theowaway213456

But how else am I supposed to know when we've reached AGI, until I see an absolutely flawless pelican?

All of the pelicans so far have had really weird flaws / quirks so I am always a little interested to see how well these models perform at this task, since I've seen all the past pelicans and have some anchoring.

Seeing a truly flawless pelican would tell me that the model has true visual reasoning capabilities as well as good taste.

You can also build your own harness for Claude, right? I think that's not against their ToS?

Not saying it's ideal, just saying it's an option

This sort of take is so tired and boring, and frankly has zero grounding in reality.

"LLMs will never <X>" is constantly being disproven every time they scale up to the next 10X and apply architectural improvements.

Their internal representations are so cryptic and complex that even the top AI researchers don't really know how they work or what their limits are. No one is going to take you seriously as a rando HN user if you're claiming to know better than them.

This seems like more competition for Cerebras? Am I understanding correctly?

Obviously the approach you mentioned has the downside of two server round-trips being necessary while the QUERY request only requires a single round-trip. Not to mention the two-request approach adds more complexity to both clients and servers, as it mandates that both the client and server have to physically create and manage those resources.

Every time you write a test that handles some data, you write an assertion about how much data is handled?

Come on, this is such an easy thing to forget to test. Don't act like there is some magical testing strategy that would have caught this

TL;DR don't have your agent write skills using only its latent knowledge, otherwise you may as well not use a skill in the first place and let it summon that latent knowledge on the fly.

Not sure if this take is correct though. I suspect self-generated skills help the agent avoid having to "decompress" its latent knowledge, which might save tokens? idk, I am not an expert

Claiming that something is someone's "entire personality" is extremely reductive and definitely hostile

Yep that's the first time I noticed it. The pocket was also numbered for me as well

Not a single mention of prompt caching in this article, which is a massive benefit of append-only context.

Certain entities seem to be displayed as numbers for me, like "You received a 6" etc when getting my first potion. Anyone else seeing this bug?

This implementation is better than stdlib's implementation in my opinion, since it respects context:

https://github.com/janos/singleflight

Old news - both of these vulnerabilities made top headlines on HN. This post isn't contributing much.

(Tangentially, the title of this article is egregiously difficult to parse.)

It's very hard to use on mobile. When clicking a graph node, something happens kind of far down in the viewport, which I didn't notice at first. Also clicking "Open post" doesn't seem to work

This feels like a suboptimal solution to me because I personally like to keep comments in "unresolved" state so that they remain visible and other folks can weigh in on them if they want, but in a way that doesn't block the PR. Basically I wish that GitHub would either separate the "collapsed" and "resolved" concepts, or add this "approve without merging" button.

Yup. No matter how much you tell it to keep things simple, modular, crisp, whatever, it generates tons of garbage much too often.

lol after reading your first sentence I literally thought to myself "this sounds like the type of person who never closes their browser tabs"

In my experience, this mistake happens all the time for native English speakers born in the US.

If we're allowed to make modifications here then it should really be lose => looze and loose => luce

I 100% understand why you are using a subscription-based model. It makes sense, and I agree it's the most honest model given that you have to continually support it and you don't want to have to either over-promise on extended support, and offer refunds if you can't fulfill that promise.

I just hate managing subscriptions.

If you gave me the option to require manual subscription renewal, rather than auto-renewal, I would 100% buy this right now. Basically allow me to purchase for 1 year then click a button to confirm that I'm still getting value out of the product. If I don't click that button then you should assume I'm no longer interested and cancel my subscription.

(I don't like using my mac but sometimes I have to use it for work, and I wish I had this.)

Sorry, my attention started drifting when you said "epistolary."

In other words, it requires a tremendous amount of effort to fully communicate your tastes to the AI. Not everybody wants to expend the time or mental effort doing this! (Once we have more direct brain/computer interfaces, this effort will go down, but I expect it will not be eliminated fully)

Feels like AI generated market research. There must be a single company/person putting out these posts because I see a ton of posts like this following a similar structure

Anyone else feel like this post is AI generated? I've been seeing a lot of posts like this on reddit programming subs and now it's happening on HN too... The pattern is:

- some generic context, containing a bunch of em dashes of course, with a vague background that isn't tied to a specific incident

- a claim like "most ___ I've talked to about this" within the first few sentences.

- ends with a series of questions as though they're soliciting more input

I feel like these all must be coming from a single person who's trying to automate market research and sell the data or something

Five years ago, I would've loved this. I love the simplicity and power of good old Make. And I obsess over my workstation's configuration. I used to have a massive bash script I would use to reprovision my workstation after every clean upgrade of Ubuntu.

But these days, I just tell codex to install things for me. I basically use it as a universal package manager. It's more reliable honestly than trying to keep up to date with "what's the current recommended way to install this package?"

I also have it keep a list of packages I have installed, which is synced to GitHub every time the list changes.

This is my impression as well. ClickHouse has tons of useful features built in that seem like they'd work well here. Though the documentation about those features has been very scattered and hard to find in my experience

table names, columns, virtual columns

This sounds solvable with clickhouse views?

automatic joins

Is this also not solvable with views? Also, clickhouse heavily discourages joins so I wonder how often this winds up being beneficial? For us, we only ever join against tenant metadata (i.e. resolving ID to name)

query optimization

This sounds potentially interesting - clickhouse's query optimizer is not great IME, but it's definitely getting better

Are there any good, non-tiling window managers that support tabs? (I struggle with tiling ones like i3 because I am a small-brained mouse user)

This is a pretty cool idea. Kind of a neat distribution hack if all you have is a TUI (and not a full GUI). Curious whether you know of any success stories yet

Thumbnails are not even needed - they can just use a plain old img tag in the readme with an external src.