bilater

Create shows and movies. I, like you and everyone else, will figure it out. And it'll be better than what we have today just like its better to drink coffee and code vs toil at a farm 20 hours a day.

Author here. Seeing comments that this is a doomer article for some reason. I'm an AI optimist. It's humans that scare me. The article is about managing the transition to Valhalla. My toy projects and my job can disappear and I'm fine with it.

I talk more about the gran future here

https://www.hackyexperiments.com/blog/machines-of-loving-emb...

Nope - I'm as optimistic about AI as they come. It's humans that scare me. The article is about managing the transition to Valhalla.

Yes! Hopefully with AI we'll get more adaptations of Pratchett. I've only read Going Postal so far. The Discworld universe would be so cool to see.

The black bill that is coming that nobody is prepared for is that the value of a token varies greatly depending on the human. Companies will quickly find out its much better to give your top 10% engineers a lot more tokens and lay off your average engineers. The 10x engineer will become the 1000x engineer.

Wrote about this and the impact of to jobs here: https://x.com/deepwhitman/status/2058324179506831372

Same haha. But for me 1.5x is the sweet spot. Anything more and I find myself rewinding a lot. I want to feel comfortable absorbing info and not on constant alert.

[flagged]

/goals on both codex and claude is insane.

I gave Claude a full end-to-end task of creating a graphic novel. I've been thinking about this idea of leveraging GPT-Image-2 for creating a comic book for a while, but didn't have the time to do it, so I said screw it, let's see if the models can figure it out.

I gave it full autonomy in choosing the story, artwork, and creating the end product. Here is a summary of the steps in its instructions:

1. Research the last 30 years of graphic novels 2. Come up with a story + art style that would resonate based on what worked in the past 3. Write the script 4. Break it up into image prompts for each page 5. Call a Codex subagent to generate the images using GPT-Image-2 (I could have swapped this for an API call, but I got a sub and didn't want to pay any more money than I'm already forking over)

Claude got to work... and a few hours later I got a 25-page PDF of the graphic novel.

I can easily extend this to more customized and longer storylines with specific artwork and maybe even swap out with Codex next time.

I was curious why it picked this particular theme (the setting is in Karachi) and protagonist, and Claude replied (accurately) that because I have a Pakistani name.

I used HyperFrames to create a quick video with some music.

except this will go away. we will likely reach a point sooner rather than later (I think 2027) where it will be infeasible for humans to review the code. This will happen at startups first rather than big corps obviously and the engineers who design systems (dark factories) fully leaning into this will have a huge advantage. And yes there are exceptions to this and a play on the other side but the this is where the buck is going. at that point even Assembly becomes interesting.

Alec Radford has been part of basically every AI breakthrough you've heard of: GPT, CLIP, Whisper, to name a few. So when he, Nick Levine, and David Duvenaud drop something new, I pay attention.

This week they released talkie: a 13B model trained only on text written before 1931. No internet. No World War II. No transistor. The point isn't novelty, it's that a model frozen in 1930 is a clean lab for asking what AI actually generalizes vs. just memorizes. Can it independently invent a Turing machine? Predict the transistor? Learn to code purely from in-context examples?

Reading the thread I had a fun idea I couldn't shake.

What if you took the premise, experts from the past reacting to the future, and turned it into a podcast?

So I did. Meet The Coming Age, hosted by four characters frozen in 1930:

- Edmund Crale, the newspaperman

- Henry Aldrige Thorne, the historian

- Dr. Walter Brennan, the economist

- Theodore Marsden, the engineer

Episode 1 covers the networked age, from PCs and email through smartphones and social platforms.

Build: hosting talkie myself was a slog and I had problems using platforms, so I used Codex to orchestrate the back-and-forth with the model hosted on the chat webui and stitch the output into a clean script, then handed it off to Jellypod for voice synthesis and production.

Let me know if you have any ideas on where to take this show!

I'd take this info with a grain of salt. You have to understand how new some of these developments are. It's only been a couple of months since we hit the opus 4.5+ threshold. I created 4 react packages for kicks in a weekend: https://www.hackyexperiments.com/blog/shipping-react-librari...

no NFTs but sadly yes it makes too much sense so might have to do it.

thanks - can't replicate this which template were you talking about?

Had a similar idea a couple of years ago but I think this is still tied to the old way of doing things. More like software 2.9 rather than 3.1.

You can generate your own audiobook (single voice or multi) https://www.plainscribe.com

Think of it as dropbox

one of the startups listed that "failed" is music.ly. the app bytedance acquired to launch tiktok

was expecting some actual reasons presented as to why this would happen. instead got some math.

big beloved projects

I have always wondered how archives manage to capture screenshots of paywalled pages like the New York Times or the Wall Street Journal. Do they have agreements with publishers, do their crawlers have special privileges to bypass detection, or do they use technology so advanced that companies cannot detect them?

Astro to Cloudflare, Bun to Anthropic. Good trend seeing people toiling away at OS financially rewarded.

I wonder if we'll get something like a CORS for agents where they can only pass around data to whitelisted ips (local, claude sanctioned servers etc).

I share some of my hacky experiments here

https://www.hackyexperiments.com

I'm curious how tools like Claude Code or Cursor edit code. Do they regenerate the full file and diff it, or do they just output a diff and apply that directly? The latter feels more efficient, but harder to implement.

Most things don't work. You can be an arm chair critic and scoff and you may be right a lot of times. But you'll also never really build anything of note and/or have crazy hockey stick growth in your life.

lol yeah its weird to me why even ppl on HN can't wrap their heads around stateless calls.

Good to have fallbacks but in reality most ppl ( at least in the west) will have internet 99% of the time.

If you are using local models for coding you are midwiting this. Your code should be worth more than a subscription.

The only legit use case for local models is privacy.

I don't know why anyone would want to code with an intern level model when they can get a senior engineer level model for a couple of bucks more.

It DOESN'T MATTER if you're writing a simple hello world function or building out a complex feature. Just use the f*ing best model.

I like how this is trending with another post called 'Slowness is a Virtue'

for now