If you are looking for clear answers, use several AIs synchronously. The answer must lie somewhere in between!
AI Playground | Compare top AI models side-by-side (vercel.ai)
This is incorrect. Current AI does not make sense of its inputs. It simply regurgitates the output that others have produced for similar outputs, possibly translated in a way that appears in its larger input set.
Current AI does not think nor make sense of things. It pattern matches. That pattern matching has gotten closer to producing the appearance of thinking, but it hasnât reached it yet and doesnât seem to be that close.
Consider the âHow many Rs in raspberry?â question. Generative AI does not know what a raspberry is nor that this is a question about the word rather than the thing. It just knows that a raspberry is a berry and that someone has asserted that there are two Rs in berry. Therefore, there must also be two Rs in raspberry. They retrained ChatGPT to fix this for strawberry, presumably by adding the assertion that there are three Rs in strawberry prior to it composing the assertion that there are two. The problem remained for raspberry and cranberry because ChatGPT has no way to understand what it did wrong.
If theyâve fixed raspberry and cranberry since then, presumably they used the same method as they used for strawberry. Try elderberry or bearberry or make up a word (e.g. burrberry or rarerberry) and see if itâs fixed there.
Current AI does the reverse of what you said. It does not make sense of its inputs. It relies on someone else making sense of the inputs. It increases the need for people who can make sense of things, not reduces it. Perhaps this is a step closer to the singularity (time when AI can do thinking jobs on its own without a human supervisor). Or perhaps it is a step in the wrong direction. A local maximum that is drawing a great deal of attention but is not seriously progressing towards actual understanding.
Wow, you necroâed a post, then quoted out of context to suit your own reply. Lemme quote myself properly.
âmaking sense of itâ does not mean making sense of it in the human context (we donât need an algo to make sense of data for us). It means making sense of it such that the algo can know what to do with it. Internally, AI algo is an abstraction of collections of multi-layered bpp neurons feeding multiple outputs back into itself. Its âlearningâ is nothing more than convergence, where said convergence is guided by the huge amount training sets. The more training data that says a=b, the more likely it will regurgitate b when given a. There is absolutely nothing to stop pollution where you intentionally use foul training data with false truths and said AI will simply regurgitate the results back. âmaking sense of itâ means making sense of it in the algorithmic context. The output at the end is whatâs useful for humans.
The correct term, iirc, is an inference engine, not pattern matching. It simply infers from large datasets what the baseline is. If your datasets show patterns, it will infer said input as part of the pattern and so on and forth. If you work the maths, all inference has a probability of accuracy. Neural net researchers always have to show the probability values when presenting/publishing said works. Current marketed AI hides this number from the user in order to sell an illusion. The âappearance of thinkingâ is our own cognitive bias that incorrectly assumes this: âThe observation of an intelligent response constitutes an intelligent entity at work.â The truth is: âIntelligent responses can be mimicked, therefore it does not necessarily mean intelligence at workâ. Two identical exam papers, both with full marks from 2 humans do not mean 2 intelligent humans, It means 3 probabilities: 1 intel+copycat @50%, 2 copycats @25% or 2 intels @25%. If we take 2000 exam paper results from both humans, all with full marks, all 3 probabilities remain except that the values now become: 1 intel+copycat @0.05%, 2 copycats @0.05% or 2 intels @99% (this is only for illustration, do not quote as theory!). As you can see the probabilities are never 0. Neural networks do the same thing. Given enuff datasets that there are 3Rs in raspberry, it will regurgitate that there are 3R in raspberry. Foul up the datasets and it will give fouled up responses.
It appears you did not read my full reply, I quote the key below for your convenience. I feel there are already enuff explanations. Feel free to do your own homework. I have intentionally diluted my explanations so general readers can understand. Do not be mislead by your own cognitive biases, else youâll risk being brainwashed by hype and media. Hope it helps, cheers !
And again, I point out that that is the wrong way to put it. AI doesnât make sense of it. As you say, itâs not sentient. It classifies, organizes, and stores the data (which may be what you want to say when you refer to making sense of it). ChatGPT (or generative AI in general) extrapolates (or infers if you prefer that term) from the data. And as a result, it makes nonsensical assertions.
âBerry has two Rsâ and âstrawberry is a berryâ are true assertions. But that does not mean that âstrawberry has two Rsâ (the conclusion that it reaches). Thatâs not making sense. Thatâs not useful for humans. And itâs not based on foul training data. Youâre not going to convince me that there are more claims in the training data that the different kind of berries have two Rs than three. Itâs based on bad extrapolation because it has no actual understanding (or sense) of the data.
If it were just bad training data, it would be much easier to fix. We could weed out the bad training data. However, making (bad) extrapolations is fundamental to the model. Itâs what makes it seem like it is working, as it can talk about things on which it never trained. It may not be possible to fix the bad extrapolation (hallucination) problem without breaking what seems good about the AI.
Incidentally, I did read the part of your reply that you quote at the end. I mostly agree with it (at least if we limit to generative AI). However, I disagree with your earlier assertion. That assertion does not lead to your conclusion and is incorrect and harmful. Itâs also incorrect, if less harmful, to say that they just regurgitate the most common statements. If they did that, generative AI would not have hallucinations. The hallucinations come from the extrapolation. The same extrapolation that makes it seem more lifelike or real than previous AI.
People saying things like âAI makes sense of itâ is absolutely the wrong way to put it and is why we have people wondering if AI is sentient. Itâs the cognitive bias that we need to put away.
Yeah, works with humans too⌠Oh, silly me. Actually forgot that the AI is created by humans
AI Granny - this is a lovely one and very useful! I love this dude: https://www.youtube.com/@ScammerPayback
The biggest security gap is small children. They blab everything they hear at home!
AI, like a small child, does not know what a moral code is. If you tell the AI to win against the chess computer, it will do whatever it takes to win. You didnât mention that it should follow the rules of chess!
I get a bit scared pipi when I think about robots being equipped with artificial intelligence and do not adhere to any rules.
The DeepSeek about Babylon.js (nice digest by the way):
Babylon.js is a powerful, open-source 3D engine built for web-based applications. It has gained popularity for its ease of use, flexibility, and robust feature set. Here are some of the key strengths and advantages of Babylon.js:
1. Ease of Use
- Beginner-Friendly: Babylon.js has a gentle learning curve, especially for developers familiar with JavaScript and TypeScript.
- Comprehensive Documentation: The official documentation is extensive, with plenty of examples, tutorials, and API references.
- Playground: Babylon.js offers an interactive playground where you can write, test, and share code snippets in real-time.
2. Cross-Platform Support
- Babylon.js works seamlessly across all modern browsers and devices, including desktops, tablets, and smartphones.
- It supports WebGL, WebGPU, and even fallbacks for older devices.
3. Rich Feature Set
- Rendering: High-quality rendering capabilities, including PBR (Physically Based Rendering), HDR, and post-processing effects like bloom, depth of field, and more.
- Animation: Powerful animation system for bones, morph targets, and procedural animations.
- Physics: Built-in support for physics engines like Cannon.js and Ammo.js, enabling realistic collisions and dynamics.
- Particles: Advanced particle systems for creating effects like fire, smoke, rain, etc.
- Audio: Spatial 3D audio support for immersive experiences.
4. Extensibility
- Babylon.js is highly modular, allowing developers to extend its functionality with plugins and custom shaders.
- It integrates well with other libraries and frameworks, such as React, Vue, and Angular.
5. Community and Ecosystem
- Active Community: Babylon.js has a large and active community, which means plenty of forums, tutorials, and third-party resources.
- Tools: Babylon.js provides tools like the Scene Editor for creating and editing scenes visually, and the Inspector for debugging and optimizing scenes.
6. Performance
- Babylon.js is optimized for performance, making it suitable for both simple projects and complex, high-performance applications like games and simulations.
- It supports features like instancing, level-of-detail (LOD), and efficient memory management.
7. VR and AR Support
- Babylon.js has built-in support for virtual reality (VR) and augmented reality (AR) experiences, making it a great choice for immersive web applications.
8. Open Source and Free
- Babylon.js is completely open-source and free to use, even for commercial projects. Itâs backed by Microsoft and a passionate community of contributors.
9. Regular Updates
- The Babylon.js team releases frequent updates, adding new features, improving performance, and fixing bugs.
10. Use Cases
- Games: Babylon.js is widely used for creating 3D games for the web.
- Visualizations: Itâs great for data visualizations, architectural walkthroughs, and product configurators.
- Education: Used for interactive educational tools and simulations.
- Entertainment: Ideal for creating interactive 3D experiences like virtual tours and storytelling.
Example of What You Can Build:
- Interactive 3D games.
- Virtual reality experiences.
- Product configurators for e-commerce.
- Data visualizations and dashboards.
- Architectural walkthroughs and simulations.
If youâre looking to create 3D content for the web, Babylon.js is an excellent choice due to its balance of power, flexibility, and ease of use.