Cuth

Saw somebody did this for ChatGPT only, so I added Gemini and Claude. Who do you thin will get there first? This seems to think it will be OpenAI.

@dev_maims Because most people don't know how to context engineer so they pay for it in real money.

About 85% of so-called “AI startups” are basically doing this: Collect your input Pass it to the OpenAI API Show you the output Then hit you with a $25/month subscription

@glagstonegame The aesthetic is elite.

While making GlagStone, I was inspired by late-80s Macintosh aesthetics — monochrome screens, tiny windows, dense little menus, and early city builders like SimCity on classic Mac. I wanted the game to feel like building a whole island inside an old black-and-white computer. That’s why GlagStone became a strict 1-bit city builder. #indiedev #citybuilder #pixelart #FollowFriday

@TheEduardoRFS It's all about context engineering and Claude Code has a better harness.

Codex is so much more stupid than ChatGPT, it's likely something about context, but it really pains me, it's just completely usless when it comes to mind reading.

@NikTek What kinda sentence

What are they feeding these eastern developers?

@cybercpu @leecronin

If you were GPT-3 I would not be surprised. You're just loosely connecting various topics and making massive assumptions about what I have said, or what I believe. It's like you read your first book and came home to tell poppa. How about this... Ask Google, Gemini, Claude or whatever AI you use if machine learning is learning and stop bothering me. You're just proving you don't have the ability to learn. Are you "sentient"? I believe you mean it is not "conscious"... It is not learning in the human sense. It takes data and experience which updates it's weights and allows it to understand the environment from learning rather than being explicitly told what to do. Whether you can it learning or not does not matter to thousands of scientists who do it. It is literally learning in a technical sense.

@ItsCuthulhu @leecronin Did you even bother reading my post? Okay, this is what I don't get. How can someone work in the field of machine learning but not understand how it works? Let's start there.

@RoundtableSpace Is this like some sort of kink? A Chinese kid solving some made up problem or making a million dillion dollars?

A CHINESE QUANT ALLEGEDLY MADE OVER $500K TRADING OIL ON POLYMARKET WITH AI. The setup combined Claude, OSINT scraping, and simulation models to analyze geopolitical headlines and predict oil market moves before they happened.

@cybercpu @leecronin I don't know one person that believes a single model today is alive... And yet they're useful. And yet they learn from their environments and gain intelligence. Why would they need to be alive?

I found that one of two things is true by most people that defend AI. They are either financially tied to it or they lack talent that they replace with it. AI is a grift that depends almost entirely on people believing that the bot is alive. That's why the majority of vocabulary related to it is anthropomorphic.

@cybercpu @leecronin Maybe explain what you don't get and we can start from there.

The way an LLM is "trained" is it breaks words down into tokens. These tokens are partial words. Think syllables but not quite like that. They're then giving a mathematical value. Essentially all it's doing is math. It knows that token a 94% of the time is followed by token b but if token a includes token C the 98% of the time it's followed by token E. It doesn't even understand the words that it's processing. When you ask an LLM a question it's nothing more than you typing numbers into a calculator. The LLM breaks those words down into tokens that have mathematical values and then compares them to its parameters that were adjusted during the training process. It doesn't learn anything. It has more in common with a copy machine or a calculator even than a human being. For someone that claims to work in this industry the fact that you don't know that is either disturbing or just proves you don't.

LLMs, they do learn in so many ways: pretraining, instruction tuning, RLHF/RLAIF, preference optimization, update runtime context, memory management, updating MD files like skills/souls/heartbeats... How the fuck else would know predict the next token? They were trained. They learned from the input we gave them not somebody writing a correct answer for every single question. Insane attempt at trolling.

@ItsCuthulhu @leecronin And thank you for proving you don't do this for a living. Because if you do You're lying. It's much more likely that you're just using that as some kind of a fallacy of authority. That's not the way LLMs work. They don't learn anything. They copy patterns. That's it.

Folks.

@cybercpu @leecronin It is literally learning. That machine does not know how to do something. We give it examples. It figures out how to do it without being strictly programmed. That is learning. You just have some definition only you know that you refuse to share with the class.

Just because machine learning has the word learning in its name doesn't actually mean it has the ability to learn. That's simply an unfortunate use of vocabulary to anthropomorphize what's happening. If you do this for a living then you should understand that computers don't learn.

My argument does not require that at all... You're connecting dots that don't exist. Have you never heard of machine learning? It is what I do for a living. Teach a model to learn. That is the foundation of all of these models. Neural networks, which is a type of machine learning.

@ItsCuthulhu @leecronin It's actually exactly what you said. AI does not have the ability to learn like people do. That's a sentient trait. Your argument inherently requires you to anthropomorphize AI. LLMs have more in common with a photocopier than a human being.

I can't imagine not making anything that comes to mind. I've been waiting for this my whole life. Things like painting and coding took me decades to learn. Now my ideas can go from mind to matter in a day or two... This is great for people who were always creative but the speed they could get the idea out of their head before it was gone was too slow. I abandoned so many ideas before. Now I make new shit all day long.

@cybercpu @leecronin That has nothing to do with what I said.

@ItsCuthulhu @leecronin AI isn't alive. It's a glorified copy machine.

@bilawalsidhu What are the system requirements to try something like that? So cool.

one 2d photo --> 3d gaussian splat quick test with Echo-2 by SpAItial -- these 3d scene generation models are getting better! already at a sufficient quality to serve as a virtual set / backdrop in your 3d tool of choice

@justinsunyt It's because you tried it 6 month ago lol.

kanban doesn’t make sense for coding agents we tried it 6 months ago. every task just ended up in the “needs review” column

We should not be encouraging companies to ship something that could lead to insane bioweapons or giving zero days to anybody with a subscription. It was selfish and foolish for OpenAI to just release this shit and then do a Bio Bug Bounty, instead of something like Project Glasswing like Anthropic.

Anthropic looked at their most capable model & said "the world isn't ready" Openai looked at theirs & said "ship it" that's the whole story one company is scared of what they built...the other is scared of falling behind and we're all just watching to see who was right 💀