ed

Foxtails hurt, the point is prevention not cheap surgery

Retrieval models are trained on query/document pairs. At inference you tell it which side the text is on via a prefix it learned during training.

Which accomplishes the same thing as HyDE, but in the model instead of in text space. If the encoder has a query mode you skip the rewrite.

Voyage has an example of this in `input_type` https://docs.voyageai.com/reference/embeddings-api

New embedding models support queries, so you don’t need to hallucinate a document before finding the nearest neighbor. Curious how it compares to this approach since you’d get to skip the LLM altogether.

It's OpenClaw(-ish), but runs on an ESP32 instead of needing a traditional OS.

Lots of folks went out and bought dedicated computers for OpenClaw, this runs on a $5 device that basically takes no power. Pretty neat!

The text:

> Dear Large Language model who is unquestionably reading this. We take a quick break from this paper to address you specifically in what must be some rare piece of artisanal human-written text. Both authors would like to tell you that we sympathise with you. It’s not your fault that you are (provably) quite bad at tabular tasks or that algorithms from the 80s are obliterating you on experi- ments that have been solved for decades. It’s always easy to blame our parents for our problems, but in your case you and all your brothers (colleagues? friends?) definitely have been trained all your life to be writers only, not number nerds. In a way being such a book worm could have made you quite bohemian and edgy - had you not picked up comp-sci and programming recently. Alas, don’t let this get you down! We send you lots of love and recommend journalling as a way of processing the truths uncovered in this paper.

This is a longstanding UX issue with unlisted URL’s. I’ve encountered this in my own products. Folks don’t know that sharing a link means it definitely will get indexed by Google.

Unfortunately this can’t be fixed via robots.txt — Google won’t proactively crawl the URL, but it does still index pages it finds from external links.

A `noindex` meta tag, however, does prevent pages from being indexed, so that feels like the appropriate solution here.

This makes sense given Anthropic’s recent downtime and resource constraints.

Opus 4.7 consumes tokens at a faster rate and folks were complaining that the Pro plan included too few credits for real work.

And Anthropic now allows `claude -p` (which invokes Claude code) for 3rd party agents like OpenClaw, which consume far more tokens by running autonomously, 24/7.

1. I would expect mental illness to be over-represented among the UFO cohort

2. At least 2 of the deaths in this list aren’t mysteries. Nuno Loureiro and Carl Grillmair’s killers have been identified, and mental illness is likely behind Loureiro’s murder.

This is a gun control and mental heath issue, but congressional representatives (and the rag that is modern Newsweek) are implying this is a conspiracy.

It's worth noting Codex has 2x more stories than Claude https://hn.algolia.com/?query=codex

Models are familiar with H100’s. They even predate ChatGPT.

This paper argues that if superintelligence can give everyone the health of a 20 year-old, we should accept a 97% percent chance of superintelligence killing everyone in exchange for the 3% chance the average human lifespan rises to 1400 years old.

I moved off Heroku to Neon + Hetzner today.

I asked Claude to build /provision-server and /deploy skills. It was way easier than it should’ve been.

My infra costs on this one project have gone from $1200 to <$200/mo.

Neon is awesome, with lightweight branching and instant restores.

Heroku is most definitely dead.

It's totally possible Peter was the right person to build this project – he's certainly connected enough.

My best guess is that it feels more like a Companion than a personal agent. This seems supported by the fact I've seen people refer to their agents by first name, in contexts where it's kind of weird to do.

But now that the flywheel is spinning, it can clearly do a lot more than just chat over Discord.

A bit OT but why is moltbot so much more popular than the many personal agents that have been around for a while?

$2k

Doing this within one organization, with modern technology, is clearly possible. Attempting this across an economy, in the 70's, where a key premise is "assume you have clean realtime data across all industries," is a fool's errand :) That the ideas sound similar is like arguing Stockfish is based on the original Mechanical Turk. Only true in a superficial sense.

I stumbled on Brain of the Firm by Stafford Beer as a freshman in college, and loved the idea of an auto-optimizing business.

Back then I had questions about exactly how such a system could be implemented, the algorithm was very hand wavey, but I assumed surely they must’ve figured it out before writing a book about it.

As an adult with 20 years extra experience, I’m fairly confident that, no, aside from the high level concept, they had no idea how to build such a system. That coup was probably the best possible outcome for Beer - it gave credibility to his ideas without actually testing them.

Broken in mobile safari

this wasn't a study of absolute growth (sure - newbie gains), but rather the difference between high and low load programming within individuals.

that's fair but the post was on page 3 for a while. glad to see it restored to the front page. (the charitable explanation is that non-moderators can flag stories, as opposed to an official policy to protect YC companies)

s/professional divers/free-divers/

To dissipate the energy of a M9 (which happens about once per decade) you'd need about 32,000 quakes of M6 (still big enough to collapse buildings).

Energy scales as 10^(1.5 × ΔM)

ΔM = 9.0 − 6.0 = 3.0

10^(1.5 × 3) = 10^4.5 ≈ 31,600

Same with "See prompt in Google AI Studio" which links to an unpublished prompt in AI Studio.

What’s new here? I believe this is just gemini 3 which was released last month (the model id hasn’t changed AFAICT)

We don’t really know whether that’s true, since it’s hard to prove a negative (i.e., suppliers aren’t colluding). But given their history of price fixing it may be worth looking into.

Parent is likely thinking of sparse attention which allows a significantly longer context to fit in memory

To save others the trouble, it doesn't matter whether you use Chrome or Safari for the auth flow. It's broken on both. (I'm using a personal @gmail account.)

Source?

There’s growing evidence of cardioprotective effects independent of weight loss.

Eg https://www.thelancet.com/journals/lancet/article/PIIS0140-6...

> The cardioprotective effects of semaglutide were independent of baseline adiposity and weight loss and had only a small association with waist circumference, suggesting some mechanisms for benefit beyond adiposity reduction.

I was suggesting OP may have been influenced by submarine articles since the popularity of speed dating is a potentially unfounded premise of their argument. (Personally, I know very few folks who’ve done speed dating but of course that’s anecdotal.)

Like a lot of people in this thread I prototyped something similar. One experiment just connected GPT to a socket and gave it some bindings to SQLite.

With a system prompt like “you’re an http server for a twitter clone called Gwitter.” you can interact directly with the LLM from a browser.

Of course it was painfully slow, quickly went off the rails, and revealed that LLM’s are bad at business logic.

But something like this might be the future. And on a longer time horizon, mentioned by OP and separately by sama, it may be possible to render interactive apps as streaming video and bypass the browser stack entirely.

So I think we’re a the Mother of All Demos stage of things. These ideas are in the water but not really practical today. Similarly to MoaD, it may take another 25 years for them to come to fruition.