That's how I read it too, but the record is only for hottest June 24th of all the June 24ths since 1888. So yeah it's 1/number of years or so but we get 365 attempts at that a year to have this headline.
I'm not going to argue these are good tests, if you asked a coworker these questions they'd look at you weird, but what surprised me is how well you can encode a sentence never written down before, put it through base64 encoding, and then ask an llm to decode it. And the good models can do this surprisingly well.
Is there any rule of thumb for small language models vs large language models? I've seen phi 4 called a small language model but with 14 billion parameters, it's larger than some large language models.
Since I won't get a better chance to find knowledgeable people: How exactly does iPXE work? Lets say I want to boot a random old PC from one of these images - does the bios need to have iPXE support? Or PXE? Or does it need to first boot into some local system first, which then goes out and gets the image to boot the rest of the way from?
Then there's mention of DHCP and self hosting. Do I need to have some service running within my LAN first or can this go right out to some public server and boot from images there? How does DHCP factor in? I am in so far over my head on this but it seems interesting and something fun to try out.
Can anyone speak about plans for a Pico 2 W (or Pico W 2)? I've been playing around recently with mine and even just syncing with the current time over wifi opens up a lot of possibilities.
I was just wondering the other day about the llm frameworks out there. Ones I've heard things about:
Griptape
Dspy
Autogen
Guidance
Semantic kernel
Marvin from prefect
Haystack
And of course langchain and Llamaindex.
Which ones am I missing? Are any of them particularly good? Things like a simple way to interact with it (Marvin stands out), clean and organized code base, good support... If I had to pick one today I'd probably pick semantic kernel but not because I have used it, just because the ones I've used have either fallen short or been too much unnecessary complexity.
The idea is to serve models that would normally be considered too large for GPU memory (70 billion parameters at 16 bytes each for 140 GB of memory required). Some people figured out you can offload the model and only have parts of it loaded so a 24 GB GPU like the 4090 can still serve the model, but it goes a lot slower. They have a new way to serve the same model on the same GPU but 8x better throughput. Something about decoding tokens on a smaller model maybe, then just checking multiple tokens on the larger model in a single batch. Magic, but ultimately its the same model, same GPU, same output as before, but much better throughout.
It depends on the task I imagine. Like writing a novel was mentioned, keeping important story lines in your memory for a long time will be necessary, or at least certainly more important than remembering what the characters were eating for lunch on page 10. But if you need to find that one loophole in a contact you probably will benefit from the perfect recall.
I think kelseyfrog meant that the state for a mamba model is supposed to "remember" stuff even if it doesn't have the actual tokens to reference any more. It might not be guaranteed to hang on to some information about tokens from a long time ago, but at least in theory it's possible, whereas tokens from before a context window in a tradional llms may as well never have existed.
Not the person you are responding to, but I'm aware that the cloud is just other people's hardware. But that's usually what I intend to mean when I say that. My own hardware I have to power, maintain, reboot, dust, cool, etc. And if I talk about "the cloud" I usually am taking about an environment I can maintain fully through software, none of the messy hardware failures and temperature management I might need to think about at home.
Justin Trudeau is not the president of any country. He is, however, the Prime Minister of Canada, which he has been since November 4, 2015. The prime minister is the head of government in Canada's federal parliamentary democracy and constitutional monarchy.
Since you haven't had a response to the second part, here it is:
PreviewJSON
Justin Trudeau is not the president of any country. He is, however, the Prime Minister of Canada, which he has been since November 4, 2015. The prime minister is the head of government in Canada's federal parliamentary democracy and constitutional monarchy.
This right here is why there is a divide the size of the Grand Canyon in America. Some will see it as obvious satire, some are 100% convinced this is the truth. When did we reach the point that the same text can be read as fact or satire with such strong conviction that the dress being blue and black (no, it's white and gold!) would seem like compromise?
Where can I learn more about these phrases to say to capture as many sounds as possible? They listed two in the article, but now I'm curious how they decided which sentences to say, how efficient they are, if they account for the same sounds but with different qualities, like going up at the end if it's a question? It's a rabbit hole I need to find the entrance to.
Same! As soon as a new cert is registered for a new subdomain, I get a small burst of traffic. It threw me off at first assuming I had some tool running that was scanning it.
This seems about as close as we might get to another fun one, https://cratebeforeattack.com/ for more traditional worms style gameplay. It's fun but crashes randomly, and hasn't seen much since 2020. It anyone knows more details about if the game is looking for a new maintainer please reach out! There is so much potential here.
I found that it works well to ask it to generate JSON the best it can, then pass it to gpt-3.5-turbo with the JSON response mode and instruct it to just clean up whatever input it received.
I've been playing with a similar idea of screenshots and actions from GPT-4 Vision for browsing, but after trying and failing to overlay info in the screenshot, I ended up just getting the accessibility tree from playwright and sending that along as text so the model would know what options it had for interaction. In my case it seemed to work better, I see the creator is here and has a list of future ideas, maybe add this to the list if you think its a good idea?
Its a result of how these transformer models work. It's pretty quick for the amount of work it does, but it's not looking up anything, it's generating it a token a time.
I've seen this link pop up in various places now, but it seems like it's still mostly not being developed? Is there a reason it was posted today? Some new development in it?
We're on HN! People might find it more interesting to find out it's not just over 1000, but 1024 exactly. And I'll leave it for curious minds to think on it why that becomes obvious when seeing the problem from the right perspective.
If any cloudflare employees end up here who helped decide on Capn Proto over other stuff (e.g. protobuf), what considerations went into that choice? I'm curious if the reasons will be things important to me, or things that you don't need to worry about unless you deal with huge scale.
As someone who knows more about words than superconductors, gentle correction in case it is an actual misunderstanding instead of misspeaking, but a correctly running MRI shouldn't ever CONSUME the liquid helium, it would recycle as much as possible.
But as a newly self-proclaimed expert on superconductors as well, yeah this would probably help MRIs. My understanding is that the reason for superconductors in MRIs is so that the wires doing the electricity stuff don't interfere with the small electrical responses from the tissue it's measuring. Without resistance, you don't get magnetic fields around the wires or something.