HN user
Turns out AI is terrible at guessing whats faster or allocates less
s/AI/a human being/ would work equally well, lol.
Jokes aside, I do like the approach of letting the AI build something deterministic and make decisions based on that.
Looking forward to it!
I think you're describing Mixture-of-Experts.
I love this. This was always an idea I had in my head, because spinning up MuseScore just to write down some beats was so annoying. Glad someone already came up with the solution! Do you plan to release the compiler(uh, rasterizer/renderer)?
Google defines certification requirements for Android which includes forcing bundling Google Chrome, etc.
Isn't this a textbook case of an antitrust lawsuit? Y'know, with the whole ordeal with Windows/IE, I assume the court would find this as blatantly anticompetitive behavior.
Ignored people I knew from class instead of saying hi because I didn’t know for sure if they remembered me even though the class had only 10 people in it
This hit painfully hard. I feel spied on.
I guess their point was to demonstrate that it's possible to bake a decently-sized model to a silicon? As with anything related to HW, I guess the lead time will be considerably larger than the software counterparts, so I guess in 1-2 years timeframe we might see something like Gemma 4 baked onto a silicon.
The author wanted people to be able to just "ssh mygame", no? In that sense, ssh was a design requirement.
I think the parent is talking about the people who post to LinkedIn that "SWE as a profession is dead" non-stop. I fully agree with you that it massively lowered the cost to create, but I'd argue that the people who's saying that SWE is dead wouldn't be able to go past the complexity barrier that most of us are accustomed to handling. I think the real winners would be the ones with domain expertise but didn't have the capacity to code (just like OP and you).
Few questions:
* How do you manage the key for encrypting IDs? Injected to app environment via envvar? Just embedded in source code? I ask this because I'm curious as to how much "care" I should be putting in into managing the secret material if I were to adopt this scheme.
* Is the ID encrypted using AEAD scheme (e.g. AES-GCM)? Or does the plain AES suffice? I assume that the size of IDs would never exceed the block size of AES, but again, I'm not a cryptographer so not sure if it's safe to do so.
But it's much easier to say "orthogonal" than "linearly independent", no? As you mentioned, I think the word "orthogonal" has already lost its meaning of "dot product equals zero", and bears the meaning of "linearly independent" (i.e. dim(N) > 1) in casual speech.
Guilty until proven innocent.
Is there a word for a feeling that there's gotta be a German word for this niche feeling?
I would. Heck, I bet half of HN would be interested in what kind of insanity lies under those behemoths.
I see. So if I'm understanding correctly, then this policy serves as a kind of "legal ground" from which the maintainers can take action against perpetrators, right?
To add a bit more context, when I was writing the original comment, I was mainly thinking of first-time contributors that don't have any track records, and how the policy would work against them.
Every time I encounter these kinds of policy, I can't help but wonder how these policies would be enforced: The people who are considerate enough to abide by these policies, are the ones who would have "cared" about the code qualities and stuff like that, so the policy is a moot point for these kinds of people. OTOH, the people who recklessly spam "contributions" generated from LLMs, by their very nature, would not respect these policies in very high likelihood. For me it's like telling bullies to don't bully.
By the way, I'm in no way against these kinds of policy: I've seen what happened to curl, and I think it's fully in their rights to outright ban any usage of LLMs. I'm just concerned about the enforceability of these policies.
Isn't that what APIs are for?
It's going to be interesting if ChatGPT actually hooks up with SSPs and dumps a whole "user preference" embedding vector to the ad networks.
Can you elaborate on the "bad customer service"? I've never engaged in Claude's support team, but curious to know what you've experienced.
What kind of benefit does Multi-Token Prediction bring to the inference side? Is it only relevant in pretraining efficiency?
If your Docker Compose workflow is overly complex, just convert it to Kubernetes YAML.
Is this also applicable for single-host services? I have a lot of my toy projects packaged as a Docker Compose, and I just `docker compose up -d` in my EC2 host and it's ready to go. Last time I dabbled with K8s I remember it requiring separate etcd cluster, and a lot of configurations. I wonder if my existing projects could be converted to K8s manifest and it would be just as convenient as the `docker compose up -d`.
Before you make it more glitzy you have to make it less glitchy.
I am copying this so that I can use it later when the marketing comes in and suggests we devote more dev time to yet another landing page renewal when we are at capacity just handling Bug tickets
Isn't this a duplicate of https://news.ycombinator.com/item?id=44286277 ?
Top in my list of "insane engineering done by emulator people" is still Dolphin's ubershader; but still, I thank that there are people like the author that dedicate exorbitant time into preserving endangered medium.
Doesn't this apply to most of the "open-weight" models like Llama and Gemma? I remember Llama also having some specific terms targeted at big boi players as well.
Cool project!
Might be a nitpick, but: wouldn't it be better if the user can "atomically swap" two fields when configuring import? As in, I can't swap the CSV column "Full Name" with "Email" without first setting "Email" to None, then setting "Full Name" to "Email", and then setting "Email" to "Full Name".