Luucas40

Some notes on how it is built, since the page mostly shows the output.

It is one index of the AI research literature, assembled mainly from arXiv, OpenAlex, ORCID, DBLP and the pages researchers publish about themselves. 496,295 papers, 790,725 people, 553,497 of them with a country. Every figure is a count on that index and carries the number of profiles it stands on. It is updated at least weekly.

The page is a single HTML file. The data ships inside it as JSON, about 100 KB of the 165 KB, so there is no API call, no framework, no chart library and no tracker. Nothing on it is typed by hand: a scheduled job recounts everything and regenerates the page and the published aggregates in the same pass. The aggregates are CC BY.

Author disambiguation has been an important part of building this. Merging records that describe the same person across four sources, on shared identifiers where they exist, and on the concordance of names, coauthors and institutions where they do not. That is where the errors that remain are hiding.

What it does not measure, said now rather than defended later: publication, not all of research, so people who ship models without publishing are missing. A country is the country of the institution, as I found it way more interesting to understand the strength of an ecosystem. Subjects come from the OpenAlex taxonomy, whose buckets are uneven, which is why a label as wide as Topic Modeling sits at the top.

I work on recruiting tools, so the index behind the map is also what I work with in my day to day life. The map publishes counts only, never an individual record. Happy to go into any part of the pipeline.

Not letting your llm write is good advice and letting him spot problems also is if you want good writing.

For anything beyond "good", we're still on our own and I love that for us.

As an institution, if: - you don’t allow people to say something - people are not happy with they way you govern/use your authority

Then these people will give 100x more credit to the ideas you try to censor, just because they are associated with “being against you”.

Twitter and what has now been rebranded Woke 1.0 is another example that backfired massively.

I also completely agree with this part. However, writing is its own muscle that needs to be trained and sometimes reading more can also be procrastinating in disguise. What helped me a lot has been: Read a lot because you love it Force yourself to write everyday, even if it’s for 5 minutes or you might postpone it forever

We've seen a switch from subscriptions to results-based pricing as AI usage would be eating all of the margin (intercom has been one of the first widely successful examples).

This is another possibility, but the amount of complexity induced for customers + staffing costs (comps for ServiceNow specialists are already extremely high) will probably push people to just buy AI access on top of the subscription instead.

I don't think AI-screening is as big a problem as people make it to be, but if I'm wrong then I guess this kind of tool makes sense.

Job-seekers are notoriously difficult to monetize for two reasons: - Often lack of money to spend as it's a period where people are careful - Churn, once they've found a job they'll not use your services anymore, so acquisition must be top-notch

Good luck it's not an easy path that you've chosen, which makes it interesting!

X is such an interesting piece of history as it was both the most interesting place in the world for years, and leaning so democrat that saying anything remotely conservative meant social suicide.

Now since Elon took over, the feed has gone very conservative also because any kind of right-wing content can't really be expressed on other social medias, and obviously it was Elon's plan.

Last point is that the old twitter was so much better at finding interesting communities for you based on your interests / jobs etc. when the new one pushes way too much political content.

Hey, curious how you tracked the %gain? They seem super high, which is entirely possible for people who didn't take they nutrition as seriously before if you managed to get them fully on board with the app.

And something I'm totally unfamiliar with, how does someone "gut train"?

It's an interesting angle for a training app as this is definitely a recurring issue in the endurance world, good luck to you!

As with a lot of these views, I always struggle with how it would apply to the real world. Organisations become important because of their singularity, what they do that others don't. It can be positive or negative, but it is the case (for the government, companies, charities, families etc.). If AI is now "in charge", then this becomes replicable and the edge moves somewhere else.

Best example of this is indie hackers where coding has become a less and less important part of the job, and the difficulty just moved to distribution.

Did you start with this or was it AI just starting random and then the website will create itself over time based on A/B tests and feedback?

We can already tell when people write with AI, let's not worry! But I guess people are worried the watermark stays even if they completely change what AI wrote