The emergence of Language Models – LLM represents an evolution in our integration of artificial intelligence, and raises questions about the future of employment and the human-machine relationship.
If some fears are legitimate, they should not obscure our vision of the future: in this sense, continuing to put ourselves in competition with the machine—seeing to act as such, only serves to further inflame our fears on an increasingly obvious whim. Evidence of the loss of such competition. Instead of succumbing to mistrust, everyone should seize the opportunity that artificial intelligence offers us to refocus on what constitutes the core of our humanity. Whether in their design or use, understanding the performance of these models and how they differ from the human brain, and beginning to re-locate our experience, opens up wonderful prospects.
Human training chat surprises with its ability to produce natural responses reminiscent of those of men – going so far as to stimulate a more explicit anthropomorphism, it is worth remembering the importance given to human experience in its training, in contrast to many other paradigms (for example, BERT). In this sense, in addition to language masking, which consists of hiding words in different sentences and giving the model the task of discovering hidden words, ChatGPT training requires humans to evaluate results according to various criteria such as appropriateness of responses, morality, and respect for human values. Once this first step is completed, reinforcement learning is used to improve the performance of the model: here, the principle is to give rewards to the model, positive or negative, depending on its actions. By incorporating these rewards, the model learns correct grammar and response strategies. In the case of ChatGPT, the more answers generated are consistent with those provided by the reward model, which has learned human preferences, the higher the reward for the model. Also, this design process demonstrates importance human experience In training, to ensure performance and ethics. Models that do not include these human preferences in their training continue to struggle to perform as well as the person. For example, recent research by Meta shows that: (1) thanks to the language mask, the LLM is able to create representations of words by looking at proximal context, as the brain can do, (2) however, the brain is able to enrich this first layer of representation, by looking at the larger temporal and hierarchical context, in order to build a richer understanding of the text. LLMs that do not include reinforcement steps based on human preferences are unable to achieve this evolving understanding.
Moreover, LLMs are random parrots that are based on possibilitiesThey do not have the ability to plan, be aware, or update information. For example, information that reaches us (for example, the outcome of a match) is instantly updated by our brain, in order to improve our ability to predict future events. Thus, when we are surprised by the information, hippocampus – The structure of the brain associated with memory, realizes that it is time to restructure information, and moves from a memorizing pattern to a refreshing pattern. LLMs don’t have that flexibility: they’re made up of billions of parameters, and it’s impossible to know which parameter needs to be updated to update the information, and a full retraining would be expensive. Thus part of the research on LLM is devoted to going beyond these limits, in order to get as close as possible to the differential capabilities of the brain.
renew our expertise
used in a way that is informed and complementary to our own qualities, Generative artificial intelligence And LLMs can greatly enhance our capabilities. Research published by MIT researchers examining the effect of ChatGPT on the performance of skilled workers on typing tasks demonstrates that the use of ChatGPT makes it possible to: (1) complete the task; more quickly(ii) create content deemed to be such best quality in terms of writing, content and originality, and (3) to improve satisfy workers to complete the task. Also, if the tool allows highly capable workers to work faster, it above all allows others, at first with less abilities, to increase the quality of their output, to the point of reducing the performance gap between the worst and the best. Other research has also been able to highlight the ability of AI to improve individual decisions, or the ability of humans and AI to mutually enrich each other. These findings call for the promotion of distributed cognition invoking the need for human expertise in an artificial intelligence-enhanced world. In this sense, if artificial intelligence represents a major technological breakthrough, it represents above all a human revolution, requiring a fundamental change in metacognition, our humility, and our relationship to the world. Ultimately, technologies do not change societies – it is their reallocation by humans that makes them evolve.
To achieve this, it is necessary to understand our intellectual limitations and our complementarities with artificial intelligence, to validate our curiosity rather than our pride, or even learn to ask the right questions. Everyone, on their emotional journey, must develop their critical thinking to be able to understand not only the potential biases of artificial intelligence, but also their own human biases. The question is not whether AI is perfect, but rather whether it can do better than the human status quo. Then intellectual liberation becomes an essential lever for creating an informed understanding of the possibilities of AI, but also of its gray areas. In this sense, artificial intelligence opens up extraordinary possibilities of access to knowledge for everyone, and aims at this learning lever exceptional. It is up to us to turn these assets into tangible and useful actions: this is our share of humanity and experience. Thus, the real threat is not the AI itself, but that we try to turn ourselves into a robot: to maintain an advantage over machines is above all not to act as such. Therefore, the time has come, for us as human beings, to understand and redraw the true place of our experience, our work, and our humanity.
Opinion article by: Emeric Kubiak, Chair of Science @ AssessFirst
<< اقرأ أيضًا: هل تؤثر Chat GPT على التعليم العالي؟ >>>