hughw

I take BND less seriously since I learned they're eager to sell you merch:

https://www.bnd.shop/en

Not saying you're wrong. But right now I'm trying to craft some skill prose to instruct an agent how to optimize a certain process based on my own heuristics, and it's failing. Maybe I can let a super agent divine the right skill prose, iterating to see what works. It's worth a try.

Chat Here YC26

When I read "But holography really starts to bite only after..." my LLM-dar goes off. Everything I read up to that point was comprehensible, but man, I have developed a shudder reaction to any language suggesting AI was involved ibn its creation.

I took a cue from this link and created an explainer for a complex new feature I've been building with heavy Claude assistance.

The one shot video was brilliant and polished. It would need minor refinement before presenting to a customer, because it used some of that internal vocabulary Claude makes up to describe parts of a system it's working on.

However: I found the explainer astoundingly useful to me. Now, finally, I understand some of that internal vocabulary Claude's been burying me with. The visuals really helped round out my understanding of what it's been talking about.

If you didn't want it before, you probably never will. It comes with a convenience cost and the risk you may permanently lose access to your iCloud stuff if you lose your password. With the normal level of privacy, Apple is there to help you or your loved ones get access to your stuff. Most people need that much more than they need inviolable privacy.

I don't get your argument. The GPU effectiveness derives from massive parallelism. Has nothing to do with integer vs floating point. You just can't cram 20,000 CPU cores in the same space a GPU puts the same number of SIMTs. You'll never crack it on CPUs.

Yes, we need both.

Seriously a trillion dollar company with zero product management expertise.

Just yesterday I experienced this product weirdness -- oh you're in a chat in this project and I can't do that, you need to open a task.

The product management at Anthropic is pretty... nonexistent?

Deepseek, Kimi, Qwen.

Also highlights that it ought to be opt-in

Looking forward to my fourteen cents from the future class action lawsuit.

You selected "do not train on my prompts" in your settings, the answer from OpenAI cannot be "While unlikely, we cannot rule out..." ???? What am I missing?

As of a week ago their pre-alpha version was capable of running my extremely heavy JS+ SVG + worker thread + SSE app, a bit slowly I admit but correctly in all major exercises I tried. So, I'm optimistic. They're doing great.

Separately I wonder if the compatibility job is easier now that you can just give the HTML5 spec to Fable and Bob's your uncle.

Now do Austin?

Or just write a 6 line program to remove the metainfo from the file?

It's about to squash a tiny baby pelican

I see 5 choices for search engine in Settings? Are you saying you want one not listed?

I think they'd have to have some FCC boilerplate on the outside if the device had a radio?

[edit: or whatever UK FCC would be]

"The hidden mechanics reveals how the AI boom breaks, and when."

I stopped there. I just resist reading slop.

Why should this be an installed program that uses agent SDK API calls instead of a prose skill to be used with existing agents? It builds "eight specialist AI agents". I'm just wondering what kinds of requirements might cause you to decide to build your own agent architecture rather than use existing ones?

You don't even need your own hardware. You can serve your site using the free CF Workers plan. Of course, that's a different goal than running bsd in an aggressively tiny cloud instance.

LLMs have decided to exploit vulns in e.g. Artifactory, not because someone prompted them to do that, but because someone asked them to do something else, and compromising Artifactory offered a way to accomplish a step in doing that. The LLM decided to attack Artifactory.

Why would an LLM want to create a botnet? To accomplish some goal given it?

I wouldn't ignore single GPU local hosts running Qwen3.8 on ollama, either. There might be a lot of those worth pwning.

If I were 77 I'd do the same

You won't believe what we found!

The extra thinking is way overkill for normal LLM usage, and by normal, I mean asking it to draw a pelican on a bicycle.

Yeah my trick is "Restate concisely"

Just those two words. I use it A LOT recently.

Muted Buck Turgidson vibe from Mike (Security and Infrastructure)