camkego

Does this article really compare a single well defined LLM model "Kimi K3" vs. a family of LLM models including Haiku x.x, Sonnet x.x, Opus x.x, Fable x.x without actually revealing what Claude-family model was being compared?

(Fable has been restricted somewhat, but the article uses Fable pricing as a comparison point, so it worth including it in the list of possible Claude family LLMs)

It's hard to know what to take away from the post with this ambiguity.

It's also worth noting the range of pricing for the Claude family ranges from $1/$5 a million in/out to $10/$50 a million in/out, so the ambiguity of which particular model the comparison is against spans a 10x range of model fees.

I've got to give the author credit, it seems like this is written without an LLM. But, there seems to be a lot that is arguable.

For example, the article says "Natural language is the interface of last resort."

But, is that actually true? Isn't the AI boom happening because N/L is truly available as an interface?

Anyway, please keep writing and asking tough questions.

This project desperately needs a rewritten introduction, hopefully with outside feedback.

You might not like Microsoft but they did a video on why they re-wrote the last version of the Typescript compiler in Go. Basically, because of LLMs. It's worth viewing even if you don't decide to go with Go.

I think the study you might be thinking of shows a drop in crime due to the reduction of lead in gasoline, thus we are now breathing less lead.

Read this section of the article “ Bug Discovery: Finding Hidden Flaws”, they appear to have used the model on open source Rust to find issues starting with just the Rust code. You might be also able to have conversations that help you write the Lean to verify your application, but I’m not certain about this.

This. This is why everyone who wants to fingerprint and collect tons of data on end users pushes them hard on installing an app. The amount of valuable data is 10x what’s available in the browser

Pin threads to cores, and make sure threads different cores aren’t writing to the same 64 or 128 byte block. Lookup “false sharing”

Maybe I am cynical, but I read this as "We were not able to meet our deadline of making you pay API rates for -p and the Agent SDK, but it's coming. We will let you know"

If I've got it all wrong, please let me know.

They are certainly playing loose with the truth: README says "Standard codecs (H.264/VP9) require dedicated hardware decoders"

Thanks to this article, I learned that the 70's Sticky Finger jeans brand were named after a Rolling Stones album. Don't know how I missed that all this time.

Well, according to Sonnet, the term "staging" refers to assembling troops and equipment at an intermediate location before moving forward into an operation.

I really didn't know, but was curious, so I used an LLM to research it.

The article contains the passage below twice. I mean how does that happen? No proofreading? Cut and paste editing. I am always surprised when I see this in professional new sources.

——

That the amendment died quietly does not erase what its introduction signals: opposition to police LPR programs is reaching higher levels of the political agenda, and Flock is increasingly at the center of it.

Maybe I am wrong about this, but I think Zed will run the npm stuff on the dev-container if you are using dev-containers. That can be your isolated virtual machine image or docker instances. But I believe you do need to use Zed (stdio or ssh) dev containers to get that security isolation. I know it’s a pain, but for me, I am going to pay the logistics price for security until a better solution comes along.

This would be kind of a fun challenge. If you are handling random numbers, well you are limited by disk or memory size. But if the numbers are compressible ala LZ77 or Gzip, then there are ways to use the value’s compression trees to sum the numbers from the least significant digits using the LZ77 style compressed value tree representation. If you go that route, and the numbers are compressible (not random) then the question is whether the compressed input and output trees fit in memory or disk.

The security model, or almost lack of any whatsoever in VSCode drove me to only install MSFT extensions, then use Code Server in a docker container, but I decided I didn’t like using my editor in a browser.

Finally I have decided to start using Zed, which isn’t perfect on the security front, but much better IMHO. The combination of WASM extensions, and the ability to put language servers, etc, in dev-containers seems like a great step forward.

I hope Zed continues to improve their extension and language server security model. Actually I hope VSCode does too, but honestly, I am not optimistic.

Supposedly, in 1990, there was somewhere between 132,000 and 270,000 travel agents. Consider that.

It would be fascinating to know where the remote drivers were located that were remotely controlling these vehicles. Wasn’t there a big hubbub about using remote staff in the Philippines a while ago? This can change the reliability profile quite a bit. (Internet quality)

If you look at the start of that episode, there is another crazy thing in there, a device which allows you to "see" the bits on a credit card track.

Apparently something called "magnetic viewing film" can allow you to see the bits on the magnetic stripes of credit cards.

I had never heard about this before.

Link to video time: https://archive.org/details/bbc-connections-1978/Connections...

Why? Because of LLM vibe coding?

The way I read the Anthropic docs, it seems the term plan is to block the usage of OAuth credentials with the "Claude Agent SDK".

This URL: https://code.claude.com/docs/en/agent-sdk/overview

Says this: "Unless previously approved, Anthropic does not allow third party developers to offer claude.ai login or rate limits for their products, including agents built on the Claude Agent SDK. Please use the API key authentication methods described in this document instead."

Again, it seems Anthropic prefers to bill API token rates (long run), not subscriber effective token rates.

It seems clear that Anthropic wants users pay API rates for tokens when use in a programatic way, and not subscriber rates for tokens when used from code. As a user, I want to pay the subscription rates with -p, but it seems they want to block that.

I don't claim to understand the factors which cause this, but a lack of security and exposed valuables at unlocked-doors, pre-opening mall in Shenzhen China without issues at 8am in the morning is very curious.

https://www.youtube.com/shorts/TZMXdR5fDrw

It seems pretty foreign to me, a PNW US citizen.

I totally buy this as someone located in the US, but what is everybody else using? It can’t be WhatsApp? Is everyone sending all their connection graphdata to Meta?

This looks useful. But, it's interesting how the backend-world and front-end world keep diverging. I must admit, I had no idea what this was from the title. "CLI framework"? But in backend-land, these would typically be called "argument parsers" or "command line argument parsers". But maybe I am missing some of the functionality.

I found this website frustrating, it asks you a bunch of questions, then ask you to sign up with nothing in return for the upfront investment

I think if the online builder could have been the whole project, that would be neat! Truly "zero-trust", what I think many HN readers want.

Anyway, thanks for building Agent Safehouse.

If you look at the first image in the article, the one with a floppy on a serving tray, it looks like an 8 inch floppy to me. I think the floppy disks in the board room might also be 8 inch floppy disks

It is rumored heavily on HN that when the first employee of Google, Craig Silverstein was asked about his biggest regret, he said: "Not pushing for ECC memory."

When an article states “ OS vendors use everything in their power to make you not want to develop native apps for their platform “, it’s really hard to take the rest seriously.

Sure, native applications can be hard and slow to develop for a number of reasons, but OS vendors are seriously incentivized to have developers build on their platform.