I call it brute force because as much as I love agents and I have created my own auto-researcher, their understanding is limited. You can try to make it take notes of its learning and get inspired by literature to reason solutions, but the main problem is that it mode collapses into the mean solution. On toy problems this tends to be fine. If you work on sota RL research, it often ends in endless loops trying to improve without tackling the main issue unless you guide it.
AI agents are amazing but the truth is that in the current state, they struggle where humans struggle. That is why they can 10x productivity but you don’t see a 10x in output. There is no human that can say “this is the way to do RL for this problem” without trial and error. Therefore AI will struggle. Agents learning is not the same as RL agents learning from rewards. It can basically add learnings to its prompt, but it cannot massage the knowledge into their neural networks. It’s like humans relying on only their short-term memory. There is a severe limitation on agents ability to learn. The best way to learn math is to solve problems with math. You need to get learned knowledge into your neurons. No true understanding can happen otherwise. Same with agents.
I have no doubts that we will overcome this problem. But right now teamwork between the human and agent is much more efficient that full on auto-researcher mode.
I would love to be proven wrong but I have yet to see anyone that has shown a proven auto-researcher that produced something of value that is not just a basic webapp or a +1 research paper.
@NeuralYogi@yacineMTB If something can still be considered an art for humans, then agents can’t really do it. We still can’t give a correct formula for how to tune RL. There are still implementation details that affects results.
AI might be able to bruteforce if more compute than what is worth it
@stacyidv@Dexerto This retarded analogy (by a woman) is equivalent to saying during ww2: “why the outrage of your allies being killed when there are nazis dying?”.
Some animals are cute. Others we don’t give a fuck about. I hope that helps woman
@BobbyEParody@Dexerto It’s not hypocrisy. Dogs are cute. Cows and pigs are ugly and deserves to get eaten. Some animals deserve to live some doesn’t.
Vegans are even more selective with which animals they think deserves to die. You do contribute to their deaths just by living
@sameercarpenter@coder_blvck I always outclassed everyone in programming and worked in backend (real software development unlike web app / fullstack).
People who hate on vibe coding never understood software development. You are nothing more than a code monkey
@4cb1063986@TrashPandaGirll@Channel6ixNEWS You prayed to him, made statues of him, painted streets after him. You burned cities to the ground for him. We do not forget how absolute crazy you went over someone who had a drug overdose. You’re obsessed with a loser because you can identify with him
@TrashPandaGirll@Channel6ixNEWS The irony of this guys statement. You literally made him your deity. You are owned by a crackhead. He like most black people thinks that dudes in dresses are disgusting perverts
@DirtyEffinHippy@Dominos_UK You’re just angry that your hair color stopped being a DEI buff. Now you’re treated the same you always were, an ordinary, uninteresting, no personality, low iq slut
All AI companies are offering discounted prices in order to collect data.
They all use same tech because they all use open source knowledge, companies are rapidly reducing the amount of publications and keep innovation in house.
I just hope all people predicting AIs failure will admit it later when they are wrong rather than pretend they knew AI would succeed all along.
@neevin It’s not even that, you naturally become more attractive if you have a good personality. Girls can be a 5 but if they have a soul it will make them a 10 in a man’s eye.
@evilyanderedev@S0UND_0F_FUTURE It’s already good. You just need to have the capability of a senior developer. If you’re just a code monkey who can’t think vibe coding won’t save you.
@XmindHQ@lochan_twt The main problem is that every AI researcher who tried it out and understand LLMs were instantly mindblown. Saying it was shit means he didn’t try it or didn’t notice the leap
@cgtwts Goes to show that people in leadership positions are out of touch. They believe current AI is agi. The current benefit is that you can fire 95% of the developers for the same results. Not that you will revolutionize your taxi app
@liwedlia You’re an artist. You became an artist because you were one of the dumbest in class. I understand why you’re threatened when your art looks like a first year student. You gave up learning to color and shadow making that crap lol. AI surpassed you long time ago
@AmazonessKing3@liwedlia I have a PhD in the subject and did research before ”generative AI” was ever in your vocabulary.
It is kind of cute when people who don’t even understand basic calculus has opinions in a field they don’t understand. Both you and AI bros have more in common than you think