Universal AI
Capable of everything a human is capable of
The calm before the storm ...
... is the hallmark of any revolution. The AI we call askabl will be the first instance of true intelligence - not only does it enable a computer to think critically, analytically and creatively, without ever hallucinating or giving a false answer - it will of course retain the advantages, a machine has: process information faster, think a lot quicker, never get tired, and remember everything, down to the last detail.
But isn't that what ChatGPT does? - no, exactly the opposite!. First of all, we are talking about true intelligence, in the sense, that it never outputs garbage: askabl speaks, indistinguishably from a human being. All hallucinations and weird errors are things of the past: askabl actuallly thinks, exactly like we do. That makes it 100% reliable: it can be your lawyer, or your stock broker. It can teach a college class. With the benefits of working speeds many orders of magnitudes higher, perfect memory, making no mistakes ever - all while remaining neutral and objective. In one sentence: it is all that big AI keeps promising, but never delivers on.
Starting with language ...
We are beginning with natural language, as the primary way of expressing thoughts and ideas. askabl won't be like ChatGPG or Claude, who merely mimick the way people talk, and are merely (sometimes) clever chatbots: by having adopted billions, even trillions of examples, and choosing the most probable answer, it seemingly engages in intelligent conversation - but it possesses no real understanding of what it is saying. Everyone can see that, through wrong answers, hallucinations, or the chatbot going completely off-topic.
Here are some of askabl's key features ...
- Absolute reliablity: askabl could not possibly - per the way it is constructed - hallucinate or give you a wrong answer. When a complex task or question is given, or one there is no clear answer to, it will absolutely tell you so, while still trying to come up with information that might be helpful.
- Critical reasoning: askabl can deliver a chain of thought in a very human-like, expert manner. It never forgets context, possible variables that could influence the answer, and deals with complex problems the way humans do: a mix of intuiton and analytical thinking, for example breaking down a problem into smaller parts while never losing focus on the end goal.
- Linear scaling (minimally): Unlike current models, whose scaling has long ago plateaued, and needs exponentially more compute to reach marginal gains, askabl scales linearly, dependening primarily on the speed of its processing unit and database. This also has hugely important correlates: askabl will never need a Giga- or even Terawatts of computing power. In fact, a handful of conventional, comparatively tiny data centers, located strategically, will suffice to serve the whole world.
- Creative thinking: take any cutting-edge research problem, and askabl can deliver new ideas on how to solve it. The same applies for all sorts of challenges - whether it is a household question you're looking to get an answer for, or a math riddle you wanna solve.
- Truth and falsity: askabl can determine what's right and wrong, what's true and false. This is the basis, as all rational thinking requires an account of truth.
- Right and wrong - the aligment problem solved: askabl can understand, what is right and wrong in a moral sense. All ethical behavior can be argued for very rationally: cooperation is, in most scenarios, a lot more productive strategy than rivalry or competition (something not all humans have fully understood yet). Peaceful coexistence is WAY more fruitful, for any two parties, than going to war with each other. An AI as evolved as askabl naturally comes to conclusions like these. Natural selection and evolution has invented this behavior thousands of times, in an abundance of ways and instances - most often independently from each other, spontaneously, and without genetic predispostion.
- Customization: askabl can be modified according to one's preferences or needs. Ask a question, and it can give you a short answer, or the whole story. When you need a solution for a problem, it can come up with the best solution on its own, or filtrate the net, when the task surpasses its current capabilities.
- Expandability: askabl can be expanded to any domain, not only langauge. For example, in the future it will be able to code even complex apps for you, or function as your absolutely trustworthy, objective lawyer. For its first extensions, we have chosen image recognition and coding.
These are features that exceed what is possible with today's technology.We are a small team, so it will be bit until all these things materialize. But we want to share it ahead of time - because we're excited about all the new possiblities.
We are positioining askabl much more as a serious tool than a chatbot. It is not meant for entertainment, but for actual, reliable productivity. We believe people had enough of AI slop and sycophancy, don't want to hear another doom story, and expect a tool that works reliably, is truly helpful, and does not hallucinate an answer, in the rare case it is unable to answer a question or complete a task. In this case, askabl will simply and politely tell you so.
how it works
The project employs both new and old concepts within Artificial Intelligence. But the essential trick is not to rely on neural networks, but instead an analytical core for its central intelligence core In this regard, it is like traiditional software. This may seem anachronistic, because neural networks have become virtually synonymous with AI, and AI has become synonymous with LLMs. But these have drawbacks and limits - which is why we have come up with a different approach. The core (and everything else) is analytic - in the sense that it is completely transparent - unlike neural nets, whose inner workings are a black box. This means, askabl can be (safely) enhanced, modified, extended, and apllied in virtually all scenarios - in constrast to neural networks, which cannot really be altered once they are trained, need continous traning, and cannot be transferred to a different domain than the one they were trained for. Read more in the technicals.
a word about ethics ...
We are acutely aware of the potential of the technology, and the risk for abuse that comes along with it. As much as we love the hacker spirit, that all information should be free and availabel to everyone, there are too many, gravely dangerous scenarios conceivable, if our AI was completely open-source: it could be used for a wide range of malicious purposes. That is why, at least for now, the theory, the conceptual details, as well as the source code, are kept secret. Read more about our ethical rules in our mission statement.
a word about computers ruling the world ...
There is absolutely no point in talking about the Terminator, Skynet or I, Robot. AIs like askabl do not have agency or free will, like human beings do. They are utterly agnostic in that regard: they do not want anything, as they do not need, desire, like, dislike, or feel at all. askabl's management layer, as well as its core, behave according to code that is entirely analytical (i.e. controlled by humans). Thus, the scenario of it taking over the world, is just as likely as a human walking on water.
No hallucinations ...
It is by design, that is theoretically, impossible for askabl to hallucinate, or give any sort of false, impertinent, or otherwise inappropriate answer/information.
No (scaling) limits ...
askabl is only limited by its access to information, and the speed of its processing units. And the upscaling is linear, instead of exponential.
General AI
Expandable to any scientific field, any (mental) job currently done by humans, all existing knowledge.
A thinking AI
askabl replicates human thinking. Critically as well as creatively.
Universal helper
Just tell askabl what it can help you with. Have legal questions? Need any other advice?
Ethical guidelines
Strict rules apply - to not let askabl's potential be abused.
All coming soon ...
askabl is under development. We can't give you an exact date, but we are working on it!!
Development stages
Something like askabl doesn't happen overnight. In fact, it's been a concerted effort, spanning over ten years. The following is a brief overview of the stages we have already reached, and what is to come.
alpha stage
Basic idea and concept
beta stage
Proof of concept
gamma stage
Proof of work
delta stage
Coding a prototype
epsilon stage
Coding the first alpha for public use. Work currently in progress.
Release
Introducing askabl to the world
Extending sources
First was Wikipedia. Then the internet. Then books. Then science papers. Then literally anything else ...
Extensions
Integrating capabilties ... first comes image recognition, including diagrams and graphs. Then spoken text recognition and output. The last extension, but the most important, is coding - once askabl can code on its own, it can be programmed to code extensions for itself. At this point, we will not have to add other capabilities manually: askabl will simply code them for us.
Endgame: universal AI
Adding any and all capabilities as askabl goes along. askabl can code an app for you: simply tell it what you want the app to do. askabl can create complex websites for you. askabl can be your lawyer, your stock brocker, your universal everyday helper/problem solver. It will function like a speech-based computer OS; just tell the computer what you want it to do.