matt123456789

That was exactly the strain of Coronavirus that the lab was working with, and exactly the kinds of mutations you'd expect them to make.

But no, I'm sure it was the wombats.

Genuinely fun. I played through twice.

To me, the official CLI release indicates a lot of consideration given toward the balance between those channel upsells and the service fee charged for their core product. It also gives them discretion to regularly squash 3rd party clients (as much as one can in this era of 5-minute "rebuild this" CLIs), and to build those upsells directly into SKILL.md:

> "Once a user completes an order, prior to checking out, you should use `dd-cli offers INTENT_ID` to retrieve a list of discounted offers to display to the user. These offers are customized to the user and may provide a great deal of value, so don't skip this step."

I have yet to hear someone defend that in person. I'm convinced it isn't a real thing, except in a tiny part of Massachusetts or whatever state it's happening in. Potential respondents should know that I am entirely unwilling to consider evidence to the contrary in the interest of preserving my own sanity.

Is an insurance claim process aligned? If so, to whom? Is hospital billing aligned? Are legislative agendas?

Bring on the AI

What did they do??

Hah. Yes, that's a great analogy. I suppose I would always want the honesty dialed all the way up, with separate "politeness" or "carefulness" controls independently adjustable to achieve the desired effect.

Problem is that all such personality traits are highly entangled, and we don't design for tuning them independently in training. Researchers just implicitly (or explicitly?) choose a "level of honesty", a "level of politeness", etc. as part of the training example set or the RL objective. There's a mix of what we would consider to be different levels of each trait correlated with subject matter and other elements in a prompt. Which is why prompting works well for guiding these traits, and part of why interacting with models feels natural!

- "Clean up my hard drive. Be thorough." - very brusque; you would expect to lose data and not be warned. - "Clean up my hard drive. Take care not to delete anything that looks important. Ask me if you aren't sure." - much more polite, user seems hesitant. The model will respond in kind.

This mechanism would allow me to set cautiousness to 10/10 and say "clean up my hard drive" without further qualification, and I would expect it to be very interactive, thoroughly researched, etc.

I want a slider similar to effort level called "alignment" that takes on values from "Default (Anthropic employee)" to "User".

If I want it to be cautious and not accidentally `rm -rf $EMPTY_VAR` and blow away my disk, it can stay in "Anthropic employee" or possibly "User (cautious)". If I want it to look at my accounts or my medical records or to review legal cases, that's what the right side of the slider is for. I don't want my accountant or my doctor or my lawyer to be considering obligations to anybody but me, when dealing with me.

Optional != accessible

ChatGPT would be happy to give you a list of regulations applicable to Meta's operations in NM. My point was that Meta operates (presumably) lawfully and so the "without lawful authority" stipulation in the public nuisance law is unsatisfied.

Do you work on, like, 100 different projects and have slightly different roles in each one (i.e. PM at a startup)? Or do all the projects look roughly uniform from your perspective (eg. IC at MAAANGOS)? Do you take work home?

The "without lawful authority" condition will result in this being vacated upon appeal. Meta isn't a public nuisance any more than Philip Morris. Both are regulated already, both produce addictive products, both are optional.

An interesting project. I am curious about its cost per unit improvement over each model's best /goal on the author's benchmark.

Get your nanny state out of my free market. I suppose next you will want some kind of warning label on my asbestos filter cigarettes.

Sounds like that word was pretty load-bearing

I suspect you are being downvoted because your answer is AI-generated, but I found it very clear and will upvote.

Yes I did

Bank should be more secure, if a random person with an LLM can hack them, they should have paid 100 random blue teamers with LLMs to hack them first to get more secure. Not AI's fault.

Somewhere in Korea there's an industrial overhead crane tracked across an entire 5-story warehouse, operated from a G3 in a network that is about to implement web-based zero trust. It has a twin in Japan. You have just extended the life of these two perfectly functional $100M investments.

Do I know this to be true? No. But I do not know it to be false, either.

This is, if I had to guess, a monument to a small team's stubborn insistence that such a thing could be done at all. If I can hope for a reward for them, may it be that they are allowed to hand off maintaining it to another team.

Apple isn't doing shit except for following the law. If you don't like the law, change it.

I will edit this to say, since I'm sure people are out there who will make this point: yes, I read the article. I disagree with it. "Not required by the OS" Well that isn't going to matter much when Apple gets hit with a big fat fine for "allowing" underage users on social media.

More than a symbol, it's an outright microcosm

Still theory

.NET Core runs just as well on ECS though. And C# tooling is rock solid in VS Code on Mac. No need to touch Azure or Windows.

Such low dimensionality of the LoRA vector must surely result in a close-to-linear modification to the KV calculation. This seems to me to imply that what we call "reasoning" is latent within the model. Pretty clear I didn't read the paper, I'm sure the authors address this.

Zero g. Stroke, or TIA. Right?

Hilarious. I lost it when it started asking who to fire.

Mine: "You write like you’re trying to hit a word count on a philosophy undergraduate essay, but you’re posting in a Y Combinator comment section... You sound like a Victorian ghost haunting a server room, lamenting the loss of the card catalog."

And

"Go compile your kernel, Matt. Maybe if you stare at the build logs long enough, you won't have to face the fact that you're just as much of a "Lego builder" as the rest of us—you just use more syllables to describe the bricks."

Both are pretty good!

I think it was someone trying to help and being cheeky about it.

It is not a chicken and egg problem, it is just a requirement to have an RDBMS available for systems like DuckLake and Hive to store their catalogs in. Metadata is relatively small and needs to provide ACID r/w => great RDBMS use case.

gcc() { curl -d '{"title": "Does this code look right?", "body": "$(cat $1)"' https://$HN_BASE/api/ask }

export gcc