would be like expecting to get paid the face amount in a single year. Many people are quite happy with 5-10% interest as a risky benchmark, so 10-20 P/E isn't wild. That puts the market cap for tech itself at 10-20T as a reasonable baseline.
I feel vindicated :). We put in a lot of effort with great customers to get nested virtualization running well on GCE years ago, and I'm glad to hear AWS is coming around.
You can tell people to just do something else, there's probably a separate natural solution, etc. but sometimes you're willing to sacrifice some peak performance just have that uniformity of operations and control.
This isn't strictly correct: you probably mean wrt compressed size. Compression is a tradeoff between size reduction and compression and decompression speed. So while things like Bellard's tz_zip (https://bellard.org/ts_zip/) or nncp compress really well they are extremely slow compared to say zstd or the much faster compression scheme in the article. It's a totally different class of codec.
an LLM can be used to losslessly compress a string to a size equal to the number of bits of entropy of next token prediction loss over the string, by encoding the extra bits of entropy with arithmetic encoding. its sota compression for the distribution of string found on the internet
When you say "fixed this" which "this" do you think they fixed? Are you imagining this is a hash table? It's not
It's an adaptor which will use two other containers (typically std::vector) to manage the sorted keys and their associated values. The keys are sorted and their values are stored in the corresponding position in their own separate std::vector. If you already have sorted data or close enough then this type can be created almost for free yet it has similar affordances to std::map - if you don't it's likely you will find the performance unacceptable.
Every lift at Snowbird has the map printed on the bar. So you can plan your route on the way up. I agree that when you get lost, that map won't save you, but I think an offline PDF is also fine.
Absolutely not true lol. I dont think any of their lifts have maps on them right now. The maps also arent super helpful at snowbird because the cliffs often come out of nowhere
Wait, really? I haven't been up this season, but it's always been there! I understand removing the printed ones when the bars have them (and the big boards at the top). Is it all just ads now?
None of the bars have anything printed on them now if I remember correctly. I have a pass and have been around 10 times this season. At least they all have footrests unlike alta where they love foot pain
You can get by pretty well with the ~$20/month plans for either Claude or Gemini. You don't need to be doing the $200/month ones just to get a sense of how they work.
That's what I was thinking, but could not find the link. Here is it working on some standard tasks.[1] Grasping the padlock and inserting the key is impressive.
I've seen key-in-lock before, done painfully slowly.
Finally, it's working.
That system, coupled to one of the humanoids for mobility, could be quite useful. A near term use case might be in CNC machining centers. CNC machine tools now work well enough on their own that some shops run them all night. They use replaceable cutting tools which are held in standard tool holders. Someone has to regularly replace the cutting tools with fresh ones, which limits how long you can run unattended. So a robot able to change tool holders during the night would be useful in production plants.
See [2], which is a US-based company that makes molds for injection molding, something the US supposedly doesn't do any more. They have people on day shift, but the machines run all night and on weekends. To do that, they have to have refrigerator-sized units with tools on turntables, and conveyors and stackers for workplace pallets.
A humanoid robot might be simpler than all the support machinery required to feed the CNC machines for unattended operation.
> A humanoid robot might be simpler than all the support machinery required to feed the CNC machines for unattended operation.
A humanoid robot is significantly more complicated than any CNC. Even with multi-axis, tool change, and pallet feeding these CNC robots are simpler in both control and environment.
These robots don't produce a piece by thinking about how their tools will affect the piece, they produce it by cycling though fixed commands with all of the intelligence of the design determined by the manufacturer before the operations.
These are also highly controlled environments. The kind of things they have to detect and respond to are tool breakage, over torque, etc. And they respond to those mainly by picking a new tool.
The gulf between humanoid robotics in uncontrolled environments is vast even compared to advanced CNC machines like these (which are awesome). Uncontrolled robotics is a completely different domain, akin to solving computation in P by a rote algorithm, vs excellent approximations in NP by trained ML/heuristic methods. Like saying any sorting algorithm may be more complex than a SOTA LLM.
Most flexible manufacturing systems come with a central tool storage (1000+ tools) that can load each individual machine's magazine (usually less than 64 tools per machine). The solution to the problem you mention is adding one more non-humanoid machine. The only difference is that this new machine won't consume the tools and instead just swaps the inserts.
There is literally no point in having a humanoid here. The primary reason you'd want a human here is that hiring a human to swap tools is extremely cost effective since they don't actually need to have any knowledge of operating the machines and just need to be trained on that one particular task.
How would you distinguish the article from an honest write up about transistors? That is, you know about his crusade in ML, but if you didn't, how would you decide that this article is written in bad faith or not?
I agree that context matters, and I had the same thought as you. But does that mean that anything he writes on the topic of "who was first" is inherently tainted?
Are you saying "worth" as a shorthand for something like annual profit? If you sort the 2025 data by earnings, you get pretty large numbers quickly: https://en.wikipedia.org/wiki/List_of_largest_technology_com...
That's not how you should measure "worth". In that world, you'd have a P/E ratio of 1. Comparing to a bond, it would be like expecting to get paid the face amount in a single year. Many people are quite happy with 5-10% interest as a risky benchmark, so 10-20 P/E isn't wild. That puts the market cap for tech itself at 10-20T as a reasonable baseline.
reply