Prgnr

We are hell-bent on being right while always micro learning from the people around us, whom we seldom consider as wrong.

Now I am certain that for 1000 years we cannot achieve AGI. Not because we do not have the resources or data. But because we are corrupted to unimaginable depths.

@francoisfleuret We should look at transformers as a hypothesis of the real neural network or of some cognitive faculty of the brain.

If you have an explanation of why the transformer is so successful, here is a rapid sanity check: if it works for a huge MLP ("depth!", "SGD!", "magic of ml!") it's a very insufficient explanation.

We are in an age where spreading anti-truths can earn you a livelihood. And unfortunately people have chosen the path of living based on falsehood. But you can also use AI to earn you the livelihood. So you can pursue truth while spreading anti-truths.

Can everything be described using language?

@Lithros Disagreement does not imply that.

my wife thinks I'm a genius but she also disagrees with me sometimes, implying she sees herself as some kind of super-genius. yet she denies this

@pmddomingos The tragedy is that we know, but still end up exploiting this nature, instead of using it as a constraint.

The central tragedy of human nature is that the mind serves the heart even though it knows better.

@pmddomingos Even if you do not think, your brain will do this type of thinking for you.

If you think the people you've met are a representative sample of the world, then you think you've met everyone in the world.

@TrueAIHound The crow, with fewer cognitive devices at it's disposal, is as conscious as us.

This crow is superintelligent and has agency. His brain runs on a fraction of a watt and requires no GPUs, deep learning, datasets or datacenters. He's smarter than Sam Altman, Dario Amodei, Geoffrey Hinton and all the sci-fi fruitcakes of the AI community combined. 😀

The AI community is scrambling to find a new architecture to supersede the transformer but breakthroughs don't happen because you want them to happen so badly, it could take a day, 50, 100 yrs or more for the next breakthrough to come. That's why research has no deadlines.

@pmddomingos What is the solution?

When you tell most scientists that their reward model is broken, they agree and quickly change the subject. They just want to keep doing their research. Pure myopia.

@QuanticASI Kevin

@MovingFramesP who owns all the knowledge?

how can we find a question that has never been asked before?

@Plinz Fluid reasoning, information organisation, plagiarism detection, information democratization, etc. and research actually exist on patterns of fraudulent language.

It's time we protect society from linguists. The linguistics industry is trying to foist alleged benefits of researching language on us. Language has only three use-cases: 1. Fraud 2. Plagiarism 3. Cognitive off-loading Which of these are linguist promoting?

@kepano I know to act without a prompt.

your edge is whatever you know that the models don't know

Curing cancer with AGI might not be a matter of AGI building a universal cure, but a matter of AGI being able to be fully focused with high intellect 24/7 on treating *your* cancer specficially.

@m1nj12 Number of page should be limited and minimized.

Reviewing for ICML and it’s extremely underwhelming. One adds a component to existing framework but write like they proposed the entire thing. Another describe their method in a very misleading way (in my opinion, intentionally, to overstate the novelty/utility). :((

The world seems to be running short of peacemakers. Multiple wars running parallel, but so far no genuine peace efforts made by anyone.

The fact that intellectual people stop posting is the reason we remain ill-informed and dumb. Intellectuals must have high tolerance to criticism, confidence to share their knowledge and be open to new knowledge. Maybe there are no intellectuals anymore.

@desimojito I'm surprised that we were never taught this yet we know it intuitively.

This is better than teaching (a + b)² = a² + b² + 2ab