low_tech_punk

I think the router applies to chatgpt app. The developer APIs expose manual control to select the specific model and level of reasoning.

Thanks for calling this out. Yes, agree. I should revise my understanding of plagiarism.

To be fair, the blog has a "Source: NASA" link near the beginning.

Can confirm that they are rolling out. It's working for me.

That was debunked by Veritasium 13 years ago: https://www.youtube.com/watch?v=aFO4PBolwFg

ah, thanks!

I find it strange that GPT-5 is cheaper than GPT-4.1 in input token and is only slightly more expensive in output token. Is it marketing or actually reflecting the underlying compute resources?

Tried using gpt-5 family with response API and got error "gpt-5 does not exist or you don't have access to it". I guess they are not rolling out in lock step with the live stream and blog article?

The ability to specify a context-free grammar as output constraint? This blows my mind. How do you control the auto regressive sampling to guarantee the correct syntax?

That blog is the student notes from a famous MIT Media Arts & Sciences class called HTMAA. Course website: https://fab.cba.mit.edu/classes/MAS.863/

Lex Fridmen has a podcast with the professor: https://www.youtube.com/watch?v=xF35Udv1DBU

Definitely something telecom is doing. People seems to be ok with their “unlimited” plans that throttle after a usage cap. Actually curious how the economics of those plans work out.

Why not do tiered pricing like OpenAI api? The more you consume, the more discount, but it’s never unlimited or free, so you can prevent abuse without punishing true demand.

I think it's short-term vs long-term discussion.

In the short term, Figma's strategy to focus on existing designer market makes total sense. And Figma, now being a public company, is indeed held captive by the market demands to sell more.

In the long term, I believe some other start ups will figure out the end user programming problem before Figma does. Maybe Figma will acquire those companies rather than building its own?

As a side note, Figma acquired Dynaboard (https://dynaboard.com/blog/figma-acquires-dynaboard) years ago, which I suspect powered their Figma Make. It took them a lot of time to develop it up though. Definitely late to market compared to V0 and loveable.

To make matters worse, there is actually a third thing named "TSX" gaining traction right now:

https://esm.sh/#tsx

I don't think it's accurate to say "without worrying about configuration". The next line is more accurate:

> tsx runs your TypeScript code with modern and sensible defaults, making it user-friendly and especially great for beginners.

You'd still have to worry about config if you want to make adjustment and when that happens, the implicit smart defaults become a friction point.

It might also surprise you with errors when you attempt to bundle the code. It'd be nice to have tsx available at runtime so I can run TypeScript code without worrying about the transpiler

On a tangent, does anyone feel bad about their HN addiction?

Are there other examples? Definitely curious to see more.

The anthropomorphic view of LLM is a much better representation and compression for most types of discussions and communication. A purely mathematical view is accurate but it isn’t productive for the purpose of the general public’s discourse.

I’m thinking a legal systems analogy, at the risk of a lossy domain transfer: the laws are not written as lambda calculus. Why?

And generalizing to social science and humanities, the goal shouldn’t be finding the quantitative truth, but instead understand the social phenomenon using a consensual “language” as determined by the society. And in that case, the anthropomorphic description of the LLM may gain validity and effectiveness as the adoption grows over time.

Wait, what is this?! It's really cool

So if we replace LSP with MCP, the UI might control the code in interesting ways!

It's a lossy compression/decompression problem too.

Probably faster than Starbucks in the US

Perfect example of show, don't tell - the project just shows how AI system design can go wrong.

Pug is an unsung hero in this demo. The template engine made it possible to brute force the UI with 4 levels of nested loop. Impressive!

I don't like vibe coding in general, but it is surprisingly a good tool to reinvent the wheel. Claude Code team allegedly built their own markdown rendering with Claude because the existing renderers don't meet their needs.

Can anyone help me understand why they changed the model naming convention?

BEFORE: claude-3-7-sonnet

AFTER: claude-sonnet-4

Thanks for drawing the connections. I think hacking on PKMS is so satisfying because it combines multiple hard problems: text editor, search engine, graph database, sync engine, CMS; and you can go down a rabbit hole in any of them!

Btw, I can't access https://notes.holeytriangle.com/admin/files/a969c5f6-fa9c-44... is that your private admin link?

I've been doing this for a while. My biggest hurdle was implementing sync between local and backend. I ended up using GitHub as my backend, Chrome extension as frontend, and implemented a simplified version of the Git client in SQLite with lots of triggers. Happy to share it with people. The app is here: https://github.com/osmoscraft/tundra and the sync logic is here: https://github.com/osmoscraft/tundra/blob/master/apps/browse...

I hope this is a reaction to Framer's rapid development in Frame Motion (https://motion.dev/). Webflow and Framer are competing as site builders, so giving the animation library to everyone is like Meta giving out React.

It's time for some severance at the office