Every MLB game I've ever been to has had that happen lol. 3 'first pitches' and then the actual start of the game happens with absolutely zero fanfair, so it's very easy to miss. In general they announce very little audibly about the actual game, it's a very different experience from watching on TV.

Apparently ~75% of the positions in the lichess database (as of 6 years ago) have only been seen once ever. Average game length is 30-40 moves, so for the completely average player it would be like 10+ moves I suppose. The stronger the players the longer it will take: I found some comments suggesting 20+ for high level players.

It depends totally on the opening. You can be out of book and database far quicker than that for offbeat stuff, or in book far longer for popular openings.

Another distinction needs to be made between positions seen and positions played. Almost every viable position will have been seen in preparation well beyond 10 moves. But seeing them on the board is rarer.

I don't think the math is correct here. The 25% of positions that have been seen more than once represent more than 25% of the occurrences. Even if all of them would be seen only twice, you should already see them in 40% occurences.

Some agent had already grabbed the gun and was running off with it when the 1st shot was fired. It's not possible to argue he was threatening them with it or something then, there was 6+ feet and several people inbetween him and the gun. https://bsky.app/profile/bellingcat.com/post/3md7banbjks2x

The phrase "ai teammate" feels popularized as a marketing strategy to position individual agents as comparable in value to a human worker. When I think about how they are actually used however, it seems like an incredibly unproductive framing. An agent is a computer program. You can copy them 100 times on the spot if you find the need. You can modify, delete, upgrade, or replace them instantly. You can keep them up 24/7 or run them only on user request. The "teammate" framing obscures all the software-type things you can do with it. Imagine Excel processes or crawler bot instances being given little human names and pictures; being slotted into an org chart. Absolutely the wrong way of thinking about it.

Yea they do a good job in the article calling this out

> A teammate is someone who shares your objectives. You’re playing the same game, working toward the same goal, invested in each other’s success.

LLMs cannot be a teammate because they are not playing the same game nor working towards the same goals.

it's also meant to endear them to the people subject to the replacement trial

if successful: the rest will be gone too

I don't understand why it has been acceptable to not upload a tarball of your data with the paper in the internet age. Maybe the Asset4 database is only available with license and they can't publish too much. However, the key concern with the method is a pairwise matching of companies which is an invention of the paper authors and should be totally clear to publish. The number of stories I've heard from people forensically investigating PDF plots to uncover key data from a paper is absurd.

Of course doing so is not free and it takes time. A paper represents at least months of work in data collection, analysis, writing, and editing though. A tarball seems like a relatively small amount of effort to provide an huge increase in confidence for the result.

This. I did my dissertation in the early '90s, so very early days of the internet. All of my data and code was online.

IMHO this should be expected for any, literally any publication. If you have secrets, or proprietary information, fine - but then, you don't get to publish.

Yeah or like a themed university library. I'm sure this kind of stuff is way cheaper in China, but $5 for a desk and food for a day seems good. Gets you out of the house and interacting with people.

Couldn't be more clearly a hallucination. If you think about the 'data' it's returned in terms of something actually stored in a database it makes no sense. Why do all the dict values contain extensive prose explanations of their meanings? Why is list data being stored in grammatical lists instead of arrays? Why are there so many categories that are minor paraphrases of each other or differ only by parentheticals?

If this was real we'd have learned about it from Twitter engineers crashing out trying to maintain that awful 'schema'. There's a long history of LLMs hallucinating their own internals.

The mental model is usually what I want most (and often find missing) from onboarding. If you give me the model and vocabulary I can find the buttons myself. If you teach me 1 workflow I'm probably helpless at doing anything else.

100%. The goal shifted from "show where buttons are" to "explain what this app is for." Once parents get the player vs admin concept, everything else clicks.

I suggest the Goode Homolosine, which thinks so little of Greenland that it bisects it.

Also, all the biggest software companies run network services. You could download your own search engine forever, but it doesn't mean anything without the Google data centers and crawlers. Likewise there's a million open source twitter clones, but without the users they have no real utility to the average person. The era of shrink wrap software has been over for a long time.