nileshtrivedi

You might like SWE-WebDevBench which tries to do this comprehensive evals for webapp development. https://webdevbench.com/

> proof assistants, traditionally, don't use our classic two's complement integers packed into words in our memory, they use Peano numbers

Why can't we just prove theorems about the standard two's complement integers, instead of Nat?

I have now switched to pglite for prototyping, because it lets me use all the postgres features.

This looks useful.

Hope would you say it compares with pgqueuer?

ETA on launching in India?

I don't think it does.

The default Bluesky frontend uses bsky dot app URL when you use the "Copy link to post". Now if one day, you lose trust in this server and switch your PDS, this link continue working depends on this very non-trusted server. If this server is profit-seeking, it can break such links.

An extension or another app is not the solution, and neither is the new at:// URI format, because what matters is the relationship the default server sets up with its majority of users. Most bsky users will lose their traffic to their own posts and therefore will be locked-in, cementing this one server to be the dominant one in all perpetuity. We will therefore get all the patterns of monopolistic abuse that we have seen elsewhere.

Not only is the bluesky network highly centralized right now, its UI is designed to perpetually lock users into the main bluesky server. Even if you use your own identity, when sharing the URLs to the posts via the UI, the URL defaults to bsky dot app domain, which will break if the author ever moves to a second server.

MOOCs are still going strong at India's NPTEL , which started distributing university-level video courses for engg in 2008:

- 5m subscribers, almost 2B views on youtube

- 30m enrollments

- 600+ courses every semester in 22 disciplines

Anyone from the world can signup. Proctored exams are optional and cost about $11 per course. Not taking VC funding and setting up local chapters for supporting students seems to have worked out well for them.

Website: https://nptel.ac.in/

ACM report about this from November 2022: https://dl.acm.org/doi/pdf/10.1145/3550473

Former-director of IIT Madras has talked about how NPTEL came together: https://www.youtube.com/watch?v=LV-QoGegFLY

Question since you are here, how long before tool-calling is enabled for Gemma3 models?

I've been thinking on these lines as well. What if computational DAGs spanned boundaries of various machines: the DOM or the UI to client memory to client persisted storage or server memory to database to other connected clients and so on? We seem to be solving the same problems again and again, in various places - mostly in the same way.

For consumers in India, we have a crowdsourced wiki for such products: https://www.isfixable.com/

Types only go far. Most of the semantics of a function's behavior is in its name, documentation and tests.

I think it is time to invent a unifying framework for Types, Tests and Evals: https://nilesh.trivedi.link/thoughts/we-need-a-formal-theory...

I find it annoying when people share direct URLs to PDFs at arxiv.org instead of the the abstract. However, if I were to use a redirect extension like this one, then my own click on the PDF link would get redirected back to the abstract page. To handle this, the redirect rule needs to consider the referer page too, with a possible exception if the referer is the same site as the destination.

That remains to be seen. Manus, a standard agent built with Claude 3.7, outperforms o3 agentic model on the GAIA benchmark.

Something like this? https://youtu.be/QeI61YmSnTA

What "other web companies"? Everyone has an app or an app store.

It should be possible to use syncthing to achieve sync across multiple devices, without having a third-party server - even if untrusted, right?

Recently, I started building a crowdsourced wiki for repairability of consumer hardware products in India: https://isfixable.com/

Louis Rossmann is putting together a Consumer Action Task Force. If people care, now would be a good time to show it.

It started as a Kickstarter campaign: https://www.kickstarter.com/projects/203null/matrix-grid-con...

Here is the documentation for the OS: https://matrix.203.io/docs/Developer/MatrixOSBasics

Github: https://github.com/203-Systems/MatrixOS

Also ActivePieces from YC

The biggest pain point with Ruby these days is: there is almost no libraries for modern number-crunching techniques of any kind. Only python libraries come out of machine learning community / academia.

We literally had a chemistry Nobel Prize winner crediting chess for making him curious about thinking and intelligence and ultimately to find DeepMind.

> "open source chatgpt desktop assistant" app with swappable llms

THIS. OpenAI, Claude and Perplexity - all have their individual desktop apps now. It's time we got a generic, model-agnostic one.

On the other hand, we will likely see the underlying platforms (operating systems and browsers) ship with their own interface. I just hope they turn out to be model-agnostic.

Relevant: PowerBI is a Human Rights Violation: https://ludic.mataroa.blog/blog/powerbi-is-a-human-rights-vi...

What happens if the account belongs to the parent but all the content is by the child?

Writing digitally is cheaper but that's exactly why distributing or getting reach is not cheap at all. You still need cost and proof of work in getting noticed by algorithms, as well as people who usually set trends. In fact, the lower cost of production means that more niche things get written than there would have been a market for.

Exactly. Most of the author's complaints can be answered with: "Use decent software. And make copies."

And I found it disappointing that the author did no attempt to recognize that digital #reading is what enables himself to reach people at all? Where is the accounting for accessibility and reach?

> I’ve noticed my near sight dropping

You mean your vision got worse after using quest2?

Doesn't the potential for abuse reduce when content is linked through user's own domain rather than a particular appview like bsky? Bsky already supports a user's domain ALIASed to redirect.bsky.com: https://bsky.app/profile/jacob.gold/post/3kh6rnpdzmp2v