The gauge-reading example here is great, but in reality of course having the system synthesize that Python script, run the CV tasks, come back with the answer etc. is currently quite slow.

Once things go much faster, you can also start to use image generation to have models extrapolate possible futures from photos they take, and then describe them back to themselves and make decisions based on that, loops like this. I think the assumption is that our brains do similar things unconsciously, before we integrate into our conscious conception of mind.

I'm really curious what things we could build if we had 100x or 1000x inference throughput.

I get they're going for the whole "look at big evil Google undermining this underdog" support ticket route, but I think it will backfire in this case.

Comments URL: https://news.ycombinator.com/item?id=47735708

Points: 2

# Comments: 0

Edit: Turns out, sadly, had one: https://geekboards.de/blogs/news/were-closing-our-berlin-sho...

I'm your usual HN-brained copious scifi novel/science non-fic reader, typically.

Right now your options are basically having a GLES renderer that you can restrict to WebGL2 (so no compute shaders, etc. and other things that make desktop OpenGL acceptable for writing a modern renderer) or having to abstract over Vulkan/WebGPU yourself, which are similar but different enough to increase your code complexity considerably.

There's abstractions like wgpu and bgfx you can commit to, and of course you can just use game engine middleware and have it all done for you, but overall things fall short of just being able to "write once, run anywhere" a renderer, sadly.

Some of the "isn't it interesting ..." type coincidences would, as people on this forum would know, be commonplace among the subculture or even just technologists, and often lack the comparison to the overall Cypherpunk corpus - for example: no, studying public-key cryptography in grad school certainly isn't a high-signal differentiating tell for Satoshi-ness.

For some he does provide that though, and they're certainly compelling.

What I like best about the Back attribution is that it totally makes sense in context of my operating model of humans and passes the Occam's Razor test: Still actively involved, interested in the governance, interested in acclaim/prestige, built up wealth masking his other wealth, etc. Ego and "Tell me you're Satoshi without telling me you're Satoshi" written all over it.

As an engineer a substantial amount of my professional effort is spent on preventing them. They aren't acceptable.

Uninformed, but not ignorant and perhaps even interested. I hope your response started with "No, actually, even cooler: ..." and you made a space fan that day.

Not always, no, and it takes investment in good prompting/guardrails/plans/explicit test recipes for sure. I'm still on average better at programming in context than Codex 5.4, even if slower. But in terms of "task complexity I can entrust to a model and not be completely disappointed and annoyed", it scores the best so far. Saves a lot on review/iteration overhead.

It's annoying, too, because I don't much like OpenAI as a company.

(Background: 25 years of C++ etc.)

- "Digital Apollo", a book about HCI, the tension between automation and human-in-the-loop, the history of systems engineering and minutae of each Apollo landing through those lenses. If you want a heavy dose of interesting, inspiring and thought-provoking HCI and embedded engineering lessons and anecdotes in context of the most thrilling examples possible you'll love this.

- "Sunburst and Luminary", the really quite charming and lively memoir by Apollo software engineer Don Eyles.

- "Apollo" by Cox and Murray, a go-to general history of the Apollo program that emphasizes program management and engineering far more than the astronauts.

This + the CuriousMarc videos and you'll feel spaceflight mini expert high, and be quite capable of maybe flying a landing in one of the emulators, actually understand the technical jargon in any of the Apollo landing videos or the Apollo 13 incident video, or appreciate some AGC source code.

It's just not super worth it.

If you want to look at a mission that did this, see China's Queqiao.