Joined February 2009
318 Photos and videos
"Coding" means a lot of different things right now. I would not spend much time on a "learn Javascript 101" course. I would absolutely be building software with AI and digging into understanding the generated code, and systems it's output and asking AI what the bits of code are doing and why they are structured the way they are.
Be honest devs, Is coding still worth learning in the AI era?
1
63
This is some drug dealer vibes coming from Anthropic. Just a taste to get you hooked.
😭 don't get too comfy with fable unless you're ready to spend lots of money.
2
52
Using Claude Fable to one shot a 90s video store simulator game in 5 minutes. Mind is a little bit blown.
1
1
151

In which I use @WisprFlow and @AnthropicAI Claude Fable 1M to oneshot a playable 3D first person 1990s video store simulator. To be honest, pretty blown away. Feeling super inspired and optimistic about where things are headed. Live demo starts at 8m 50s. Full prompt in thread.
30
In which I use @WisprFlow and @AnthropicAI Claude Fable 1M to oneshot a playable 3D first person 1990s video store simulator. To be honest, pretty blown away. Feeling super inspired and optimistic about where things are headed. Live demo starts at 8m 50s. Full prompt in thread.
2
4
256
Full Prompt as one shot dicatated through WisprFlow. ====== You are working in a totally empty directory. There is no code, no documentation, nothing in this directory. There are no external resources, no remote Git repository, et cetera, for this project. This project is bootstrapping from a completely empty directory. I want you to completely self-manage this project. At every turn, if there is a decision to be made, you can go ahead and make that decision on your own. You do not need to ask for input from me.The criteria: we're going to be building a game-like experience, a local game that will run on my Mac desktop here. We don't need to worry about hosting it or publishing it or anything. Whatever framework or technology stack makes the most sense, you can build it; it only needs to run here on my local machine for now.This is going to be a game experience. It is going to be a video store, a 90s video store simulator. It is going to be an immersive 3D first-person game where you're walking around a video rental store. You can pick up VHS tapes off the shelf or off the floor or off the counter and look at them, flip them over to look at the back, place them back on a surface, either out nicely on a shelf or drop it on the counter or the floor. That's all there is to this game.The key elements are:- We need some kind of a framework system that can do an immersive 3D, realistic first-person walk-around experience.- We don't need to worry about doors or anything. This can all be self-contained in one room.- There should be obstacles in: video shelves (freestanding video shelves as well as video shelves on the wall).- There should be some kind of a checkout counter or something.- Just model it after any reference images or reference media you can find on what 1990s video rental stores were like.- There should obviously be decor and decoration. It should look and feel immersively like a 90s video store.- To the extent we can incorporate sound, that is a fun element.- We do not need to represent the user's physicality in any way. It can be completely first-person. When you pick up a video, it can float; you don't need to see a hand holding it.- The one object that should be manipulated is the VHS tapes, and you should be able to flip them front and back and then drop them on the floor or place them nicely on a shelf.That is all the instruction and so sorry, input controls: AASD or WASD movement, mouse click manipulation of videos. I think that is all the input that you need. Everywhere else, where there is some decision to be made, I want you to just make a decision that optimizes for a fun, entertaining, nostalgic experience. I want you to continue working on this task until you have a working local application that I can play and experience. I'm going to give you one single pass of asking any questions that are going to be key technical blockers for you, where you can't get started on this project. If there are hard dependencies or something that you are not capable of installing, otherwise I want you to just proceed fully autonomously with building this experience. ======
1
81
Someone who's smarter than me about LLMs explain to me how you make a model like Mythos "safe" without just making it dumber at coding? Isn't cyber security in the way Mythos understands it mostly just a particular domain application of coding?
1
21
Wait a minute... selling shovels is how 99% of people make money in a gold rush. 🤔 What's the take here?
The AI gold rush has started. Most people are still selling shovels.
1
5
230
🌶️ take: The "you need to loop" concept is a wildly unhelpful artifact of programmer brains meeting current model/harness limitations and trying to be thought leaders. Truth: Once agents can perform a long running task, the bottleneck becomes how many sufficiently specified tasks you can feed those agents. A "loop" is a very much NOT a properly nuanced way of describing the process of effectively feeding agents sufficiently specified tasks. For me this "loop" none-sense currently reduces to "have you tried /goal"? Everything beyond that is performative. The people who are truly loop-maxing are NOT shipping quality software.
1
1
23
I only prompt via goto statements
Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore. You should be designing loops that prompt your agents.
27
Travis Fischer retweeted
reminder: it's okay to ignore the /loop discourse
10
1
59
4,130
My feed here lately is 95% slop/engagement bait... Getting tired
28
It is actually quite hard to make good evals… More than half the benchmarks you see online are pure slop.
53
lololol. Why does this feel right?
Opus 4.8 is like a very good Staff engineer who doesn’t want to work here anymore. Every conversation feels like it’s starting by shrugging and stopping whatever Reddit browsing it was doing and then coming back with the minimum viable, but totally skillful, answer.
43
Wait a minute... Finished that Anthropic piece, am I losing my mind? It very specifically did not say that we should stop/slow AI progress as of today. It said we should research and have a conversation about what would be required in case that becomes necessary. Which is like, completely self-evident?
18
They DID say they wanted to slow AI development...
Claude must be down HARD; without it nobody at @AnthropicAI can string together a coherent sentence. "Current ongoing issues with frontier models are with Opus 4.7 and Opus 4.8 and other models have recovered" I'm sorry, what?
4
121
There HAS to be a better word than "taste" for what we are describing here. "Good taste" just implies the ability to identify what other "fancy people" are going to appreciate in some subjective art form. It's a REALLY low bar. When it comes to research, wouldn't a word like "judgement" or "discernment" express this more precisely? I think we started using "taste" as a placeholder for "judgement that is hard to quantify precisely". So sick of hearing about "taste".
23
Weird human behavioral quirk... When Claude / Codex, prompts me for their next step, I more frequently reply "sure" than "yes" or "ok" or "go". Why is that? I think it's something about the gap between the way it is framing it's next step and how I would frame it. I think it's approximately, "sure, that's reasonable, proceed" vs. "yes, precisely as I envisioned it"
1
53