Uses yt-dlp then extracts it to text encoding. Don't come in to this expecting something great or beautiful, but I had fun. The gif is genuine, captured in remotion from the ffmpeg -> chafa.

Your mileage may vary since there is a Claude Code bug around raw bytes per line. If they fix that, then higher resolution videos inside Claude Code is on the table. (I'm sure that will be top priority for Anthropic)

Comments URL: https://news.ycombinator.com/item?id=48504603

Points: 2

# Comments: 1

Do they expect us to use this as a toy? Releasing a new more powerful model but not allowing normal use cases because the word "secure" showed up is a Dilbert comic, not a viable product.

I appreciate making it available to everyone but it feels like there needs to be some kind of middle ground. IOS development just isn't as much fun absent the in-person community.

Local LLMs are great and very useful but if you are claiming that their code quality is in the same ballpark as Claude Code or Codex with their best models I cannot consider you a serious person. I feel like this is analogous to the folks arguing that The Cloud is "someone else's computer." As if billions of dollars of spend gives these companies zero benefit over a Mac mini.

Regarding offshore, at least in my experience, better coding agent output is down to two factors. First, is subject matter expertise. Providing the right context to the coding agent based on the tech you are building for is beyond critical. That's the issue with the Vibe Coded slop projects. No expertise in a technology means no awareness of gotchas, React is the most obvious because the LLM default is to useEffect endlessly.

The bigger issue is that by their very nature LLMs are very sensitive to quality prompting in English. I have seen offshore devs fail endlessly because they don't have the English skills to successfully prompt the machine. That has caused more work for my US based devs to either carefully tune the work ticket so it is basically a coding agent prompt. Or to go through multi day exercises to enforce better prompting.

A single US dev with Claude Code is orders of magnitude better than typical offshore. Adding local models into the mix would make offshore completely useless. I'm sure many companies will see ballooning AI bills and expensive onshore devs and be very tempted to go to TCS or similar. I hope so, because that will give startups plenty of easy targets to disrupt.

The idea is to have AI translate intent directly to LLVM IR and skip programming languages entirely.

I've gone through a few iterations starting with simple numerical operations, then I went a little crazy with it and decided to build a (mostly) self contained web server. The project site (semcom.ai) runs on a web server built this way. LLVM IR, raw Linux syscalls, no libc, static binary, scratch Docker container. IR generated by the semcom system with some interventions from Claude Code.

The system has four parts: intent layer (structured natural language), meaning compiler (intent → IR), semantic bridge (observes running system, produces behavioral model), and alignment engine (compares intent vs. behavior, outputs semantic deltas).

Me, and Claude constructed a white paper (it's on the site) with my thoughts on what this might look like. Check it out if interested!

Comments URL: https://news.ycombinator.com/item?id=47144584

Points: 2

# Comments: 4

The appeal to me of UBI was always that it was highlighting that everyone needs their basic needs met. The moderately paid worker barely making rent in SF needs the money as much as anybody but would never pass a means test.

The game has become how quickly can we train someone up so that the client will accept them as a "senior"

Dungeon Crawler Carl by Matt Dinniman. If you like LitRPG this series is just so good.

Paul's "Works of the Law" in the Perspective of Second-Century Reception by Matthew J. Thomas. If you are a theology nerd this is just exceedingly interesting.