Joined April 2008
- Tweets 1,104
- Following 261
- Followers 505
- Likes 273
120 Photos and videos
14h
1/
Google released OKF this week, billing it as a vendor-neutral standard for the knowledge AI agents need: table definitions, metric formulas, runbooks. The gap between that framing and what version 0.1 actually specifies is the most interesting thing about the release. 🧵
1
21
14h
7/
If you are evaluating OKF for your own stack, the practical read is this: it standardizes the package, not the vocabulary. You get a reliable way to find and read context across systems. You do not yet get shared semantics for what that context means. Registered types, link vocabularies, and declared profiles are the work that later versions will need to do before "interoperable" means what the announcement implies.
1
5
14h
8/
The right first move may well be to standardize the container before arguing about the vocabulary. But naming a thing "interoperable knowledge" sets an expectation that v0.1 does not yet meet, and the gap between what OKF delivers today and what that label promises is exactly where the interesting design problems are waiting.
Read the full piece here:
medium.com/@marc.bara.iniest…
14
Jun 11
1/
Six million GitHub stars are suspected fake, and you can buy them for three cents each. The star count is the most visible number on a repository, it feeds GitHub's search ranking and trending page, VCs scrape it to source deals, engineers cite it to justify a dependency, and it ends up on résumés. I started looking into this after noticing a repository with 157,000 stars that almost nobody was actually using. 🧵
1
1
1
47
Jun 11
8/
When both checks point the same direction, scarce forks and empty profiles, the star count has stopped being a signal. The CMU researchers recommend GitHub replace the raw star count with a weighted metric, because a count that treats every account's action as equal is trivial to inflate. The loop sustains itself regardless: investors use stars to source deals, founders inflate stars, investors see traction that is not there, more investors adopt star-tracking, more founders inflate.
1
13
Jun 11
9/
A GitHub star is evidence that an account clicked a button, not that anyone is building with the code. When forks, active issues, real stargazer profiles, and release history are missing, the number is decoration.
Read the full analysis here: medium.com/@marc.bara.iniest…
14
Jun 10
1/
For three years, @AnthropicAI's public policy position was transparency: disclose safety procedures, let binding regulation wait until risks took concrete shape. @DarioAmodei just abandoned that position entirely. He now endorses mandatory third-party testing of frontier models, government power to block releases, and a coordinated geopolitical strategy, with a promise of "substantial financial backing" for legislation. A funding promise is not legislation, and no figures are given. But it is more commitment than the usual CEO think piece carries. 🧵
1
14
Jun 10
8/
Strip away the framing and the essay does three concrete things. It proposes a specific, scoped regulatory mechanism with money behind it. It concedes, from inside a frontier lab, that enduring labor displacement is a real possibility. And it rejects, by name, the idea that public concern about AI reflects misunderstanding rather than accurate perception of real risk. You do not need to share @DarioAmodei's timeline to find the document useful. The four risk categories in the testing proposal are a preview of what compliance conversations will look like if anything resembling this passes.
1
12
Jun 10
9/
Whether you read this as genuine conviction or regulatory capture in early formation, the document is worth reading in full before you form that opinion.
Read my full section-by-section breakdown here:
medium.com/@marc.bara.iniest…
9