caprock

The last couple of pictures, of the eyes and the bird, are really neat. Göbeklitepe continues to provide interesting results.

It's kind of comforting and exciting that we have so much yet to uncover about the past.

The details are not clear with regard to whether or not the chats have been accessed by the FBI:

> We also don’t know if the FBI ever actually got ahold of the Xbox chats in question, or merely got Salamone to confess by making it seem like they did.

From the closing section:

> but we should not have to self-manage our own productivity in imperfect systems laid atop programmer thinking to simply do our work.

And

> And we can’t expect each user to be both a maker and a self-manager, especially with the imperfect tools on the market.

Why can't we expect these things? Lots of life seems to expect this of me.

Project management isn't about tools or even process. It's about doing the hard work of analyzing, defining, and organizing the next batch of work. Then do the work. Rinse and repeat. This gets harder with more people because alignment and communication takes more time and effort.

The best situations I've seen were always because one or two people took the brunt of that thankless work. The tools and process were merely supporting that work.

The answer is dependent on the task(s) to which the embeddings will be applied. For general search in industry, the e5 models are well regarded.

A good place to start is this eval system:

https://huggingface.co/spaces/mteb/leaderboard

RIP. RedBox works as a half decent replacement.

Most company types will have concrete details for defining ownership proportions (controlling interest) and/or assigned roles with broad but clear allocations of responsibility and authority. Hopefully you have similar contracts defining the rules of engagement with your angel investors.

If you're worried about litigation and liability, fall back on what the concrete legal docs say you can do.

If not done yet, step one should probably be updating the company docs to reflect the reality of the founder separation. This of course requires negotiation of an agreement with your cofounder unless you set things up where you have controlling interest already or want to outsource it to a lawyer.

If you clarify the region, company type, and whether the separation was formalized, then folks might be able to provide more specific advice.

There are some:

https://huggingface.co/spaces/mteb/leaderboard

Erring on the side of openness is fine as a principle. That doesn't mean the actions to follow the principle won't cost you time and money. Everything has tradeoffs. As a small SaaS business, I don't think public repos would make it very high on my list of priorities.

A good balance can be open sourcing an internal library or two, so you get some of the benefit without the work needed to polish the mass of your core code.

With regard to recruitment, I think the openness is mostly a benefit to employees, because they can point directly at their work product like a portfolio. So it helps recruitment, but it's about them not about your cool problems.

People are nuanced and complicated. Almost all activities have tradeoffs. So it's intriguing that we sometimes seek to categorize famous people as completely good or bad.

The continuum of Dr. Jekyll and Mr. Hyde lives in all of us.

The biggest initial hurdles I see:

Understanding the process, tools, and little details of local development for the team and product.

Understanding who does what on the team, and how to deal with each personality, role, and processes.

> "It would stymie the government's legitimate and crucial efforts to address threats to the public welfare."

> The State Department reportedly canceled a standing weekly meeting with Facebook officials to discuss the 2024 election.

Is there a line, where this becomes effectively state run media?

Which side is driving the agenda at these weekly meetings?

These are good:

* Eleuther AI

* Yannic Kilcher

* MLC Open Collab

* Open Assistant

It's nice to see more pathways to good tech careers. We need more of this experimentation.

This sounds a lot like the chunking concept from learning theory.

You don't learn it, you practice it. Just start trying, failing, and trying again. You'll build endurance over time.

It's counter intuitive, but the mind and body get better the more you work, exhaust, and challenge them.

Good luck.

PBS Frontline episodes are interesting, and available on YouTube.

Reuters is a decent replacement. Regular news, less slant, and minimal editorials.

https://www.reuters.com/

The corresponding paper:

MTEB: Massive Text Embedding Benchmark

https://arxiv.org/abs/2210.07316

And the repo:

https://github.com/embeddings-benchmark/mteb

You might want to look into what the creator of htmx has to say about this topic. He's here on hn and may even show up for this thread.

Here are the essays:

* https://htmx.org/essays/

The link to the feature: https://you.com/search?q=what+was+the+recent+breakthrough+in...

I believe so, yeah. The dates are derived from a grouping query run directly on my local copy of the hacker news data. It's synced from the official api via firebase, not the algolia api. As best I can tell, the api stores all the stamps as UTC unix time. (today is of course incomplete)

This is tweets-as-news, and of course is not news but instead just thin, speculative commentary. It could just as easily have been written about Lex Fridman's similar, and probably more serious, response to the new ceo tweets.

A more interesting story is the interview of Mr. Beast on the my first million podcast. He talks about about how the "company" operates, and it's very fluid and unstructured. He just focuses on making content and surrounds himself with folks willing to help fill in the gaps.

Thank you for sharing. Do you know how many visits per post were received on average?

And here's the number of comments.

  --------------------------
  :: comments per day
  --------------------------
  [('2022-12-22', 6860),
   ('2022-12-21', 12008),
   ('2022-12-20', 13784),
   ('2022-12-19', 12418),
   ('2022-12-18', 11652),
   ('2022-12-17', 10372),
   ('2022-12-16', 13487),
   ('2022-12-15', 13468),
   ('2022-12-14', 13463),
   ('2022-12-13', 14500),
   ('2022-12-12', 12176),
   ('2022-12-11', 9787),
   ('2022-12-10', 9437),
   ('2022-12-9', 11439),
   ('2022-12-8', 12157),
   ('2022-12-7', 12013),
   ('2022-12-6', 13760),
   ('2022-12-5', 12971),
   ('2022-12-4', 9748),
   ('2022-12-3', 9416),
   ('2022-12-2', 12519),
   ('2022-12-1', 13578),
   ('2022-11-30', 13375),
   ('2022-11-29', 12077),
   ('2022-11-28', 12057),
   ('2022-11-27', 9024),
   ('2022-11-26', 9285),
   ('2022-11-25', 9848),
   ('2022-11-24', 10405),
   ('2022-11-23', 11677),
   ('2022-11-22', 4477)]

Here's the number of stories per day, for the last 30 days or so. These counts don't filter out dead, deleted, or anything else.

  --------------------------
  :: stories per day
  --------------------------
  [('2022-12-22', 774),
   ('2022-12-21', 1261),
   ('2022-12-20', 1306),
   ('2022-12-19', 1175),
   ('2022-12-18', 811),
   ('2022-12-17', 820),
   ('2022-12-16', 1285),
   ('2022-12-15', 1340),
   ('2022-12-14', 1221),
   ('2022-12-13', 1262),
   ('2022-12-12', 1133),
   ('2022-12-11', 851),
   ('2022-12-10', 795),
   ('2022-12-9', 1125),
   ('2022-12-8', 1250),
   ('2022-12-7', 1284),
   ('2022-12-6', 1306),
   ('2022-12-5', 1337),
   ('2022-12-4', 852),
   ('2022-12-3', 830),
   ('2022-12-2', 1245),
   ('2022-12-1', 1323),
   ('2022-11-30', 1481),
   ('2022-11-29', 1324),
   ('2022-11-28', 1348),
   ('2022-11-27', 866),
   ('2022-11-26', 764),
   ('2022-11-25', 922),
   ('2022-11-24', 1022),
   ('2022-11-23', 1177),
   ('2022-11-22', 448)]

    ...a loss function based on the cosine similarity of any given text with statements made by individual HN users.

A. Could you elaborate on the mechanics and particularly the loss function calculation? For example, did you take the top N users by karma, volume, something else? Does the loss get calculated as an average sim against each of the top N, or do you mean pool the top N first, or something else? Or is it one model per user?

B. How do you manage the posting and accounts? Just manually?

gruvbox. for a long time. maybe it's time for a change.

Very intriguing ideas. Is there a relationship here, with the concept of starting an application by thinking more about the data structures and getting those right, before proceeding with the rest of development? Or are you speaking of more abstract mental models?

That's really neat to hear. Can you comment on how replit has managed to foster a culture of fast delivery? Are there any interesting trade offs?