󠂪󠂪

Nobody got the irony in my last bio so now you get this

[󠂪󠂪 @TaiiwoDev]

󠂪󠂪

[󠂪󠂪 @TaiiwoDev]

@_Mira___Mira_ Unfortunately that's now how it works. Running the model slower would just occupy the memory for longer. They would need specific infrastructure that has enough memory but less compute. Best they can do is access to smaller models.

[󠂪󠂪 @TaiiwoDev]

@blacknredtext I don't know the rules there. Do I tip at fast food? Which ones are fast food? Tip for beers? Things that include food? I trust people to be honest because it's better than my wild guess.

[󠂪󠂪 @TaiiwoDev]

@blacknredtext When I'm in the US I ask the waiter how much I'm supposed to tip and what the total is. I don't mind paying, just tell me what the price is.

[󠂪󠂪 @TaiiwoDev]

@johncrickett That's not how ai works. It's not an average of the inputs, it's the best of them. It's seen the most popular solution more than the rest, but it's also seen every time that solution was refactored and why. That's why more params on the same dataset gives better results.

[󠂪󠂪 @TaiiwoDev]

@DeepStarts If a tractor drives itself, a farmer can control more tractors. If the farm runs itself, a farmer can run more farms. There is limited potential farmland, but unlimited potential software.

[󠂪󠂪 @TaiiwoDev]

@DeepStarts AI writes better code than me now. I'm still needed to tell it what to do for now. Next step is selling what it can do to people that need it. There's endless software that could exist.

[󠂪󠂪 @TaiiwoDev]

What has improved your AI coding workflow more:

A better model, better prompts, better harness/tooling, or better project structure?

For me, the biggest gains have come from improving the structure of the codebase rather than any of the above.

[󠂪󠂪 @TaiiwoDev]

@cryptobeast Well at least if we lose they'll be a silver lining.

[󠂪󠂪 @TaiiwoDev]

@_rad_o It really does lol. I have no idea how long ago I set that up.

[󠂪󠂪 @TaiiwoDev]

First 5.6-Sol prompt edited a migration right after fetching it. If it doesn't understand why that's bad, what exactly is it smarter at exactly?

[󠂪󠂪 @TaiiwoDev]

Unrelated to actual coding performance, but I find all the frontier models talk like lobotomized AI bros. I unironically use Gemini Pro when I actually want output that reads well.

[󠂪󠂪 @TaiiwoDev]

I found GPT-5.4 significantly better to work with than 5.5. They're removing it on the 23rd of July, and my initial tests of 5.6-Sol are not great.

[󠂪󠂪 @TaiiwoDev]

@HeyItsBonkka @MrBeast Ok what are you gonna do with the other 400k?

[󠂪󠂪 @TaiiwoDev]

@MrBeast Regardless of if they're bots or not, if you get 1m followers for no reason, they're worth nothing. You can buy 1m followers for less than $1m

[󠂪󠂪 @TaiiwoDev]

@RealPostFolder Wow they must be real files, that's the only reason they would hide them in hackthissite level 1 security. They must have thought nobody was going to find them.

[󠂪󠂪 @TaiiwoDev]

@thsottiaux Create new project from GitHub link. Sandboxd environments. Port discovery in the browser. Annotations without screenshots. The ability to create a new chat in an existing work tree. View+commit changes for this session instead of just last turn.

[󠂪󠂪 @TaiiwoDev]

@rateupx8 @MensFertility @reddit_lies I think the diet coke can? It doesn't seem ai generated to me

[󠂪󠂪 @TaiiwoDev]

@xanderwasserman @ChadCarleton @wholemars @Tesla_AI So you don't think self driving cars use LLMs, but you're just using LLM terms to describe how they work, and using LLM terms to suggest improvements? Okay.

[󠂪󠂪 @TaiiwoDev]

@lauriewired Isn't this you? https://t.co/0MCI0fyvSH

[LaurieWired @lauriewired]

I’m convinced that old games feel “hard” because modern emulators have like 30 layers of abstraction between the input and the screen.

Today you’ve got controllers going through bluetooth driver stacks, on general purpose operating systems, with all sorts of thread coordination, GPU APIs, not to mention display latency…

Much easier for the brain to learn when the original hardware is extremely deterministic!