I encounter Codeberg as a forge first, then have to dig in to discover that it's a highly opinionated service with a different business model than most similar entities.

Microsoft and Amazon both provide web snapshot services that purport to give you a kind of agent-first internet archive. You can approximate something like that using common crawl, but it's a huge amount of data. Downloading the internet is impossible or a bad idea for almost everyone.

You could just as easily point out that subsistence farmers exist. That’s great, but we still need industrial agriculture to feed everyone. No single human fully understands that domain either… even ignoring the chemistry, biology and meteorology, those farms use digitally controlled tools! That means all of computer science is embedded in farming.

It’s great to pursue broad knowledge. It’s important to understand your dependencies and respect the abstractions you build on. It’s pure hubris to think an individual auteur can challenge a multi-billion person civilization.

I convinced the customer to accept the delivery by pointing out that (1) our app had zero lines of ocaml and (2) the feature had been implemented in the ocaml driver since the CVE was issued.

This cycle happens when you have a well defined use case with high volume. It is the far opposite end of the spectrum from the general purpose intelligence on tap that frontier AI models purport to deliver.

I think focused fine tunes and big general models will coexist. Ideally with smart routing and caching to use small, specialized and local options when appropriate.

Skylight is making some progress as a family hub of operations. I could see value in something like this in that ecosystem, especially if there was a way to control data sharing.

For example, could this run on my Mac, review all the data, then create a memory/fact feed that I can review and control?

   - They explicitly require all hosted software to have a FLOSS license

   - Private repos are only allowed in support of Public FLOSS repos, with a scary ambiguous note about "really small and personal stuff"

   - Projects with heavy use of coding AI are explicitly forbidden

For reference, here's the parts of the TOS that define the above:

> 1. Public repository content shall be made available under a copyright licence [...] referred to as ‘licences for free and open source software, free and open source hardware, or free cultural works’. Examples of such licences include those created or approved for free software projects by the Free Software Foundation, for open source projects by the Open Source Initiative, and for free cultural works by Creative Commons. Note, however, that the Creative Commons licences with Non-Commercial (NC) or No Derivatives (ND) restrictions are not licences for free cultural works and thus do not fulfil the licensing requirement.

> 2. Private repositories are only allowed for things required for FLOSS projects, like storing secrets, team-internal discussions or hiding projects from the public until they're ready for usage and/or contribution. They are also allowed for really small & personal stuff like your journal, config files, ideas or notes, but explicitly not as a personal cloud or media storage.

> 7. You must not share projects that mostly consist of code written by "generative AI"-tools (including services such as Claude, OpenAI Codex). Such projects having an unclear copyright status (see requirements § 2 (1) 1 and § 2 (1) 3) and furthermore have little safeguards to ensure that they do not include harmful code (c.f. § 2 (1) 5).

[1] https://codeberg.org/Codeberg/org/src/branch/main/TermsOfUse...

> But again, some combinations might be just harder to draw than others.

> To account for that, I fit a fixed-effects regression on all 1,008 images: score ~ lab + animal × vehicle, plus per-lab interaction terms for pelican, bicycle, and the pelican-bicycle cell, with robust standard errors. The animal × vehicle terms absorb the inherent difficulty of all 48 combinations. The interactions measure each lab’s benchmark-specific boost relative to the average lab, with confidence intervals.

> If these tools demand expertise, yet the tools can actively circumvent the friction that cultivates expertise, then what is the path for one to become an expert so they can effectively use these tools?

Comments URL: https://news.ycombinator.com/item?id=48893308

Points: 10

# Comments: 1

https://nostarch.com/download/samples/automate-excel-with-py...

There’s probably some ideal blend of locking regions, in-excel validation, in-python validation, and clean separation of human inputs and machine outputs. Has anyone figured out what that is already?

“It’s now difficult to prepare for the world three months from now if the median LLM-produced computer science paper is better than that produced by the median grad student.”

https://news.cornell.edu/stories/2026/06/digital-research-re...